Transcript for:
AI Applications in Trading Strategies

I'm going to show you how to actually use AI to trade so not like everybody else I'm going to show you how to actually use artificial intelligence to trade and it's not the way you're thinking no it doesn't just predict price I'm sorry to break it to you right here right now but I'm going to show you the actual way how to use AI to trade now I've been at this for over 3,000 hours today I'm going to show you everything step by step over the course of this short short video I'm going to hold your hand through the two most effective ways to how to actually use AI to trade by the end of this video you will know exactly step by step how to use AI there's two unique approaches that I show you through this video number one no I'm not having AI predict price no way the reason is because if everybody uses AI to predict price the price is going to change and the price is not going to be that predicted price anymore does that make sense let me say it again if everybody uses AI to predict the next minute price or hour price or day price of any asset then that price is not going to be the price anymore because people are going to buy and sell around it it's not like the weather when you predict hey tomorrow is going to be sunny well if 2,000 other people predict tomorrow is going to be sunny it doesn't matter it can still be sunny or if you predict there's going to be six car crashes yeah that's morbid I don't even want to talk about that if you predict that your your sales numbers in your Fortune 500 company is they're going to have 500 sales or something and then you have other 20 other data scientists predicting there's going to be 500 sales that's not going to change the sales but in trading if everybody's predicting price that's going to change the price so that prediction is not going to work that makes sense right so I'm sorry to burst the bubble right off here but I'm going to show you two legit ways to actually use AI to trade the first way to use AI to trade is the most obvious way dude we have Claude we have chat GPT we have luxie we have all of these different AI platforms that we can just ask questions now you do need to kind of know how to code in order to get the most out of them I think that they gave us way too much power with these AI platforms to be real but the thing is nobody knows because nobody takes the time to learn how to code so they don't see how much power we actually have I literally sit here I build trading systems that used to take months if not years by hand and multiple team members so that's the first way we're going to use AI we're going to use llms those are large language models to help me take my ideas out of my brain and put them down in code so then I can go ahead and test them I follow this system all day long this is my simplified system to algo trading so algo trading is the process of automating your trading and I start with researching and then back testing so llms AI can help with researching and back testing because it can take an idea and expand on it it knows everything I only know some things I don't know much and then it can help me code out these back tests to see if that research those strategies I came up with if they actually work in the past so llms are the first way I'm going to use AI but the second way is rest in peace to his soul his family everybody that loved him I loved him but not like anybody loved his family member you know how that goes but rest in peace the king Jim Simons Jim Simons if you want to Google him he is the best algorithmic Trader ever to touch this Earth and we've been studying him and you're going to see a lot of it here in this video but one of his favorite models before I get there he said he did not he let the he he made all strategies based off of data so if you're making strateg trading strategies based off of data that kind of infers that you're using machine leared because if you're starting at a Zer Point origin with just a bunch of open high low close volume data and some math well that's telling me that he said he starts with data that tells me he's doing machine learning we watched a ton of videos and he finally said it he really likes the hidden Markoff model the hidden Markoff model is a model a machine learning model that helps predict different states now I know this might be confusing but I go over it so many times throughout this video so it's going to get less confusing it was confusing to me too but the hidden Markoff model which I refer to over and over again as hmm is a machine learning model that helps predict hidden States now what does that mean mondev well if you go over here to some data here you can see there's like a a bull market State a regime so State and regime I'm going to use them interchangeably a lot in this training you can see this is like a sideways a consolidation state or regime this is like a bullish regime this right here is like a bearish regime so Jim Simons repeatedly said he likes the hidden Markoff model which is a machine learning model and machine learning is AI so not only are we going to use llms probably the the newest form of AI at least that we have access to in a userfriendly way but we're also going to use the hidden Markoff model which is a machine learning model that Jim Simons the goat used by the end of this you're going to have a machine learning model that predicts different regimes and it actually back tests dang I should pull up the back tests right now so you can see it let's see if I can pull it up really quickly if I can't then you're going to have to watch the whole thing dude I just wanted to CH you show you real quick so you can believe me I want you to stick around and watch this entire video because it's a journey dude it's a journey and you're going to learn so much and by the end of it you're going to have every single thing that I know about hidden markof models about using llms and AI in order to code for you help you flesh out your ideas you can see here these are a bunch of back tests here that we outputed and for example this first one has a return of 52% while Buy and Hold is 35% this is just the start we're just getting started with this stuff dude it's wild and the thing I like most about this is some of of these returns that we see they're only holding 11% of the time so that's one of the best returns that I can remember I didn't show you it right there I didn't have it prepared I'm sorry let's go ahead and just dive in and show you step by step by the end of this you're going to be able to know everything I know about how to actually use AI in trading and I show you everything step by step as always I go live every single day and I show you everything every single piece of code because I believe code is a great equalizer I believe if you know how to code you can pull yourself out of any position because you can build for the rest of your life and like I said I think they gave us too much power here with this this AI these llms I think if the the higher ups they knew about how much power coders have now with AI they would cut it off that's why I'm going so hard every single day because I don't think we'll have access to this forever I hope I'm hopeful we do cuz the can's already out the the worm's already out of the can whatever the saying is I hope we do but I'm going to show you everything step by step here right now let's get into it stick around for the whole thing Jim Simons love the hidden markof model so I wanted to investigate a little bit more got all the code written out so it looks like we're just importing pandas here and import numpy MP and then from hmm learn import hmm Google this B pip install hmm learn then map plot li. pyplot as PLP PLT PLT SK learn. preprocessing okay we're starting the hmm analysis here I'm using like uh what year is it it's 2024 so six years of BTC hourly data it's data 51,9 or 5 51,9 122 rows of data um I guess I could get more I could absolutely get more absolutely absolutely and different time frames but I just want to get this set up and working and then worry about all that stuff later but essentially loading in the data okay we're naming the files open high low close volume pd. read file path names I don't think that's right but let's go ahead and look at it daytime yeah I don't think that's right but we're just going to pretend like it is for now and no let's fix it let's fix it date time date time open high low close pd. read file names what if we just dro the read yeah yeah what if we did that okay okay and then just say print data let's start there print DF doad fine time. sleep oh we don't even have time in here AI sometimes huh load pre-processing data okay let's go down to the bottom here and it says training start executing uh starting main execution okay and then it starts this okay we'll start there so we'll just start there and then go step by step because then you get to see all the code I get to see all the code and then we all get all the code you know you see the code I see the code we see the code and then we can do cool things with the code maybe but I did watch that video the other day that you know compared the the hidden Markoff model hidden I'm going to keep writing that out because it's such a nerd word hidden markof model uh was the was better less than the RNN the current neural network it was less than let but it's only on one day only on one day and one person's features really because this is just feature engineering right that's what that one dude said at least Andrew ning I'm so sorry to Butch your name if that's true you I love you so so much yo they really they really Andrew really has all of these things let's go ahead and watch a couple of them not right now but maybe this is look at that deep learning AI stop it I feel like he had like a some class at some school oh this right here stanf or something engineering this is uh I watched a good amount of this dude I think we watched this whole thing okay so the reason I guy here is because he has said something about like it's just feature engineering it's like that's where the edge is and machine learning is feature engineering it's like coming up with the ideas what you going put what data you can put in what are the uh yeah that's it that's all I got I learned I learned machine learning before learning how to code it's crazy that's how I learn how to code so whatever it is what it is weird path but I'm going be able to pick this up easy dude not easy but it's like a refresher that's it that's it dude so lecture 19 we might go up in there any anyways let's go ahead and print out this data here boom got it unnamed six yeah my data is janky dude I forgot about that this is good this is a good practice how do I do that ah how do I select all the rows no all of these there's a way to do that um let's go ask AI okay so what I'm going to say here oh my God I love this what am I doing this is amazing I can't believe it I can't believe the opportunity we have here with AI we can ask anything so my data above is super janky it has a comma at the end of the column names indicating a new column how do I I know I could just drop it probably I can I can drop I know I can drop that column okay I got that but I've seen people do this little thing where they select the end column and it it's like a keyboard shortcut and it selects you have a selector on every line yeah that's it all right how how do I make it so on the CSV I I use that keyboard shortcut to um have a selector door blinker I don't know what these are called blinker on each line and then delete the last column I know I can drop with pandas but I'm curious about that keyboard Bo shortcut and how to do the same thing on on multiple multiple multiple lines okay you know what I'm talking about though right you know what I'm talking about I know you know what I'm talking about I see you're looking for a way to efficiently edit multiple lines yes there we go and your text editor remove the trailing commas this is a common task and many text editors have features to help us the T teni the ten the teni uh the technique you're referring to is often called multic cursor editing or column selector here's how to do it in some popular so press alt shift and I that's it that's all I need to know now I'm I'm off alt option maybe no control no dang what's all on Mac controll shift I oh okay option shift to add cursor end of each line okay let's try that option shift I maybe this not an i no I mean option shift I does it have to be in order maybe the same time n dude ain't it visual Studio code I'm just going to drop it I'm just going to drop it I'm just going to drop it because it doesn't work for me blinker cursor bro you should throw that data into a DB like postgress or my [Music] skill my skill how come how come that is the question I'm just going to drop it so let's go back to it my bad my bad I thought I would be able to pick that pick that up real quickly and it couldn't so drop drop the last column thank you okay it should be gone now look how much easier it is to do things with code you can't tell me code does not give you advantage in life can't tell me that I'm sorry I've been I've been living I know I'm not the oldest old head out there but I've been living and this stuff is Magic dude drops a column calls coding magic let's see print creating date time range okay let's say this let's say print creating date time in index calculating returns and volatility I'm not going to PR everything that'll take forever calculating returns and volatility okay the returns is the percent change of the close Okay volatility is returned the rolling standard deviation of a rolling window dang this is some good stuff be volume change volume percent change okay dropping nonv values data process and let's just print out our DF let's start there make sure everything's looking good this is we're calling the main main execution down here this is the main execution we just went through the loading and pre-process data okay you saw it all dude print data not not I want all of it though whatever I'll take the I'll take that I'm not sleep five no way 555 deal remember that you remember the 555 deal at dominoes I was good I was good I think I'm a little hungry so these returns has to be a lowercase close this needs to be a lowercase volume who did this dude should have gone with industry standards no clothes in it okay well let's go look at it close looks like a close to me did I spell it incorrectly I have to make the columns yeah I got to make the columns okay that's crazy because there are columns here but let's see here let's see what line is this on 201 108 um 27 here we go so returns it's having an a right there no it's not it's on returns this one right here so close do you have close CL print the columns then show me what the columns are yo it's so funny because there's no reason to code stop it I don't get it how did we get here because I would rather just write in English you know that's what I'm saying like I I know I know I need to print out the columns to check it out but it's like okay easier way to play the game dog easier way to play the game dog daytime open yo these are janky look at this there's a space in between that is crazy that is insane all right so I'm just going to go say close and then volume here I'm going to say rename the columns this is easy this is light work when you when you have this stuff rename rename the call cols two open high low close volume dude thank you D have. columns okay let's just confirm it ah no no no that ain't it that ain't it we got to have date time Daye time date time that is so janky come on dude come on you got got this there we go daytime now let's see what the columns are looking like okay perfect the columns look good now just getting things all formatted okay time. sleep no more sleep now let's go back come on bring it back okay so now we have the close calculating returns of volatility perfect perfect perfect creating the volume this should all work now I mean we'll see but let's run it okay great so we got the returns here we've got the volatility here and the volume change perfect all right let's keep it moving then I think it's important to understand this code base because this is like just a just a start B just a start with path all right so training hmm model okay so let's go check out the training so we're passing in this data here passing in the the above data to Now train the train train the hmm model OKAY training hm model let's go to train definition get go to definition okay so print training hmm with n components the features are returns volatility and volume change x equals data of the features values it's got the standard scale so it's normalizing the features standard scalar and we're going to go through all this don't worry don't worry scalar equals standard scalar X scaled equals scalar so we have to fit and transform it and then we have to fit it fitting hmm model and then this is that pip andall that we made earlier hmm. goian hm m n components equal n components ciance type equals full n iterations random state so the random State 42 model that fit okay this is the actual training of it I believe hmm training complete okay I don't think this works I think we we get into an error here so first off let's explain it so we can kind of understand and help the AI out explain the above the above code with two different uh examples that a 12 year all can understand um yeah let's see what are components uh what are components um n iterations covariance types um standard scaler all right this is good fit transform this is good this is good okay so certainly I'll explain the code using two simple examples that a 12-year-old might understand this secret mood detector imagine no I can't do this B secret mood detector all right imagine you have a magical mood detector that can figure out if your friend is happy sad or neutral based on three things how much they smile how loudly they speak and how fast they walk the components in this example the components Nal 3 n component equals three are the three moods happy sad and neutral the features are smile amount speaking volume and walking speed the standard scaler is like adjusting your eyes to see these features clearly no matter if your friend is a quiet Walker or a loud Runner fit transform is like calibrating your mood detector with all the information you've gathered and it equals means the detector will practice guessing moods a 100 times to get better at it yo this is actually pretty good covariance type equals full means the detector considers how all features might be related like maybe people walk faster when they speak loudly the gazum hmm is your magical mood detector that learns patterns to guess your friend's mood even when you can't ask them directly and remember we're trying to guess the market regimes maybe you don't remember maybe you weren't here I don't know come on bro keep it keep it up keep going that's it come on we going example two the weather prediction game let's say you're playing a weather pred addiction game you can't see outside but you have three Clues temperature humidity and wind speed you're trying to guess if it's sunny rainy or cloudy in this scenario the components n equals 3 n components equals 3 are the three types Sunny rainy and cloudy so that could be like trending consolidation downward consolidation upward I mean what Jim said there's like nine I'd be curious but those components could be could be uh those the different reg regimes regimes the features are temperature humidity and wind speed the standard scaler is like converting all measurements to a scale from 1 to 10 so it's easier to compare them fit ter transform is like studying lots of weather reports to understand what combinations usually mean n it equals 100 means you'll practice guessing the weather 100 times to improve your skills covariance type equals full means you consider how temperature humidity and wind might affect each other the gajian hmm is your brain trying to learn patterns to guess the weather correctly even when you can't look outside give me two examples me two examples of trading alos and boss for crypto um I mean I understand it but again this was hard for me to get through my thick skull oh I didn't want actually code examples I'm just going to move on let's see if this works there's plenty of videos on this people much much smarter than I okay so look at this look at that that's dope actually because yesterday this didn't work but Mund went up in there and raw doged it a little bit and fixed it if you remember the end of yesterday it was like oh well we did good but um there's an error is there still an error I don't want to start acting like I figured it out or anything oh I got to sleep there okay okay stop it I was hyping myself up real quick all right get off me get off me oh my goodness so explain what the outcome this should send back our return the model does it actually did it say training complete I don't think it did I think it was training right there yeah States predicted let's go ahead and say where does it say States predicted okay predict States dude we didn't go over this one hmm training complete okay so we didn't go over this part predicting States there's only three states analyzing States okay got you got you there's a pause there that's our pause that's fine but we didn't go over this part so I'm going to say explain this one explain this code explain this code with two examples not examples with uh the theme being trading alos and trading Bots okay so certainly let's explain predict States function so we're looking at this function here just so you can see it in the context of trading algorithms and Bot for cryptocurrency markets imagine you're running a sophisticated crypto trading bot that doesn't just look at price but tries to understand the overall mood or regime of the market this function is like the Bots Market psychologist trying to figure out what's date the market is in at any given time here's how it works features equals the returns the volatility and uh volume change and we have all that remember I printed it anyways your Bot is looking at three key aspects of the market returns how much the price has changed volatility how wildly the price is swinging volume change change how trading activity is changing this is like a Trader watching not just the price but also how nervous the market seems and how many people are trading x equals data x equals D this line here x equals data passing in features or in the features section values debot collects these three prices or pieces of information for each time period it's looking at X scaled this is like normalizing the data Maybe returns are usually tiny like 0.1% but volume change can be 50% the scaler makes sure the bot doesn't overreact to naturally large numbers that's cool but finally clicks this is where the magic happens States equals model. predict EX date so this is where this is where the magic happens the bot uses its trained model like an experienced Trader in int Traders intuition intuition to guess what state or mood the market is in based on the returns volatility and volume changes it seen print States predicted unique states the bot announces how many different states it found for example it might say it found three states which could be something like state zero calm Market low returns low volatility normal volume state one bull market High positive returns medium volatility and high volume Panic Market High negative returns High volatility and very high volume in the context of a trading bot this function would be called regularly maybe every hour or day to reassess the market State the bot would use these State predictions to make trading decisions for example in state zero which is a calm Market it may might make small careful trades what is Dar I don't know in state one bull market it might make larger buy orders in state two a panic Market it might sell to protect assets or make short trades the key idea here is that the bot is trying to adapt its strategy based on what it perceives as the market current market i s just like a human Trader would it's not just blindly following a single strategy but trying to understand the Market's mood and adjust accordingly this approach can be particularly useful in cryptocurrency markets which can rapidly switch between periods of calm explosive growth and sharp declines by detecting these states a trading bot can potentially make smarter decisions when to be aggressive when to be cautious and when to stay out of the market just move on I mean I understand that to you I hope so I mean if you don't that's fine though because it's this is like super deep stuff you know it's deep it's deep it's deep let's look at analyze stats States all right so print analyzing States DF do analysis is a data. copy and then analyze State equals state for state in range of model. n components so it grabs the components analyzing state State data equals DF where State equals State print State returns volatility and Dot describe number of periods in state okay let's just print it oh there's nothing to print plot results let's check out plot results print plotting results so we're getting subplots PLT do subplots so our good old PLT I've never understood why it's called PLT let's try to figure that out right now Matt plot lib as PLT what are you talking about my dude I don't know I don't know plot maybe yeah maybe because plot okay got it figured it out plot plot without the oh [Music] the ax1 plot index data okay we got the close now I'm glad I switched it to close because it's just all up in here bitcoin price in hmm States okay for state and range of model. end components mask equals what is a mask what is a mask great question in the context of data analysis and visualization a mask is a Boolean true or false array that's used to select or highlight specific parts of your data let's break this down in a way that's easy to understand imagine you're looking at a long series of cryptocurrency prices and you want to color code different periods based on the market State like bull market bare Market or sideways Market The Mask helps you do exactly that here here how it works in the code for state and range this Loop goes through each state the model has identified let's say there are three states 0 1 and two mask equals State equals States this is creates a Boolean array mask where it's true for every time point where the Mark is in the current state okay got it so it's just like true or false let's use a simple example to illustrate suppose you have 10 days of data and the states array looks like this States equal 00 0111 22 011 this Loop goes through each state the model has identified let's say there are three states 0 1 and two mask equals States equals State this creates a Boolean array mask where it's true okay when creating mask for each state You' get true true false false false false false false okay I got it so it's for each state so you get two it's for you that's where I was a little confused like where's this two come from but it could be a seven it could be a nine it could be whatever for State 9 is it true or false you can see for State one here it's not going to be true until 2 three 4 or whatever that is 2 three 4 all right figured it out understand it let's keep it moving B in the context of a training B algorithm these masks could be used to back test strategies for example you might want to see how particular trading strategy performs only during periods identifies State one duh okay okay okay you could use these Mass to calculate State specific metrics like what's the average daily return since when we're in state 2 in live trading identifying the current state could trigger different trading rules or risk management strategies yeah I feel like I should I don't know why I say explain it to me like I'm 12 just explain it to me for outo trading always here on out all right you heard it here first Mev live Live on YouTube live on YouTube every single day dude I don't know why I just like to show it I like to show this stuff opposed to just doing it by myself I'm sorry it's just like kind of like deep it's deep you know it's deep so like I'm not going to be able to comprehend all of it that's it that's it I can't comprehend all this you got to see it bro you you you you have a hack at this stuff dog I'm G show you everything every day every day okay ax. Legend so we're just building a legend here for the uh ax one and two so it looks like we're getting two plots I wonder if it will work let's go ahead and just run it up plot plot plot plot time do sleep I don't want to see the err that's why I keep putting this these time do sleeps I want to understand every single line of code you know why not okay so plotting results yo pzy plots pzy plots pzy plots a lot of thoughts get your head out the gutter be in your mind thoughts in your mind all right so I don't know but you know what we can do from here hey guess what guess what guess what guess what we can do from here dude we can do anything this is crazy this is crazy AI is nuts I can't believe it I can't believe what I'm about how I'm about to chop this up chop it up explain to me like I'm however old I am okay and in the context of trading Bots and trading algorithms we are building out a uh hidden markof model and Below attached attached no no no attached is the plotted stats I don't know stats or States now please explain it to me please explain with examples okay let's see it not code examples though don't give me any code hey not right now no thank you certainly let's break down what this hidden Markoff model plot is showing us in the context of trading Bots and algorithms for BTC the hidden Markoff model overview in this context the hmm is trying to identify different hidden states of the BTC Market these states are directly observable but are they aren't these states aren't directly observable but are infirm inferred from the visible data okay let's look at this data too let's just look at it while while I'm reading it because then that's how you do this stuff [Music] dude all right so these states aren't directly observable but are infirmed inferred from the visible data price returns volatility Etc the three states the model has identified three distinct States state zero 1 and two represented by different colors in the top of the chart so you can see three states zero one and two zero is blue one is yellow oh I'm sorry that's like orange or peach peach and two is green all right let's get it state zero this appears to be the most common state it might represent normal or Stable Market condition State one orange this state appears less frequently and often coincides with price increases it might represent bullish or a volatile uptrend this state is least common it repres it might represent a bearish or highly volatile State price chart top I mean let's start doing to-dos here to-do too too too I'mma keep it in the ideas I'mma keep it in the ideas um put um all types of States look into I can't get my words together look into look into Jim's favorite favorite yeah gy regimes regimes and make those the states that are predicted I guess guess what that's what I'm I'm trying to say okay so whatever this St is least common okay highly volatile I haven't I don't see any greens on here to be honest do you make it big bigger I don't see any greens that's okay um the Blue Line represents btc's price overtime I think that was obvious the colored backgrounds show which state the model believes the market is in at each point returns chart which is at the bottom this shows the percentage returns of BTC spikes indicate large price movements up or down so this is co and that coincides right with the price so this is lined up which is nice that hourly though this is hourly since 2018 on BTC I want to put Liquidations in there I don't know all the data that we got anything throw it all throw it all at it you got ideas throw it in the chat 777 every single day you're up inside my brain b it's kind of weird you you hav been here a lot honestly so just share your your 777s in the chat if you got a good idea share that with your brethren meet somebody in the Discord you don't have to share with me dude I just 1 plus one I just want 1 plus one to equal three like that one dude said all right price chart trade B applications example one state based strategy a Trad B could be used could use the states to adjust the strategy use a conservative strategy during light blue maybe scalping small profits State one orange Implement more aggressive buying strategy anticipating price increases in state two green either stay out of the market or Implement a shorting strategy example two Trend following notice how State one orange often appears during uptrends aach could use this to enter long positions when transitioning from 0o to one 0 to one exit or reduce positions when transitioning out of state one example three volatility trading the returns chart shows periods of higher volatility a bot could increase position sizes during state zero when volatility is lower reduce risk and tighten stop losses during States one and two when volatility might be higher example for mean reversion after extended periods in state one or two so one or two okay I mean I I think this data just needs to be I don't know I think we just need to keep going through this we're not even done we're not even done yet so risk management the rapid State changes in some periods like early 2022 indicate a highly uncertain Market a trading algorithm might reduce its overall exposure during such times Market regime detection the long stretches of state zero in 201819 might indicate a bare Market or accumulation phase a sophisticated algorithm could adjust its long-term strategy based on these Extended State periods in essence this hmm is providing a framework for a trading bot to understand the current market context Beyond just price it's like giving the bot a sense of the Market's mood allowing it to adapt its strategies to different Market regimes this can be particularly valuable in the volatile Andy cyclical nature of the cryptocurrency market okay so let's just keep it moving then let's see how much else like I know we got something else here so how much else we got so printing the transition Matrix let's go ahead and see that man I don't know if it's going to let us because we'll see let's just let's just see it let's just see it feel like it gets stuck on the the output of the the other thing you know plotting results yeah it's going to stop there showing plot okay I can close the plot now and then it should move on there we go transition Matrix okay I want to save this plot though analyze the stats predict the stats plot the results let's go save it dude let's save it to current directory tight layout pt. show save the save the plot to my uh data folder B in this directory all right that's it data dot okay but as as a plot B not as not as the data that's looking weird that's looking like the other thing okay let's go ahead and make a data folder here look at this look at this just just keep swimming be just keep swimming this is fun I like it if Jim liked it I like it all right get off me B get off me let's go ahead and check it out I don't even need to show it anymore I could it's plotting okay perfect perfect perfect it's still plotting but now I'm going to close the plot and then I see transition Matrix okay so I might go ahead and say n I'm not going to do any custom stuff there right now not the point transition Matrix okay let's go ahead and print out the transition Matrix let's get the stats of it oh snap okay yeah we've always got that though transition Matrix okay I'm going to copy that in a second I want to First go check out the transition Matrix walk ourselves through it see where we're at say see where we're at oh model. Trans okay there's no code there there's no code there dude there's not one piece of code there I mean mod model. transmat so I think that's just how you call the transation transition Matrix which is fine that's fine with me but mean State variance I just want to explain that explain this uh for my hmm can you you explain this transition Matrix what dude that's it stop it explain this transition Matrix please transition Matrix where are we I don't know I don't know anymore okay so certainly let's break down this transition Matrix in the context of your hidden Markoff model for BTC trading the transition Matrix shows the probabilities of moving from one state to another in the next time stamp in your case you have three states let's call them 0 1 and two The Matrix is read read as from state to state from state to state so transition from state zero you can see it's the 0.92 wait wait wait wait wait wait how are you going to change up on me oh it's negative 1 okay got it got it there's that to the 10th power thing up there so it's negative one so 92078 okay and point blah blah blah blah blah 254 92% chance of staying in state zero 7% chance of moving to state two and very small chance of moving to state two sorry one and then two transitions from State one there's a 24% chance of M moving to state zero and a 75% chance of staying in State One essentially 0% chance of moving the state to transition from State 2 100% chance of moving into state zero so that means there is no state two right interpretation for trading state zero likely normal or Baseline State very stable tends to process occasionally it transitions to State one rarely goes to state two State one possibly a bullish or trending State moderately stable often persists sometimes reverts to state zero sorry to wake you bro sometimes revers to state zero almost never goes directly to state two state two possibly a volatile or shock State extremely unstable never persists always transitions back to state zero immediately trading implications state zero could represent represent a calm Market a trading bot might use Conservative Strategies here State one seems to represent a persistent Trend a bot might use Trend following strategies when in this state state two appears to be a very rare instantly selloff correcting state it might represent sudden Market shocks or extreme events a bot should be prepared for quick reversals after detecting the state the Rarity and instant transition of State 2 back to state zero suggests it might be capturing outlier events in your data you might want to investigate what market conditions correspond to state two and consider if it's providing valuable information for your trading strategy it's not printing means and variances co-variances of each state okay let's see it dude yo I wonder if this is popping out sure is sure is cool cool cool we know how to code just a little bit just a little we know a little bit of code then we got AI my God come on let's go let's do it dude anything that's in your brain it's already been learned before dude stop it co-variance yes give me all that walk me through all this please print print print okay walk me through all of this now it's the last part of the prints of my code my code your code our code my code your code our code our code so let's look at it here let's close this up a little bit tieing up tieen up close that yes okay here we go we should get it here soon where you at let's see it why you moving so slow she's tired she's tired all right certainly let's walk through this output in the context of a hidden markof model for Bitcoin trading these statistics describe the characteristics of each state identified by your model for each state we have mean the average values of the feature in this state covariance how the features relate to each other in this state the features appear to be in this order returns volatility and volume change state zero mean okay slightly positive returns below average volatility in state zero and slightly decreased volume covariance shows low correlation between features interpretation this could be a calm or sideways Market State returns are small but positive volatility is low and volume is slightly decreasing State one the mean slightly negative returns above average volatility increased volume covariance shows stronger correlations especially between volatility and volume interpretation this might represent a volatile or uncertain Market State returns are slightly negative but volatility and volume are both up significantly state two strongly negative returns decrease VA extremely high volume increase covariance all values are 01 which is unusual and might indicate a computational artifact or a very rare State interpretation this could represent an extreme event or Market shock returns are very negative volatility surprisingly drops and there's an enormous spike in volume trading strategy implementations implementations implementations in state zero use Conservative Strategies look for small quick profits be prepared for potential transitions to more volatile States in state one Implement risk management strategies due to high volatility look for short-term trading opportunities in both directions monitor for potential Trend formations in state to be extremely cautious this state might represent rare extreme events like Market crashes or major news events consider implementing stop losses or hedging strategies be prepared for quick reversals okay I want to see that though where it how can I see where they because I can't see it on the chart maybe we can maybe we just need to make it bigger let's make it bigger can you see this chart can you see it I hope you can I don't see any green up in here I see no green all right green green green green Green where you at Green I don't see any green maybe I'm color blind maybe I'm color blind but that's okay so where do we go from here I feel like these states need to be chopped up a little better so um below attached is my code attached is is my code I know that um Jim Simons said there were like eight plus regimes should it we have a regime uh state for at least six then I can think of a few please add them in and walk me through the changes okay so let's say bullish trending bearish trending uh sideways consolidation upward consolidation idation downward consolidation uh consolidation uh downward capitulation and upward capitulation right you got more dude 1 2 3 4 5 67 so you got bullish trending there's not that many things in in the market right I don't know I don't know dude I don't know anything I'm just here I'm just here with you my dude me and you every day around this time or another time I don't know just when you see me dude come on get up on here all right so there you got like bullish you got bearish uh you got like sideways upward consolidation but then in in here it's like downward but then there's a capitulation downward there's a capitulation upward I guess I need to describe these areas yeah yeah let's do that okay I'll describe it a I'm not going to describe it um can use volume I wish I had the [Music] uh I do dude I was going to say I wish I have the liquidation data but I do let's just see what it does first I'm just curious I'm going to put that on the Note sheet though put Liquidations in there these could be good to help identify points of capitulation upward or down okay because I got volume um I could also put things like open interest inter interest funding rates okay now we going now we got the wheels turning B open interest what else okay that's good for now um here's my code 144 lines look at us look at us go there it is all right let's uh let's let it run I'm gonna make a this is the end though I'll see you some other time this is long enough we I've shown you enough we are diving into unsupervised machine learning for trading does machine learning work for trading can you predict price well I'm sorry to not bury the lead but no you can't predict price maybe maybe if you got some hidden model but the way I look at it from testing this stuff is if we're all using similar models they're all available with Claud or gbt or whatever if we're all using the same models to predict price well that price will be predicted then by you you that dude over there your grandma me and then that price will not be the price anymore because it's already been predicted by thousand thousands of people so thousands of people run these models I don't know how many models there are 20 30 50 100 whatever like we tested them all right collectively predicting price the next price I don't know if that's the way maybe though maybe if you build some unique model I don't know I'm kind of thinking thinking thinking this through with you as always because that's what I do here I just show you everything I'm doing live and what this has led me to is well what was Jim Simons using well Jim Simons he was using hidden Markoff models okay so let's build a hidden Markoff model and then let's watch everything about Jim Simons okay that's what we did that's what we've done that's what we're doing I've got a bunch of notes here and a bunch of ideas and we have a couple models here and you can see we even tested some out of sample data so we have multiple models here I think this was the template model so so this was the first hmm with just putting in the returns volatility and volume change okay and then the next one we put in as the features we put in these features here so we added some more returns volatility volume change BB width RSI and ema2 okay and then this one we tried some other things you can see the states we're trying to predict we're trying to predict seven different states but the thing with hidden Markoff models so far and excuse me if I or any of this I'm not I'm not like a gilfoil type senior engineer here I'm a womb engineer I'm still in the womb so bear with me gil foil but the thing about hmm is it doesn't actually know what the state names are so we actually have to rename them after it finds the states it finds different states different regimes and that's kind of what Jim Simons was doing supposedly but do you think he'd actually tell us what he's doing no way but maybe we could put the pieces together that's my my hypothesis here we we are just passing in returns volatility and volume change top three Indies these were the top three ones volume change is taking up most of it though and how can we know that well when we run this we can see that it's mostly getting weighted to the volume change I'll show you some cool things here first let's show the plot it's making so it's fitting the model this might take a second take a little bit of time it's doing its little AI thing machine learning thing it's it's fitting the model boom all right saving plot perfect perfect and it also plotted the plot so you can see these are the different regimes here it colored them but the thing is it doesn't know the names of these regimes so in reality we got to go back and we got to set up these regimes ourselves I'm sorry baby I've got a dog here that constantly stands right under my foot so I almost just killed K her that was very very scary you okay baby girl I didn't put my full weight on her thank God that I know she does that because she likes to I I stand on one foot a lot of the times while I'm cating and then she'll go right under my my other foot that I come down on and I'm a big boy you know that's a big that's a lot of way for a little dog so I almost had a heart attack right there I'm sorry for the for the side show but this is life and I'm sure you got kids or dogs and things like that but anyways here you can see the seven different regimes we have bullish bearish sideways but like I said it doesn't know the regimes it just cuts up the data this is Bitcoin data here for a few years and now you can see this outputs the the analysis the analysis has been completed not fully though you can see the state prediction is 89% but we already know all this I'm just doing a little review because again this is fairly new to me as well and every day I just go live and show you what I'm working on so it's still doing some other things here and I can show you the final results we have thus far if I go to the read me I've been tracking them here so down here we you can see this hmm the seven states it's our first attempt and it had 87% State prediction accuracy the log likelihood was negative 15,000 you can see this one is done now and this one is uh you can see what I was trying to show you is volume change is taking up all of the importance can you explain this so I'm I'm using H&M mm hmm with three different features and it converges to volume change being the most important feature I'm using an hmm and this is the output can you explain can you explain what the importance means in my mind this means that this is the most important feature it's pretty self-explanatory but I just want to make sure that I'm correct the feature importance analysis you provided shows the relative importance of different features in your hidden Market off model let me break down what these numbers mean volume change 94.5 1% Bullinger band width 2.72% volatility 2.77% these numbers represent the estimated contribution of each feature to the model's prediction or performance the value sum up to 100% indicating the relative importance of each feature in your case volume change is by far the most important feature accounting for 94.5% you can see that here it's 0945 so we just got to do a little math not really but of the model's predictive power this suggests that changes in trading volume are the primary driver of the patterns or States your hmm is detecting volatility and BB width which I assume stands for Binger band with have much lower importance this analysis suggests that hmm is relying heavily on the volume changes to make its predictions or determine State transitions while other two features play a much smaller role it is important to note that feature importance doesn't necessarily indicate casuality cause causality sorry but rather how much the model relies on each feature for its predictions High importance could mean mean the feature is truly more relevant to the problem or it could indicate that the feature is easier for the model to use in predictions so what I want to do here is I want to look at some of our notes and these notes are going to show us the differences and one of the tests I did was since you can see here uh where is it I think it's one of my last ones I said no volume since volume continuously took up 90 some per 94% here of the importance 93% of the importance here and 97% here I took it out I took it out because I wanted to see something a little more a little more what do you call it balanced yeah a little more balanced so I took it out here and you can see this made it way more balanced and you can see it actually increased to 95% here so it was before 89% the state prediction accuracy and that prediction accuracy is how accurate is it changing states in predicting changing states so when the CH colors change how accurate is it in and uh deciding predicting the change and again I'm not the best in the world at this stuff guys so you know bear with me I'm kind of showing you as I learn as always that's what this whole channel is so every single day I just go live and I try the hardest things and try to understand it uh I was so scared to learn how to code when I got started dude I I waited 15 years I was in t typing on this computer fully in Tech without knowing how to code for so long dude just cuz it's scary but I now I'm at the point of life where I need scary things I need hard things so I just go after them viciously every single day here live so then the the common thing was or the obvious thing to do was Let's test all of these models I think there's five different models on out of sample data data that the model has never seen before and you can see here are the results state zero so these are the two best ones I have so far one has volume change one doesn't and these are the results says model one has log likelihood of negative 65 6956 and this one has 3890 so I've got these two models explain these datas I've got uh I tested I tested two different uh hmm models and the code uh on out of sample data models on oos and the code is attached below are the results and the plot please watch walk me through them so I understand okay let's put the code the code in there as well copy okay and here's the code okay and then also I want to put this screenshot in there and do you think I saved it here no I did not I just want it to to have everything everything I got and I know I already reviewed this actually yesterday but dude this is the only way I can learn is with repetition 500 shots a day baby 500 shots a day I've been doing that since I was a kid come on you think I'm not just going to do the same thing over and over again till I learn it of course I am worked for me in the past it's going to work for me in the future repetition is key just keep attacking every single day four hours a day 4 hours a day if you're not doing 4 hours a day then don't talk to me don't ask me me questions because all your questions can be answered in four hours a day we got AI my guy don't tag me don't DM me nothing if you're not putting four hours a day of focused work at this goal to automate your trading don't hit me about nothing about automated trading nothing I'm sorry dog we're competitors at the end of the day but I'm showing you every single thing cuz I believe code is a great equalizer for me it is at least can't stop me now dog I know how to code I can build anything you can build anything we got AI my guy thank you for providing the code and results I'll walk you through the key aspects of your hidden Markoff model analysis let's get these up model comparison you've tested two hmm on out of sample data model one uses Binger bands volatility and RSI as features model 2 uses volume change Binger band width and volatility as features so the log likelihood scores did I write this down in the out of sample data no I didn't unfortunately that's okay we'll write it down right now so model 1 has - 6956 model 2 has 3890 the log likelihood is a measure of how well the model fits the data higher values indicate better fit model 2 has a significantly higher log likelihood suggesting it fits out of sample data much better than model one visualization the plot shows three panels top panel bitcoin price over time this gives context on the market conditions during the out of sample period middle panel shows the I need to get this up if we're going to I need to get the um what they're talking about up or else we can't understand it can't understand anything you can't see even if it's in your mind B you can see it in your mind your mind doesn't know what's real or not because your mind is real so if you can see something in your mind you can have it I believe I can fly I believe I Could Touch the Sky I think about it every night and day spread my wings and fly away I believe I [Music] can I see me running through that open door I Believe I Can Fly D you got to believe it before you do it D the log likelihood okay here we go um top panel is the Bitcoin price over time just so we can see it this is the out of sample data here middle panel model one States shows how model one classifies different time periods into seven distinct States 01 through six each color represents a different state bottom panel model two State similarly shows State classification for model two this is interesting so while the price is here you can see how the state changes it's kind of hard to see to be real but State transition both models show frequent State transitions indicated by the changing colors in the lower two panels model 2 appears to have more distinct and longer lasting State periods which might suggest it's capturing more persistent Market regimes feature importance given that model 2 performs significantly better it suggests that volume change which is in model 2 but not in model one is a crucial feature for predicting bitcoin price States this aligns with the importance feature analysis you shared earlier where volume change had the highest importance model stability model 2 seems to produce more stable State assignments longer periods of consistent colors implications implications the superior performance of model 2 suggests what volume changes are more import informative for predicting bitcoin price than RSI the frequent State changes in both models reflect the volatile nature of cryptocurrency Market to further understand the models you might want to analyze the characteristics of each state interpret look at transition probabilities between states consider how you might use these State predictions in trading so analyze the characteristics of each state okay how can I make this how can I make this more robust and test more features I was thinking I can add in more indicators I have un limited from pandas ta um and ta Li I was also thinking about trying just four states and what were those four states somebody shouted it out the other day so much love to them and those four states he said that Jim Simons looked at we those four states dude let's go look at our notes Here notes are quite dope four volatility regimes this was the comment on four states okay let's just let's kind of IDE ideate ideate yo 777 Smokey what up bro what does each state mean the different regimes uh you were flying bro I saw it happen uh puppy is Gucci Gucci Gucci bless you thank you thank you thank you apple apple apple little apple little mananita okay your ideas for improving the robustness of your hmm model are excellent let's explore how excellent you gave me excellent hey shorty you ain't never said hey I'm excellent dog I don't I don't care I'm not I'm new to this game dog and I already got Excellence from a a AI my guy is this excellent is this am I excellent dog I did not know that that is the biggest compliment I could ever get from Claud I think I'm on CLA forever yo they probably coated that in they're like yo juice them up juice them up hey hey hey if moev ever gets on claw juice him up say he's excellent all the time that's crazy excellent I never heard any AI tell me excellent okay let's explore how you can enhance your approach excellent GE your boy ain't excellent I'll tell you that much we did the math the other day I've been at this for three and a half years 3.5 * 365 days time. 85 CU I took some time off 1,000 hours Time 4 hours I'm only at 4,343 hours I ain't excellent bro you know who excellent is Jim Simon's excellent rest in peace my guy let's explore how you can enhance your approach testing more features leveraging panel ta and ta lib gives you access to a wide range of technical indicators here's how you could approach this add more indicators Trend indicators moving averages SMA EMA magd P parabolic star momentum indicators stochastic oscillator CCI volume indicators on balance cool cool ATR nice keler channels nice feature selection after adding more indicators you might want to use feature selection techniques to identify the most informative ones consider using methods like correlation analysis PCA random Force importance try and force state Jim Simon's insight about four volatility regimes is valuable here's how you can implement this modify your hidden hmm initialization to have four components instead interpret the four states you could potentially interpret these four states as low volatility Rising volatility High volatility yo that's F four that's four making the model more robust cross validation instead of single train test split use time series cross validation to ensure your model performs well across different time periods Ensemble methods consider creating an ensemble of hmm with different feature sets or parameters regularization if overfitting is a concern you can add regularization while hmm doesn't directly support this you can implement it by adding a small constant sensitivity analysis test how sensitive your model is to changes in the input features or parameters this can help you understand which aspects of your model are most crucial back testing Implement a back testing framework to evaluate how well your hmm States predicted predict future price movements or volatility okay so these are pretty good ideas honestly I'm only saying that because she hyped me up though you know it oh yeah you got some pretty good ideas is gbt or Claudia but in real you know is just cuz she hyped me up and you know I'm a lifetime I'm a lifetime now I'm a lifetim now to do next next things I can do to improve hmm dope dope dope okay so get him out of here get that BTC little boy out of here 150 racks 150 racks I hope that's not you dude I hope you're here instead tap in 777 Claud B glazing yo that's fire B I don't even I've never heard that in my life but she do be glazing be CLA be glazing y'all keep me hip I'll tell you that much y'all keep me hip CLA be glaz in be I'm about to start saying that all the time she be glazing be I be glazing too let's just double check make sure that's not something bad Urban de dictionary glazing the act of f no when you are meat riding someone oh I don't be doing that I don't be doing that I be doing tomor so it's me kby glazy okay okay okay see this is old man old man trying to be hip over here he trying to be hip saying the word hip ain't hip dog she be glazing that's good I need to be glazing all the time that's how she going to keep my subscription she better keep glazing this she hey what year were you born dog that's crazy that's crazy I love it I love I love how Lang language changes I love it I love how slang changes and I feel like we're going through a pretty big shift right here for us old heads yo Brian says hello sir I wanted to ask the mentorship I can use code build trading Bots like Jim Simons you got to you got to build it yourself for sure but um the boot camp I think is what you're asking about the boot camp will help you I've been doing this for how many hours did we just say I've been doing this for 4,300 hours so I have a boot camp that I try to just share the things that I know most and the things that work for me and the pieces to the puzzle but at the end of the day everybody's puzzle is going to be different Jim Simon's puzzle is going to be much different than my puzzle my puzzle is going to be much different than Brian's puzzle Jane's puzzle is going to be different than Brian's puzzle so it all comes down to your unique Ed your strategy your approach to the markets but I can show you everything in between step by step how to automate your trading how to test if it worked in the past gives you a better hope that it's going to work in the future but it's not guaranteed and uh yeah everything's in the boot camp 69 bucks stupid cheap high vow you can read what others have to say that's it landlord says are we continuous in the hmm from yesterday yeah yeah dog yeah we send me back send me a simple script that will print out all of the indicators for pandas ta and ta live yo Brian says saw you your channel and you're a King sir I appreciate you dog that is so much love much love to you dude glad you're here I'm so so happy you're here let's say uh all indicators all sh no all indicators please AI my guy ai ai dude this is over it's over for me and you all of the indicators yo all of the indicators you see that dude you want that script come with me dog I'll show it every day dog I'll show you everything everything output them to a output to a txt in my data folder yo I don't care who you are dog I don't care if you've been coding for 30 years dog I'm coming at your neck every single day I got the power now we got the AI my gu stop it stop it do you see this it's just one idea and boom I get every single indicator just like that I don't need some long tutorial I don't need a c stack overflow I don't need to go to school for 82 years like you did dude we're not the same I go harder than you you can't keep up I'm sorry yo print them out too cousin whatever that's cool so these are all the indicators here for ta lib pandas ta indicators so we have all these I'm going to use adx yeah let's do that let's do that so I'm going open this to the side here split right show all indicators sorry burped in your ear my bad that was super in poite yo Mev you need to start using multi-processing and multi-threading why it's good comment though it's good comment but why do especially when accessing large sets of data okay okay I seen somebody play with that the other day up in the Discord just so I can test faster back test faster multiple times multiple multiple processors I guess I like the idea I appreciate the share much love to you for that you can use Python and C++ e for all right so I'm going to go ahead and start writing some of these ideas now so test two test three Indies to try so I want to try the adx because homegirl talked about it a lot um one from my first podcast I did uh Mansion says I use multi processing it's much faster to train the model okay there we go that's a good why that's a good why that's on track use multi-processing to train models and also to back test faster appreciate you guys this is like you see that right there like it might have taken me two years to get there and y'all just dropped a dime like that you just Dro you just dropped Mev a dime you threw the alleyoop and I just dunked it I haven't dunked it yet but I'll dunk it soon landlord says he wishes he had a supercomputer Connor says uh was using my uni one for the last couple years and I miss it dearly dang dude what's the supercomputer looking like Connor says plus one for the faster training thank you landlord thank you appreciate all of your your Brilliant Minds I don't know a lot I don't know most honestly I only have 4334 hours in this game so far and um quantum computer dang I got the I got the quantum MacBook no I'm just kidding let's see what other indicators we want to try I'm just going to use ones that are um I guess I don't know that's the hard part like which ones do you use adx try that I just the ones that I guess I know ATR there's just so many Don Chan I'm just based off memory of tests I've done in the past or ideas I've heard I'm just going to use those I guess and if you got any ideas that you see here holler at your boy just be like yo Moon try that one Lyn regression macd yo how much like RAM and stuff does a supercomputer have I'm just curious to see like yeah how far off I am I know I don't have anything near a supercomputer but like what is yeah cores 2 I guess again I'm not very I'm not very technical I'm just kind of a new but appreciate y'all yeah I want to Cur I'm curious how far off I am prom super computer status what is out there and then can we use gcp Google cloud or GC whatever to get supercomputers is it worth it I'm curious about all that stuff but this is fun this is fun true range I love this game game dog this is just this is my element here feature engineering this is where I'm at this is this is my game right here come up with ideas I love it I love it I love it ta live they got all the math here cosign seal RSI Stoke Stoke RSI how about that what's Kama Kamala Kamala CDL CDL engulfing what are these I don't even know what these are D CDL is chart something doy Hammer let's put it let's just try it I don't know dude I'm just trying everything dude this is the game right I don't know we'll see see I mean we'll see if it's better or worse ly standard DV T range ATR we got the ATR in there already the VAR the obv you down with obv yeah you know me candlesticks CDLs got it thank you bro um so Connor says it was mainly the GPU access training models was like day and night when looking at using supercomputer versus not angga says you can use tpus tensor processing units in Google collab nice landlord says you can run your scripts on AWS and gcp but you pay for the increase in Ram I think I'll stick with gcp for now if a Candlestick is 100 it's bullish if it's negative 100 is bearish good Intel bro for you good I didn't kill you over there did I okay let's try this um I'm going to say top three Indies no no no do not there top three Indies is this use the volume change okay let's say update this code update the attached code to not use the features or the yeah features it currently has but use the ones below from pandas ta I think that's the first one right let's try these three yo is this is this a approach is this how people approach this stuff or are there like smarter people out there I found that three is better than seven so Myra says you come from data science background Munda I mean I don't know not really not at all actually so I learned how to code 3 and a half years ago but I learned through watching like 10 machine learning courses so I've seen everything not everything but I've seen a lot of machine learning stuff I was just super fascinated by it but I also was learning how to code at that same time so I was just typing all this machine learning code out in Python and kind of learning python at the same time I don't think it was the correct way to do it but I've seen it all now and now I'm three and a half years deeper and I actually know how to code in Python I can build any Bots anything like that build a bunch of alos that I run every single day and now I'm diving back into machine learning so I think it's going to click it has been clicking a little easier so if you want to say that's a background then yes but no I didn't go to school for any of this stuff dog I did not go to school for coding machine learning nothing I didn't read a lot of these boring books and try to absorb what I can so all this all everybody commenting here you guys are Lifesavers bro teaching me what you know because I know we got some guil foils in the audience Connor says for individual sessions I could run I could request up to 72 CES dang wow so 72 time eight I can't do that math bro all right so I'm way off I got 96 gigabyt of RAM on my computer though altered says I want to enter your community I don't speak English is it possible to add French language or Italian dude you literally just said that in English you do speak English give yourself some credit first off but yeah um yeah everything's got subtitles for you bro so you can pick French or Italian that's cool you know French and Italian that's amazing but yeah you can join the boot camp everything in the boot camp has subtitles to every language pretty much so that's dope uh Myer says that's cold straight to ml had to go for the top man had to you know that's that's who I got inspired by it seems super hard super fascinating too and it was before AI was here well before llms were usable from the desktop like they are so I'm glad I learned it kind of by him before I just demand AI to do everything all right I just had to finish my little Waffle there what do we want to call this one adx ATR Don chin adx ATR Don I like that adx ATR Don Chim let's just pop it in let's pop the code in dump it too yeah dude dump it I want that update this though I want to say train models here data I want to make a folder this make this a little neater trained models and let's put all of these models here copy path update the code so it I'm not going to do that just going to do it myself dude did I put it in there already dang that boy nice that boy nice look at this I can type code in the in the thing dude all right let's see what it does run it do fit in fitting somebody just got licked we got licks always let's see what's going on today oh my God people getting wrecked today I'm looking at the 15minute chart that's why this taking a long time to fit or whatever there it is there we go so be in the market in green and out of the market outside of the green I like this in the market green out of the market outside green yo call trading what up bro oh is JPY going crazy again what's going on over there Connor was training models to predict gra gravitational waveforms of binary black hole collisions on the supercomputer very resource intensive stuff not as applicable as I'll go trading though oh man that sounds intense though okay so this is interesting data so let's go ahead and analyze it we just got to do repetition dog I don't know where you're at in your journey but for me I just need reps I need reps reps reps reps reps reps 98% likelihood wow I'm curious about the importance here count okay that's fine I was like why are they all the same it's still working here nice to see you here again Call of Duty what is happening is the expectation over the September meeting in fomc 25 bips versus 50 bips and inflation is running relatively high in Japan which leads them to have to increase rates they will do so slightly but what matters most is future expectations over falling rates in the US interesting okay look at this I like this too because this uh see how it's pretty even I don't know if that matters but and compare it to our best model below and what was our best model below is the question damn um hey it didn't update into our models uhoh ibly just overwrote it huh h I put it in the data folder yeah my bad it's typically me that's what I love about coding when something goes wrong it's typically me it's like such great accountability and then I want to compare it to let's see where are our other hmm models here noall dang I don't know how to tell which is the best one I can look at the out of sample so out of sample data here and X2 X1 so X 2 model 2 is the best volume change BB width and volatility Vol BB and VA I love this game dog you'll never get me to stop I'm sorry you'll never get me to stop this game I'm sorry BB volatility and RSI I'm sorry dude I'm not going anywhere for 60 years at least I'll be standing my black ass right here for 60 years at least BB width volatility and RSI how do you evaluate Indie importance I don't know dude it says right here it's a good question though um and then compare it to our best model which is model two so I just don't I don't know any of this yet I'm I'm getting there though I'm getting there I'm starting to understand a little better a little bit better every day every day everywhere uh it's Jan says is each console log a unique trading bot no it's not right now I'm working on the hmm which is a hidden markof model which supposedly Jim Simons liked but you know that boy capping ain't nobody going to tell us what that was actually working but we did figure out yesterday that he definitely starts with data first and if you start with data first that means that you're not looking at indicators and stuff like that you're looking at the data so we're going to let the data be do it Thug fizzle you know we're going to start with data First Data first and that's what we're doing now we're going to we're sending into the the AI the machine learning what's this JPY you're talking about I don't even know how to get there let's see if I can get there JP I I'm not a I'm not a Forex guy I see oh boy is this what's this mean what does up mean is that bad or good because I know with these pairs and stuff cuz the other day didn't it like okay the JP I capitulated down here honestly I do not know I need to get a little better at currency understanding I think cuz it looks like it's important but see if this is done yet being I feel like you need to understand currencies and how they move USD JPY oh so this was opposite oh that's a different story so was that just backwards did I just look at it backwards this is what yeah okay yeah this crashed the other day but that doesn't look like a backwards chart does it JPY I don't know JPY USD uh that's not backwards though is that a different thing sorry for my ignorance again appreciate you guys' love though okay so it's the inverse okay so this down bad today huh I should add this to the list here what are some other good ones I should start looking at that like move the economy let me just add them to the list because you know I canceled this you know I canel the uh trading view there's no reason for it what are some good symbols I should watch in order to get more in tune with Forex because I know a lot of you want Forex over here and um I've heard that Forex is more mean reverting so it could be interesting to look into but let's go ahead and analyze this data so we got two new models done or one new model and I want to keep just working through this until it just so so clear okay yo thank you for this so we have we've got some good suggestions as well I know I'm always doing this but always getting over here so Euro USD that makes sense just got to get them on the uh the thing before it cancels on me so Euro USD what else GP GBP GBP the British pound GBP USD and USD JPY those are the main three sweet yo I appreciate you dude caller trading and landlord I appre appreciate both of y'all so these are the three movers so Euro USD GBP so the British pound who uses the British pound Britain oh the UK okay so Jane uses the British pound and then everybody else uses the Euro right so I see why this is important is there not like one for uh China China's dollar and stuff whatever that is is not important or something Jane uses the British pound everybody else in Europe uses the Euro Japan uses the JPY I feel like the two missing from an outsider standpoint are the Yen yeah the Yen and then also India and Russia those are all the big ones I don't know if those are tradable though so it's interesting Swiss frank okay dang so yeah I'm not going to go down that that rabbit hole because I guess there's a lot more not everybody uses the Euro I was thinking everybody uses the Euro but I guess not like everybody on over there over there where y'all at everybody over there uses EUR Yan yen is Japan oh yen is you w you w anyways let's go ahead and analyze this data because we got a new model here and this is what it looks like it looks good to me did you see it look at this look at me 98% prediction the thing is the whole system's all dollarized so the majority of the flow is dependent on the US the Swiss frank is interesting because of how tied into International banking it is wonder what info could be inferred from analyzing that dang UJ is going to fly higher in this coming week what is UJ is that Euro or us Japan dxy is a good indicator of USD strength I think I got that right here yep dxy so dollar down risk assets up dollar up risk assets down that's what I learned at least I don't know if that's right but dollar down today and risk assets down so down BTC down interesting this game is super fun because there's so many different levels to it in different Arenas like y'all are over here in the uh Forex Arena and I haven't touched that I read baby Pips when I was in college maybe or whatever and I I was bored to tears I was bored to tears dude I was bored to tears that baby Pips there's so much information there but I probably could I probably could dive back in now that I'm a little more financial Savvy I guess that's my interpretation too Connor says it is half right it's half right until like a big down day like this I feel like then it's like everything just hits the fan that's what I've observed at least is that like on days that it's slightly down risk assets up if it's mad down then everything down dxy is a basket of currencies against the dollar so if people are in Europe and want to invest in US Securities they need to raise the US dollar interesting well let's stop let's stop talking about that stuff because I I don't know where that's going to take me so maybe in those days there's a rotation from Euro to USD dang it's awesome there's so many games in this game that's why like I'm not worried about allocating time to algorithmic trading like you you allocate time to like new trends like Shopify or apps or I don't know social media you don't know if it's going to be here in 5 years I needed something that was so hard and I know it's going to be here for the next 60 years and I know for a fact there's going to be some sort of Market here for the next 60 years so every single minute I put into this stuff I'm compounding and I know in the future it's going to be useful I'm getting better and better every day and thankful to you guys that like accelerate this stuff too so if you're in the chat you see everything I see I wouldn't miss it I wouldn't miss it if I were you thank you for providing the results of both models I'll compare them based on the given metrics and features State prediction accuracy new model adx ATR and Donan 98% model 2 volume change BB width volatility 89 the new model model shows higher accuracy in predicting the next state new model log likelihood minus 93,000 model 2 plus 14,000 model 2 has a much higher log likelihood indicating a better fit back to model 2 Bic basian information Criterion damn 187,000 verse negative 27,000 lower Bic indicates a better model model 2 has a significantly lower negative Bic suggesting it's a better fit cross validation score new model -47 47k verse 20 2000 model 2 has a higher and positive cross validation score indicating better generaliz ation to unseen data import feature importance new model adx 27 85 ATR point4 I like how it's a little more spread out but model 2 is all about volume essentially visual inspection the new model's chart shows more distinct and prolong State periods especially in the later half of the time series this could indicate that it's capturing longer term Market regimes conclusion while the new model adx ATR Donan shows higher state prediction accuracy and more balanced feature importance model 2 volume change BB width and volatility outperforms it in terms of log likelihood Bic and cross validation score this suggests that model 2 is likely the better model overall as it fits the data and generalizes more effectively to unseen data the heavy Reliance on volume change in model 2 appears to be capturing something fundamental about bitcoin price movements that others don't however the new models more balanced feature importance and distinct State visualization might offer different insights using model 2 is a primary model keeping the new model as a complimentary tool for additional insights all right this is dope dude I love this game cuz you know what's next a new idea and then you know what's next after that a new idea I don't care if I miss the shot bro I will just keep shooting I will keep shooting I know I'm a shooter I'mma keep shooting I'mma keep shooting dude I'm just going to keep shooting and keep shooting and then I'mma shoot again and then I'm going to shoot again and I don't care how much time this stuff takes dog cuz I'm here for 60 60 years B I don't care I'm here present do you use volume as a percentage change yes sir volume change all right let's try another one use these three new features instead of the old ones in the attached code they are all from pandas ta yo I haven't watched pton fo probably for 20 years though so I don't know what I'm even referring to paid and full when did this launch yeah 20 2002 probably watched it when it came out or around that time I was a kid I need to watch this again as an adult because you know all that stuff goes over your head all of it goes over your head so uh I started watching this show I grew up on like uh fresh prince of belir I love that show yo they have a new show oh because you said B remind oh okay they maybe that's where I got it from B I don't know but um I don't watch TV too much but I start watching the new Fresh Prince it's pretty it's pretty good A lot of like a lot of references to old black movies which is dope Kan do you care if you live or you die that threw me off I was like I know what he's about to say that's exactly what Kane's grandpa said to him and they just LED this they started the show the new fresh PR show oh it's called belir sorry it's called Bel Air um it started with will you care if you live or you die and that's like you got to be you got to be watching all the old movies in order to understand the show fully which is dope I love it I love it I love it I love it and they give I love how they ALS like totally changed the show like Carlton is somewhat cool um jazz jazz finally swoops his girl whatever her name is uh Jeffrey is like an assassin L key he'd be holding it down it's crazy I love how they just like switched it up a little bit at first I was thrown off like why is Uncle Phil some like skinny dude rest in peace Uncle Phil I like how they did that at first I was like apprehensive towards it like why is Uncle Phil like is he he who is this guy but he's still a great father figure takes care of will I'm only on like the seventh episode so I was happy to see that there's three seasons to it it's a good show and I don't say that about many shows I I don't usually like watch this St a got time B yo I need to watch this movie again though got Regina a young Regina Hall we got Cameron in it hey Ma that used to be my head I used to sing that all day long damn there's some good movies I'm a old head dude there just some old head stuff to be saying back in my day back in my day we had some good movies cast is Elite absolutely damn I can't wait I might watch this tonight I mean I'm sure it's on somewhere Netflix eight more where to watch YouTube I got Amazon Prime thank you you're an territory I know it's uh it's crazy dog it's really crazy I mean what it is what it is but I am I'm there I'm there off officially an unk you never think you're going to get here till you get here dude I'm literally an uncle I got two two little Young Bucks and it's cool it's cool man it's cool to see cool to see cool to see him jack my SCH SW my swag is irrelevant but they Jack it anyways because I'm you know so somebody got to give it to him all right so I'm writing down Pon full real quick all right let's get back to it I'm going to go ahead and test one more so use these three new features instead of the old ones in the attached code Lin regression macd and true range see I like just switching it up entirely so our best one is volume change but we just tried the adx and it at first it looked better because of the 98% it looked better but was it better no way uh send me back send me back all the code all right I love this game dog we literally are competing with the the machine learning people over at name your Fang name your Fang you gave us too much power dude you gave us too much power Claudia you gave us too much power Gigi you gave us too much power luxie you don't understand it okay go ask AI I say CO's the great equalizer but like damn AI is really the great equalizer but you got you got to know how to code you got to know how to code to be able to get the best out of AI because what are you using it for otherwise right like emails and stuff okay I see through I I can see through that you can't see through my code you know what I'm saying like code is code so like you can kind of feel when AI is talking to you almost I mean not always but like how often are you writing papers how often are you writing emails I mean how often are you writing Facebook posts I mean it's cool whatever like that's a cool use case for it but coding is the killer app coding is the killer app for AI right now and if you don't know how to code you can't use it properly I mean you can learn through it at least but you got to you got to be able to learn it so that sends me to let's let's let's actually go through it because I know there are people here that weren't here at the start of this so this one's Lin Rag macd and TR l ly reg Mac and tr. py paste it in and then we're going to just do a little study session here because I need to see it dude like I said I I grew up getting 500 shots a day dog and this is the same game this is literally the same game you get better over time get wetter over time every single day I don't care how tired I am I don't care don't care how little I slept last night I don't care that's when I really double down and I make sure to get my hours in so import these are a bunch of imports then we're starting the hmm analysis on BTC we're loading in pre-processing we're dropping a column because my data is janky we're creating the index okay open high low close this is the linear linear regression with the TA pandas ta Library pandas ta Library so linear regression is a type of machine learning as well so cool we got macd here calculating the macd we're getting the true range here then we're dropping any na any non I'm actually about to go get some mother naan right now B train hmm seven components lyre macd true range I swear there's a new non place I mean it's Indian place but you know you know how you know how that goes train hmm we've got seven components that's seven different regimes we are training on but we have features three features linear regression magd and true range we normalize those features meaning put them between zero and one I believe then we run the hmm the goian hmm are there other hmm that we can run put it on the read me put it on the read me we are using the goian hmm are there others that we can run where we at where we at ly red okay so and now we the Asian's completed we predict States so it's predicting States remember this is the actual prediction here we have seven different states but the thing is is it doesn't actually know the states it's just ch them in the states and then we're going to have to do that later but that's okay I just want to see a lot I want to see a lot and then see what the best is of a lot since I'm starting at a zero point origin where moev knows nothing I want to see a lot of different tests here and then out of those tests imagine you've never played a game before I have no idea how to play this game I'm going to play the game a 100 times and see my score over the course of a hundred times and then I'm going to go with the method of the best score that's all I'm doing that's that's all I'm ever doing that's what this whole this whole world is about to me right now that's what life is about really just a test your way into figuring things out predict next state save changes it's just saving the changes St changes pending it to a folder and we haven't even been looking at them but okay model scaler where are those I can delete these actually need to update them real quick I'm going to update them in my thing so you can see here that I updated the file path so here at the bottom I added a folder to all these so I got to just update this real quick um hmm model 2 here every time I run this it's going to try to save it again so this is our best model so far model one here second best actually I don't know if it's second best but I don't really care about the second best I just want the best be that's it just trying to be the best I can be I don't know how good I can be but been able to figure it out thus far so I'm G keep going plot the results okay that's just a bunch of plotting I just want to make sure you see all this code so you can leave here with your hands full of ideas and code data saved to a plot calculate prediction accuracy calculate Bic Bic is a evaluation technique so something about basy and something something something death time series CV okay analyze feature importance okay starting main execution training the model predicting States these are the names of the states but remember these names are just made up really the model we have to place the states later this is just running all the code but I just want to make sure you see it all try to show you all the code but sometimes I'll be forgetting dog that's my my bad Mal M bro Lyn reg 14 key error come on cousin I showed you all the wrong code then it's not done yet calculating lyre okay I get this air but I can see in the available available indicators that let's see what the available indicators look like these are all the available indicators on Lin rag I have these two options I have uh indicators on Panda ta that I have these options here okay um let's see they gave us too much power llord said I think they did and they can't take it back now was the crazy thing you can't go backwards with this they did give us too much power they gave us kids out here too much power now noobs you can't tell that I'm a Noob anymore like if you look at my code if you look at my code you can't tell that I'm I'm just at the start of the game I'm only at my 3400 hour and I have to get to 10,000 hours before I know anything before I'll ever people ask oh moev get on my podcast tell me this and it's like bro I don't know nothing yet I'm not even halfway there yet stop it maybe I know more than some but 4,000 hours no way I need to get 10,000 hours I need to get to 10,000 hours before I can confidently say I know anything I'm sorry dude sorry to do it to you I'm sorry I'm not your Guru sorry I don't know everything in the world the only thing I do know is I will keep going I will keep finding new Secrets live every single day there's aha moments every single day dog every single day take care make sure you're here I think uh man man Manuel maybe I think learning writing and coding separately are very important like me I'm learning the English language because of AI sick that's Fire Dog you're born in New England so you're learning English because of AI That's fire okay I could see that I can see that learning just in general with AI is crazy like anything you want to learn it knows mostly everything at leasts a structure to everything that's awesome you're learn English and coding this would be killer for resumés too absolutely dude the bay indicator thing is a comparison of the log likelihoods it's used to gauge whether you could choose one model or the other thank you bro so the Bic it compares the log likelihoods which we've seen a lot of and we're about to see it again Bas factor basian basian factor base Factor so that's cool man you show up to these lives dude you learn at exponential Pace because I'm pushing us into a direction that I I'm interested in every day and then there's people way smarter than me up in the chat that are so gracious with their time and gracious with their thoughts and they went to school for this stuff dog I ain't go to school for this let's try it again doggy doggy style sad sad did I not put the right code in h you nerd I apologize for the confusion you're right and thank you for providing the information the error occurs in the Lin R function in pandis ta with different column names okay let's modify it so we need to put the load and pre-processed data we need to change that here are the other parts of the code that need to be updated okay I'll do that all right do it load and pre-process data here beaters yo what up Nick how are you how are you Nick have never seen your name before so welcome welcome welcome welcome in the predict States function so I'm going to go to all the Lin Rags here train hmm here what did it change do I don't think it changed anything anything think is plenty man that looks guzy to me beef all right let's just run it see what happens send back full code with any prints you need to debug my dog does unknown says I upgraded Twitter to get grock but can't handle heavy coding challenges wonder why PO is the best amount for free interesting Nick says he a newbie here salute salute to you dude much love 777 how pass Love Good Vibes all the time you'll see it all the time in the chat light the chat up with the 777s for our new friend Nick so much love so much love all right let's go ahead here and try this new code out see what you got Litt Mama nothing linear regression calculating time series dang we're stuck this is good this is good exercise though good exercise Jane says 777 welcome to the fan fam fam wow dollar huh look at that dollar we look at the dollar now BTC this liquidation Point still got me tripping here 150 mil liquidated right here going downwards came all all the way back up did a cherry kiss to it and then reverted right back down and then jump got liquidated right here that's wild that's wild that's all I've got to say though because I don't want to predict Market movements or anything like that here it's not what we do here dude we let the AI do the work yo I like your pfp bro Nick it's a good one yo what happened to the live stream yesterday I cut it off randomly didn't I I forgot I had a hard stop at 3 and then it was 305 West Side see what she found out here I apologize for the confusion it seems that ta Lin red function is returning a series instead of a data frame run it back then there we go shorty we can do it all dog we can do it all where's the m d at though okay there we go beautiful looking chart I'll tell you that much beautiful all right let's analyze it do exactly what we did with the last one repetition is key dude for me and this is my world I'm sorry if you're way ahead of me and this is too easy for you but for me it's a challenge I like I like challenges but I got to understand this stuff fly FY because I got to be able to grab it by the mother balls and then just understand all okay you know if I understand it all I'm a threat if I don't understand it all I'm not a threat dude I'm just following in the footsteps of Jim Simons he he did this till he died dog that's amazing it's amazing I'mma do it till I die too I'm do it till I die I'm just waiting on this stat these stats to be done here and I actually think I have my um best one here show all indicators the if I got to read me here I wonder why it takes a little bit of time in between this okay here we go let's see if she can just compare them to my model to to the previous ones let's just not even give it to her let's see what she really about let's see what Claudia is really about see if she can really memorize things nine more messages get off my D you know I'm running it up every single day nine more messages why do I even pay you fool what you even pay for who even pays for this stuff dog I do dag NAIT Jiminy Christmas certainly I'll compare the results of this new model using Lin Rag macd and true range with the previous two models we discussed let's break it down by each metric and let's get that thing back up here let's get the image the image open revealing finder please plot it plot It Like It's Hot Leg is hot baby my leg is hot yeah you don't get much I mean you get a good amount Jane but you can see I only got eight more messages left for like three more hours this crazy stuff so I think I'm going pretty hard on it today but um I feel like it's less than yesterday I'll say that much feel like it's less than yesterday I've been using kote says I've been using open AI API plus pine cone DB to avoid the rate limit okay okay I'll feel you what is pine cone DB oh the V vector vector bab I remember this tripo yo I got luxie so I'm Gucci like I got a backup I got luxie is you know unlimited so she's my backup and she does great it's using all GPT stuff so I'm Gucci on that on the backup so I was just complaining to complain cuz but I appreciate all the tips because this is how how we find out about new stuff you know okay my monthly spends looking pretty solid here on gbt I was trying to keep it down a little bit but you know it's all good we're going to use luxie here when we get worn out of or we're done with the whatevers so here's the model here that we just tested and this model uses adx the ATR no sorry new model is linear regression macd and true range 97% and we're comparing it to our other model or other two models now previous model was 098 so that's better there model two though which is our best model has a lower prediction State prediction accuracy the new model performs very well in terms of State prediction accuracy almost matching the adx ATR daan model and outperforming the volume change model new model is 30,000 minus 30,000 and the other one was minus 93 so it made an improvement model 2 is 14,000 the new model shows a significant improvement over the adx ATR but still doesn't reach the positive log likelihood of volume change that's funny our first one of our first model or second model Bic Bay in information Criterion new model 61k previous model was 187k and then model 2 was minus 27k the new model has a much lower which is better Bic than the adx ATR Donan model indicating a better fit while accounting for model complexity however the volume change model still has the best lowest Bic cross validation score new model is- 100K previous model is - 50k and I think this is the important part the standard deviation model 2 is 2000 still the best the new model shows worst cross validation performance compared to new previous model with a lower mean score feature importance new model Lin regression um 47 so this doesn't matter too much it shows more balance which is nice visual inspection the new models chart shows more frequent State changes compared to the previous ones particularly in the earlier part of the time series this could indicate that it's more sensitive to short-term price movements conclusion the new model linreg macd true range shows improvements in some areas particularly in state prediction accuracy and log likelihood compared to the adx ATR Donan model however it still doesn't outperform the volume change the new model's feature importance is more balanced than the volume change one which might provide more diverse insights into Market Behavior the frequent State changes in the visual representation suggests that this model might be more reactive to short-term price movements overall while the new model shows some promising characteristics especially in its balanced feature importance and high State prediction accuracy it doesn't conclusively outperform the volume change model across all metrics the volume change model still appears to be the most effective in capturing the underlying dynamics of Bitcoin price movements okay sick well let's keep it moving then let's try it again I'm going to take all this information just plop it in there because why wouldn't I just plop her in plopped it in perfect perfect I'm getting comfortable with this I can see ly regres macd true range like sometimes this stuff's so intim then you can't even see it you know you can't even look at it like don't even look at me don't yo you are so complicated I don't even want you to look at me right now CL no don't look at me stop looking at me dude could you imagine if I was on camera there's no way the show would not go on if y'all fools were looking at me this whole time I would stop it stop it stop it stop it stop it stop it y'all would fall in love with me you already fall in love with my voice voice 97% males in the audience by the way weirdos oh my god let's go ahead and try some more so what I'm going to do do here is I'mma paste this in the things I do without actually thinking I did is crazy you see that over and over again like moev you already did that look at this look at this professional looking file dude this is a good looking I'm GNA push this to GitHub right now this is just a good looking non moev like dang really cool really cool I remember back in the day when I saw folders I would just oh pretty much just puke I pretty much just puke because it's so intimidating looking at this code over here and then folders of the code how do you even oh man that's why it's hard to work on other people's code bases in my opinion that's one of the reasons what is this folder called hidden Markoff model rest in peace Jim so much love to you dog you don't understand how inspiring you are dog appreciate every every thought you put out every thought you put out in the universe I know you're not going to show us everything and now we'll never know but I feel like you gave us enough enough pieces I'm grateful I'm grateful for him thank you Jim rest in peace 777 to you your loved ones I hope there's somewhere amazing that you had it after death and other gy too I love you like a pops to me much love Eerie that you both died in the same year Eerie Eerie stuff when I say Jim I'm really talking to both y'all much love pop let's go ahead here and let's try these new ones so instead of using the current indicators in the above code use these three from ta all right and one of them is a Candlestick that's weird I don't know if it'll work but tripo metac CLA GPT did you check grock NOP nope so the vector database I remember that from uh um Auto GPT days Jane says grock many apis it looks like it's a drop in code for open AI compatible so not much changes that's dope feel like I got my trippy kit for now but keep it coming keep it coming so I need to go to H mm models what was the last one we did we did Lin rag so I'm going to copy all this code over and this new one is called Stoke comma Hammer yo am I doing this right like that's my question is this what is this what machine learning is it's just coming up with a bunch of ideas and testing them anybody in the audience know like you know are you supposed to switch models more often am I doing this right yeah I guess that's the question I feel like I'm just going in a loop which is good because I'm seeing new data but again I'm trying to get like 50 shots up just so I can see where to go but I'm curious if there's another approach to this I'm actually just GNA ask AI I'm running out of I'm running out of chat so I'm not gonna I'm not going to ask her not right now but I'll put it on the Note sheet let's go to read me ask AI am I doing this right am I literally supposed to just just test all of these different indicators indicators and ideas with different models and features to predict different things I know price isn't the best to predict but test uh predicting Market r regimes and when to run each um strategy could be good ideas but just curious if this is the right route route to take When approaching ml for trading is it really just about testing all these different ideas features indicators Etc trying to find a model that works best what is best I still I'm not clear on that my best right now is just I'm trying to compare it against others you know a different other ones that I do but you know Rick RI what's up bro good to have you here dude Anga G says you're just fine-tuning the parameters and putting the model through various tests you have designed to get it to reflect reality as close as possible essentially so do you feel like this is a good path I mean I feel like I haven't found anything better than the volume one yet and I'm going to keep going obviously keep going but I just want to make sure like when I come back at it that um it's the right path to take but I I like this so far I also want to try some should I be trying should I be trying rnn's or other um models where can can I find all the suitable models to use on this time series data do you have a GitHub you says Nah I mean I got a GitHub but it's a private GitHub um Kieran says first time here and I don't understand a thing on the screen LOL that's fun man or lady yo B I'm going stick it to B that's fun I'm I I'm really proud to hear that honestly because I didn't understand any of this when I got started either and it just takes like persistence day after day um I believe code's a great equalizer if you don't know how to code then you got to learn bro you got to learn that's the first step cuz then you can build anything riss says I'm pursuing a data science degree at the moment and have beginner coding experience what concepts do you recommend I study needed to make a trading bot damn you're going to school for this you should be recommending stuff to me honestly no but in all seriousness I've been coding here for 3 and a half years so I don't know where you're at but I show it all on my YouTube channel I've been tra building trading Bots pretty much specifically for three and a half years on my YouTube so you can just start here and just kind of pick through however you want I know you're in school so you probably don't have a lot of cash but the boot camp I have a boot camp that walks you through all this stuff step by step you can read about it here and probably reverse engineer it if you don't want to join but $69 I know you're already spending a ton of money at school so no worries if you can't afford it um you can always just learn stuff at a little little slower speed on my YouTube because everything that I've learned I've done here Live on YouTube or somewhat live and um I just put the good stuff in the boot camp so my YouTube has everything and people always ask me to do like short concise videos but this helps motivate me to keep going when I'm I'm kind of like showing you over my shoulders it's like you guys are watching me do this entire thing every single day for the last multiple years and I can't stop I can't stop because we're a squad now you know yo kieran's from Singapore and you is from Singapore yo you y'all are on it your whole country is on it I never been to Malaysia but um I know Singapore like y'all are high tech over there angan G says I think you're on the right path Industries like Aerospace weapons and GPS have un or utilize hmm the most but finding the position of something in a Newtonian space is way easier than markets ah yeah makes sense but I still think it's worth looking at if you're stuck for examples ah it's good idea it's a good idea to look at that stuff over there in other spaces okay so I was just I wanted to ask AI this eventually but for now let's go ahead and compare the data or what were we doing oh I was doing a new one Stoke comma Stoke Kamala stoked stoked Kamala is about to be our president first woman president first second second Brown president dude yo yo yo yo yo yo that's sick if we got two brown presidents in my lifetime I don't care what side you're on dog I literally couldn't care less I vote for race that's kind of whack but that's kind of whack but if you know you know you know and now gender I'm voting for a woman every time I can't vote but that's my input I'm voting for a woman every time way more organized and then you put you put the race card out there you know I'm voting Brown so you do you dude but you know my take on that I ain't going to vote anti Brown that's for sure that's pizy all right States my opinion doesn't matter though I'm not able to vote unfortunately or fortunately probably fortunately okay let's go ahead and paste those in there main execution okay past her in okay let's see what she's got let's see what Cala got I know I say her name wrong but I should I should correct that I should correct that kamla kamla didn't work on this one let's go ahead and see if what she can do come on baby girl there she go down bang bang your trading bot can automatically place orders on centralized exchanges yep um as long as you have an account you should be able to Via bance yep hey Moon dad you've inspired me to learn dude that's so dope I'm so happy for you bro learning coding is like the best decision in your life in my opinion that is the best decision you could ever make so let's go ahead and copy the load and pre-process here and our our comma code comma code code see what our comma code does now oh but this might be a problem we'll see I've never used the Candlestick before what AI model you use uh so I'm using the hidden markof model right now just on day four though so I'm still barely new what type of models do you use for time series data we're trying to predict the state so not necessarily the price predict the uh regimes this is based off of some hints that um our guy Steve Jobs oops I mean Jim Simons okay dude same thing same thing I don't know I don't know dog Visionaries B Visionaries anyways we're using hidden Markov models because homeboy said he used them or his people used them but it's not only the only thing he used so any ideas you got please do drop them in the chesy predict States and analyze States here so this is a hidden Markoff model predict and analyze States we're trying to predict seven different states I think I'm going to start testing after this one with less States though because I think seven is aggressive use that it can help but in combination with something interesting so how would you combine it I guess and what like combin it with what like lstm [Music] or change oh did you take out she took it out why did you take it out of here look at that let's see what she said at least apologize for the error the issu is that likely caused by the nature of CDL indicator which returns discreet values usually minus 100 yo homeboy said this earlier dude listen to the chat the chat is where everything's at dog the chat is where everything's at somebody said this earlier and I just didn't understand it yet see that's the thing sometimes y'all say stuff I don't understand and I love that I love that because then I see it later and it's like okay now I understand what homeboy was trying to tell me earlier I love it it's a whole another element to this game and if you're here live you get the same you're here like you get the same things and you probably come to even better ideas cuz you see the chat too people leave with love around here this ain't scary scarcity set mindset Finance no way not around here I know there's a couple scarcity MOS up in the the chat right now but they ain't chatting they ain't chatting in the audience I say that there's some scarcely LED mos in the audience but not in the chat the chat chatly would love chat brings all ideas I apologize for the error the it's because the negative 100 or zero or 100 doesn't continuously like other indicators this can lead to singularities Let's modify approach we'll keep stoke SI I and comma but replace CDL Hammer with A continuous indicator a good choice could be the average directional adx all right I'm with it I'm with it if you with it I'm with if you with it she sneaked me though she sneaked me real quick on that that's good though I love it I love it I love it I love it let's go push me to the limits dog push me to the edge all of my friends are dead push me to the edge all right P her in okay let's run it there we go so Vladimir what up bro I'm glad you're here dude never seen you before he says use the hmm in combination with something find state for example and then combine with other Model H boost or some signals interesting interesting stuff you got there dude I don't quite understand it but I will so yeah yeah let me just ask then if you don't mind let me ask you a question about that but the question would be wait a minute the question is okay so he says to use the hmm in combination with something like hxg boost okay comfortable with that find state what does that mean find state for example you mean like the regime like the states okay that's I've been working on for the last 3 hours and I don't even know what state means come on mondev come on mondev so we find the state and then combine it with another model XG boost or some signals so do you mean to predict price after so predict price in the states while predicting price is pretty Noob and everybody is trying to predict price is everybody trying to predict price during specific states there we go see now we're a level deeper see everybody wants to predict price I don't know is this right am I going there right Vlad I don't know everybody wants to predict price right away but if everybody predicts price the price ain't going to be the price anymore get it does that make sense if everybody predicts price and the price ain't going to be the price noo but not everybody he says few working on algo trading already few years dope dope dude but not everyone predicts regime and then predicts price the this is one level deeper is it the answer maybe not but but now we thinking really I like it I like it though I love it I love this game so grateful to play so so grateful dog that we have a long Runway here a unlimited Runway to play this game come on B let's go come on you do you whatever you do do it hard dog rest at the end not in the middle rest at the end come on be let's get it I'm talking to me dog I apologize for for people that don't like hearing that but it's me I'm going every day this is the day to do I mean we could back test things too oh my God bro stop it this game is too fun there's too many levels there's too many ideas dude we can back test some of this data too like buy when in state one and then sell when in state three unlimited ideas dog unlimited Come on B let's go go um let's see here vad says yes you find state for example one bullish signal plus then XG boost can predict your direction for example with 80% confidence if bot match you put long if both match you put long you're fire bro you are fire you are fire I like your mind dog I'm so happy that you're here I'm so so happy that you're here I hope I hope you come back jump in the Discord you're you're my people bro it's good to good to meet you I like how you think that's a bu idea buy idea I'm giving you the crown I'm giving you the crown for the day I'm sorry just you just hopped in dropped the mic and yeah yeah duh like of course of course you do it that way and now it's like cuz predicting price everybody's predicting price just put some put some get some separation in between predicting price and everybody else you know okay if you want to predict price everybody's got access to XG boost everybody's got access to open high low close volume data landlord said send details bro oh man you ain't got to send nothing you ain't got to send nothing more Vlad you drop mic you can you don't have to even come back honestly N I hope you do you're a homie and I hope you do but you provided your value thank you dog so much love C all right I said we going to be here for 60 years I D there's unlimited things two tests all right sorry I'm just tripping now I'm just tripping relax relax I only got five more messages for today is I'm GLA I'm I'm over here glazing Vlad I'm over here glazing Vlad look at that I be learning hypers speed cuz of yall I learned a new I learned a whole new word today and then Vlad came up in here and dropped some dimes and then I use my new word because I am lowkey glazing right now oh man I love this game so grateful for this life that we live every single day we get to wake up and do anything that we want to do just close your eyes dude whatever you want just close your eyes 10 minutes a day dude I meditate for 1 hour a day I don't just say this stuff dog I really believe most people aren't thinking about what they want in life most people don't don't have the right direction they don't know which direction they want to go in they're just reacting but when you close your eyes and it's not just for sleeping when you close your eyes and you just sit there you can see where you want to go and you just go there it's going to be hard but it's supposed to be hard life supposed to be hard dog I love I'm so grateful for this let's get it let's keep going okay so here's the data analyze it so oh oh this one is poor I've never seen one so bad this is our worst one in the accuracy but let's see if it does anything else here's my here is the latest uh data output please compare to our other models so a Transformer model to predict price or Transformer model to yeah I guess that's that's the question right that's the question Transformer model to do what I've heard that a couple times now in the chat which is fire because when I see things a couple of times you know that people are people just leave with love around here dog this is the leave with love with Munda because I'm showing you literally everything I ever see every single day no matter what and y'all do the same thing back stop it I love it I love this world I love this world so so much you get what you put out dog you get what you put out all the time that's always you smile at somebody what do they do do they frown back no they don't frown back bro uh I found meditation very powerful and after was guided to Lord much closer dude don't look at accuracy I feel no Dimitri I don't want to skip over that I'm glad I'm glad you found meditation to be a powerful thing I it's I think it's in my mind it's one of the most powerful things I can do each day it's just magical it's magical it's magical how like or uh your room's all messy your brain your thoughts are all messy and then you just meditate for 10 20 minutes and now it's all organized everything's organized and now you know exactly where you want to go and how to do it and it's just amazing it's amazing so don't look at accuracy profit and draw down we don't we're not even back testing yet though how do we even start back testing this we're not even back testing yet maybe I should start back testing build this into our back testing model cuz Vlad said don't even Lookout accuracy so I need to move this note here it's all coming together dog it's all coming together you know it too you're here you're present build this into our back testing models I don't know whatever cuz FL said don't even look at accuracy profit draw down obviously so that's profit draw down it's calar ratio I believe what is calar ratio I believe it's profit draw down draw down but yeah why wouldn't we be testing on actual data funds maximum loss maximum loss from Peak to tro over a given period not quite but whatever um or Direction and then we can use some things from the past to enter well or Direction then layer in straps do enter well ah I'm not going to flush that out right now but I think that's the next Play From Here I'mma I'mma update this test three damn we haven't got test two yet get back testing. py into the equation and back test on OS o o data and yeah yeah super dope Vlad much love to you what's up shorty what you doing hey hey glad my dog's been quiet this whole time and she just started congratulating you she's said woof woof she said woof woof such a good girl much to you do I'm saying you puppy Vibes I'm send you puppy Vibes dog back testing is the most important part and most complex I have great tools for it nice what type of tools you use I just use back testing. py I just use back testing. py which is wild old okay um um see what else we got here this is this is comparing our other two get big body get get them get him that's a big body bag right there somebody just got taken out on a stretcher BTC let's go look at it I love seeing these body bags no offense when it's on the short side I love to see it yo Zeus what up good to have you back bro it's been a while it's been a whole day it's been actually hasn't even been 24 hours so I'm glad I see you every day now much love how's it going I'm doing great dog um we're just we're just hacking our way through these hidden Markoff models and got mad ideas got mad ideas I'm just watching somebody get body bagged on the chopper here they just got liquidated for 341 thow y'all probably can't even see the numbers on here huh yo is this screen stuck no it's not stuck is it can you see when I move it around oh my God how long has the screen been stuck for how many hours that's wild how many hours has this screen been suck for all right well I think that I think I got a bounc we're going to look I'll talk through this I'll talk through this I'll talk through this long time landlord says yo ain't nobody said nothing what's this a podcast now Jane says she's listening not watching okay so it was fine before okay maybe it just happened recently um I'm just going to end it with here and then I'm going to go grab some food and then I'll be back live later just because I got to reset this or something um Jane's got meetings she's getting ready certainly I'll compare this new model using Stoke RSI Cam and adx with the previous models we've discussed yo have a great day Jane much love to you always 777 landlord said about 20 minutes okay that's not bad Zeus I'm going to I'm going restart this here soon I just want to compare these models so new model is 41 previous 97 98 89 new model likelihood 3us 37k previous model minus 30 and the best one was 142 the new model has a lower log likelihood indicating poor fit okay so this model sucks this model sucks that's okay that's good that's a shot that's a shot we took and um at the end of the day and I know you can't see this so it's kind of awkward but at the end of the day this Stoke comma unfortunately doesn't work well you know I'll let Vlad go ahead and say his seven steps though I'd love to hear it I'd love to hear your seven steps because you've been dropping dimes today other than that I'm Aude all right so I've tried seven different hmm which are hidden markof models and the best one so far is model 2 so model two was there another good one this Stoke did not do well so we passed in some different indicators oh that's a good idea right there so model 2 is the best what are the main predictors here it was Stoke here let's see let's see no it was um it was a volume change so this is the best thing for these inputs thus far best I don't know how to define that but it is what it is let's go ahead and see okay this is 0 41 but that that model didn't do well what is our second best model yeah from what you've seen from what you've seen which are our top two two modelas so model two okay this model consistently outperformed others with high State prediction accuracy 89% but then the lowest log or the highest log and the lowest Bic the second one is the linear linear regression macd and true Range model this model showed good performance though not as strong as number one so let's go ahead and say so far the best one is this model 2 where can we put that let's just go ahead and put a date here dat it I don't know the thing is his time is irrelevant so let's go ahead and say 820b and Top Model is model to best obviously and second is Lyn rag Mac TR okay let's go ahead and just say hey okay well let's test this on how to sample data for both of them um update the above code in order to test model 2 verse the um and model 2 and the linreg Mac one as well because you said it's second best this is oos data where are we dude how do we get here what am I even saying at this point oos wow wow wow wow wow wow wow wow wow wow wow this is why I don't like coding B it's a whole different language that's it's they speak in a different language y'all speak in a different language too it's just it's just like any industry though there's jargon so just got to get with it or get lost essentially and they had me lost for a long time okay this is O data L ma above please what is the please please send new code back that will test these models ah man I feel like I'm missing something there she's going to need something else but I'm kind of BL certainly I'll update the code model l r so these aren't quite it but okay I see I could have just done this myself really but it's okay why would I type why would I waste my I only have so many keystrokes left in my life I'mma save them I'mma save them o testing this one here is oh that's testing 2 it's the same thing let's see how much it changes 78 79 okay what did I have up here nothing perfect so this isn't quite right but that's okay I just have to say something in front of it model two scaler two I said trained models trained models I was about to use some of my key extra keystrokes right there but I'm just going to copy paste I mean that's a smart thing to do o s okay slope what is going on here dude why do you have an air why would you ever have an air on me dog but you know I got Shooters be Claude Claude and then gbt that's it dude two messages left you are kidding me right you are joking right you are a silly silly B Billy right that is some wild wild stuff I apologize for the oversight the error is occurring because ta. linreg function is returning a series instead of a data frame it's all good shorty don't worry about it I'm gon let you slide this time obviously don't do it again but you decide to handle it at that time there we go shorty that's all you needed all you needed a little threat little threat that's all you need but Claude Claude is Claude is nice I like it so far I haven't switched back to GPT have you have I I mean other than like N I like it I have no no complaints I think it's is mobbing too but you know I can always jump back on the GPT when they need me um when I need it honestly when I'm out I'm actually about to jump on it so I better stop pumping up claw because I guess that's the complaint that's the complaint I got what do you mean you're cutting me off dog for an hour get out of here dog how you going to just cut me off maybe pay more that's fine you're just telling me that you're burning VC cash then you're exposing your hand dog that's all that's all it tells me if you have to cut me off at $20 and you won't let me you won't let me uh run it up then that tells me that these AI companies are they're upside down dog the metrics ain't right B something's going on there I'm sniffing it I'm sniffing it why you kind of why you charge and then not let me buy more I guess that's the question what's going on over there is it super expensive to run these models no yeah it could be it's cuz you're buying chips like a bajillion years out I don't know I'm not edicated enough to have this argument I'm sorry but this is the data so let's just go look at it let's use the use our last couple shots here CLA I love you so much you got to understand that it's all from love I love this you you got me through some hard hard times with rust the gbt couldn't do she couldn't do it and you could dude let's uh use one of our we only have one message left and this is going to be the end so the end of me and Claudia I'm going to make it a good one so I'm going to put a bunch of questions in it yeah let's run it up dude let's run it up one please follow the following follow the steps follow the following steps one help me Analyze This return print out and the chart to understand which did better in the out of sample data why did it do better I don't know this is a lot of pressure to have a limit on I can't just I can't just be thinking like that come on I'm closing this I'm gonna put a bunch of good questions in here though because I got this question ask AI get that up in there PA it paste that in what else what are we are using go yep yep there's another good question see just stack the brain dude I I mean I would smash the me of the past not just only because I know better python but just like we have ai now my guy they gave us way too much power but only coders can use it if this was widespread if this if they get if everybody could have access to the power that we have access to with python and AI or coding and AI in general they would shut that stuff down quick they just don't know they don't know dude we can launch a startup in an hour like a SAS we're literally sitting here in building alos all day long I know I'm just getting started know I don't know anything yet I'm only 4,300 hours into this game and I'm not saying anything till I'm at 10,000 I'm not saying anything I don't have to say nothing I'm just going to code you can watch the code and feel the Vibes send love 247 777 maybe not 247 but if you really want to watch me 24/7 you could think about that bro that's all love it's all love all right use multiprocessing to train models okay I don't really want to take a hard pivot but I might need this I might need this how would dang I was excited to I was excited to get get this going today I'm going to use gbd4 mini yeah EP it see see how she can get down you know let's see if she can move all right we're going to gbt for mini and I'm not even going to ask this question well man no I am going to ask this question let's get one more good question then so what else is there use multiprocessing we are using gazi and hmm good ask that predict price um or send me back some code that predicts the state change and then uses back testing. py like the below the attached the attached gript does in order to combine processes or whatever the most efficient way of doing this is the goal is to be able to to test the a strategy where it switches from where where when when price switches from not price when the state switches from one state to another buy or sell based on what you see in the image attached make things uh use the optimizer use Optimizer like the attached code shows so that you can test for all types of State changes okay and whether to buy sell or hold I don't know dog I'm asking the world of her right now why wouldn't I why wouldn't I now we got to figure out where that code's at because I was making big promises over there I was making very very big promises and I got to figure out where the codes at do so just get up in the little hyper liquid action here push a little code here and there can cancel because that's not going to open Visual Studio code okay it's already open the now I want to just see what type of back test any back test that looks good that has Optimizer in it that runs good thing we do this a lot right that's a good thing we do this a lot okay there we go okay let's run it let's see what you got dog oh she can't even handle this she couldn't handle that paste right there where you at oh you're down there okay I was going to say all right so I threw a back pest in there asked a bunch of questions from her Ox Auto says is this live it's not it's crazy pre-recorded though DUS Monta says how do I find the moonshot coin before it explodes and a bot to take trade quickly any video you would suggest none of them bro nah no moon shots around here dude actually the sniper I guess was kind of a moonshot but but um back in February I was building the salana sniper if that's what you're referring to 777 OE what up bro good to have you here I'm asking the biggest I'm asking the most out of out of whatever her name is right now and I just did because it's my last chat it's the last chat that's the finale so certainly let's go through your questions step by step one analyze analysis of the out of sample data results the model volume volume change BB width volatility performed significantly better on the out of sample data with a likelihood of 3859 compared to 27 798 for the Lin regge model a higher log likelihood indicates a better fit to the data looking at the chart the volume model shows more varied and frequent State changes which might be capturing more nuanced Market behaviors the Lin rge model seems to get stuck in one state state six for most of the later part of the data which suggests it's not adapting well to changing Market I conditions the volume model likely performed better because volume changes can be a strong indicator of Market sentiment and potential price movements the combination of volume volatility and price range BB width captures different aspects of the market Behavior it seems more responsive to changing markets as evidenced by the frequent State changes approach to ml for trading okay so this was the first part and that was interesting um you know I think the winner out of this whole test so far is modeled to that volume percent change is producing best results okay there we go I just got so much in that read me dude proach to ml you're on the right track H that feels good feels good you're on the right track the process of testing different indicators features and models is indeed a crucial part of developing machine learning strategies for trading there are some key points feature selection and Engineering is critical testing different combinations helps identify y what's most predictive for your specific problem when predicting price directly while predicting price directly is challenging predicting Market regimes or optimal strategy selection can be very valuable it's important to have a clear objective maximizing returns minimizing draw downs and to evaluate models based on these criteria not just statistical measures consider using other models like random Force gradient boosting or neural networks including rnn's and lstms for time series I mean yes you are on the right track cool just want to make sure I'm on the right track you know on the race I'm racing but I want to make sure I'm racing the right way look into libraries like s learn t flow and Pie torch for a wide range of ml models suitable for time series data hmm variations okay got some more Alpha here so hmm variations dang I can't see it okay here we go discreet hmm higher AR kle hmm factoral factoral hmm input output hmm W you might also consider consider other state space models like cman filters or particle filters oh goodness sorry oh the Inner Circle this your shot that was your chance that was your chance what's going on here so looks like a wild day everybody's getting liquidated both sides I want to feed in liquidation data dog feed in liquidation data and Clump it to be hourly Daily 4 Hour Etc dude that's going to be far you know that's going to be far that's why you here dog you're like yeah that's going to be far that's going to be far it's for sure going to be far let's start that now yo 777 blessings to you dude and drct or Mr tomur re what's up dog uh can you teach me what you do um yeah I can for sure hey man he said he said he is literally teaching in every video watch all the old videos yeah dude that's the way that's the way just watch the old videos pick out what you want um good to have you guys here much love Jason hey dude how are you I'm doing I'm doing well thank you thank you very much we uh working on these hidden markof models right now and um man I love this stuff I love this stuff dude find some good ideas all right um we ran out of credits here but this is going to be interesting here's sample code that combines hmm State predicting with back testing all right so let's see how this finale is go to hmm models dang I knew something was Eerie around here I haven't had a music on bro wow spooky hours like what is going on dog e this was this was I'm going to listen to a little I'm going switch it up a little bit oh it's not on my computer you nerdy nerd but yeah I'm going switch it up a little bit let's get a little something going you know let's get a little vibe going all right now I got it there we go now we can hear it okay um feed in liquidation data and Clump it into hourly 4 hour so let's go ahead and see if we can get this back test uh hmm bt. py this is a new Avenue here we just took a left here not full left but here's the sample code let's just see what it does does it work we off bat I got to put the data in let's not hold it not hold it to that high of a standard no way this ain't going to work but maybe hey this is looking nice where's the day to go dude back test cash hmm strategy M cash why does it say it twice though that's interesting I don't know what's going to happen here if this is going to work but let's just work our way through it final results is stats they okay so yeah yeah let's just walk through it because I can't even ask oh your data path here we go I was like what is going on dude load data but I want to see the yeah so I don't know if this is going to do it not quite what I need I didn't explain it well enough it's always me that's okay run back test run optimization Run final back test with best parameters okay let's just read through it see what we can get out of this load and pre-process data okay we got that we know that we load the data in so I'mma put the data here but the thing is dude is I want I want different data this is out of sample data here I don't know I got to figure out how to piece these together but we we got let's just move forward let's just move forward I'mma put this data in there for now copy path past in okay and then let's and get the first ER whatever that first error is let's get it no clothes what are you talking about no clothes I don't have ai on my side anymore yes I do above is the sample data of the CSV attached uh CSV for the code attached I got to say I like the desktop app big time I like the app I don't want a summary dog maybe she's going to do a summary and what did I put in there dude that was stupid of me I'm sorry I put the wrong thing in in there so I'm sorry AI GG first minute back I'm already dogging you no way it's me it's always me that's why I love this game so so much oh you lose money that's you oh you make money that's you it's crazy dude it's a good game the error you're encountering in case the code is trying to access a column name closes in your data frame but the data doesn't exist typically happens the CSV doesn't match what are you talking about we have it I thought is it because I have the file maybe because I have the file let's see if it drops it let's see one time no I think it's because the B CSV is a bit quirky and has an addition additional row we need to drop B like it's hot all right let's run it let's see what's Gucci wow you're really thinking about it that's that's good yeah but I got to have the code in there too you dumb dummy nah she's know she knows the code oh man it's not a good first show showing here but let's go ahead and uh make sure that we get everything that we have in here because remember we started with the uh clock and now we're over on GG send back full code send back fixed code with correct paths uh ideally ideally we train on the data that is above and then we save the model no no I have a model I already have it trained dude let's go look at it let's go get it trained model OKAY model two model two copy path copy path GG are you here are you here to stay we'll see I want to scratch this no and when we save the model ideally we train on the data which we already did and now we are just using using the O data below in order to test the model I guess I don't understand why we need to train the model again when I have one already trained I just now want to use it in a back test and be able to change the variables of the back test this is a the back test the back test should be testing what happens when the bot buys at state two and then cells at State four with changeable variables making those making those making those variables easy to change okay okay here's the other data um scaler to copy path what else do you need OS data copy pass sample of os data CSV okay let's show it all right let's see what you got gg40 mini whatever your name is let's see what you got we need you okay to utilize your pre-trained hmn model and scaler of Al s day for back testing here's how you would modify your script load the trained model and scaler from the provided paths process the oh as data to use the same features as used in the training phase process the OS data to use the same features as used in the training Fe phase perform the back test use the loaded model to predict the states and evaluate the strategy try to prepare features predict stat using pre-trained model find the strategy for back testing okay she just needs a little help with the back testing part It's Tricky GG don't notice uh above is a back tester I use and it it works please use it as a template in order to recode the above just recode it for me please all right hmm models BT okay this is where we're at attempting to back test the open out of sample data I missed the start what's your main goal of this rewind how did you train your model or did you download pre-trained models um I don't know honestly I don't know the answer to that these are pre- Trend pre- train model loaded the hmm oh no I trained my own model yeah now it's pre-trained I think that's the answer I'm sorry sorry I can't answer yeah but thanks for getting that in my mind okay let's pass it in okay so let's see here how does this look now dude okay we're passing in the model path okay this is the path model 2 is the best scaler 2 is the best okay the model equals jb. load I don't know what that does dude but it's okay I don't need to know everything this is the out of sample data this is out of sample data it reads it in okay okay process the data okay but what about ta dude what ta are we using that's why we look it's why we look at it dude it doesn't matter though cuz you know you know how that goes we got unlimited unlimited uses over here at open AI I guess we open AI guys again look at that we switch up quick I don't care dog just get the job done let's get it done dude okay print out the best optimization I I'm I'm not I'm not I'm not too confident in this but it's okay CU we'll just keep going so Define the strategy Define by state Define cell State I like this actually how did we get here I don't know last state none okay yes there we go buy and then sell and you can see it sets it up correctly but we don't have a stop loss why would we have a stop loss I think that I can just remove all of this be honest but let's just let it let's let it do it Thug fizzle let's add one more zero there because no why do I even have this dude I'm taking it out I'm just going to take it out sorry look at that your boy's learning maybe I don't want any of this here I'm just going to take all this out no stop loss wait wait I think it's just buying so all let's check it no way dude that's wild that's really wild 989 minus 46% cool that we got that the first time though line 89 I mean we kind of got it you know no attribute called take profit online 89 I get it out of here I don't want it then I don't want it I don't want it I don't want it get it out of here bye byebye good night all right dude we done did it we done did it look at this oh how can we do things so fast this AI is crazy man it's really crazy dude really crazy so the original or the best it converged on was one trade pretty much Buy and Hold whoop whoop whoop fun fun fun and it optimized from buy state to four sell state to one are you running the code locally yeah I am I am I am I'm here locally oh what you doing boo I didn't even know you were down there I'm local yeah I'm running on my computer so this is an interesting thing here that I mean what are we testing through let's look at that what is and component I don't know what that is optimizing optimizing for n components yeah but I want all that seven that's how many states I have perfect perfect I was just double checking just double checking you know I reckon we all double check a little bit more but insure buy and sell states are different all right well need a better idea than this clearly because I don't see anything wrong with this I said buy CH from buy from changing one state to another and with the hardcoded stuff it's down 45 but that was like a total random guess and then the optimizer just converges to buy and hold but you know you can always do that that's one way to do this but that's not really trading and there's a lot of risk risk risis aiz says hi bro I'm a CS student how can I start into trading and what's your advice my advice is to go watch my oldest YouTube video and wherever you're at in this process find the ones that Vibe with you and what you learn from and if you ever need extra help I got the boot camp but the boot camp is not required by any means it's just like it's like the fast track at Disneyland bud it's like the fast track but it's all here it's all on YouTube so I want to make sure it's like a equal plan playing field you know I mean it's unequal because people get the fast track but it's unequal enough and other than that I've learned everything on YouTube like while doing it on YouTube all right so this is just not that good of idea but it's a good Baseline peran okay let's copy path here this is a good Baseline solid Baseline for now chopping the market up into regimes and then testing only in one regime the code attached is um I don't know is is predicting the next date and then just buying or selling so that can be useful in order to code out something that is only buying or selling during specific periods blow blow up says have I made an outgo for stocks one of my first alos was for stocks but I just got distracted by crypto and it's just more interesting to me essentially so it's not really a distraction I'm just more interested in crypto and but yeah one of my first or was U interactive stockbot interactive brokers so I got it in my bag let's just say that I got it in the bag and um maybe 2025 or something I'll play around there I don't know though to be honest I've been saying that just more interested in crypto and you got to do what you're interested in in order to do things a long time I amo in the boot camp do you teach machine learning and how to get your data sets um no I teach you how to get data sets though so this is a boot camp you're asking about and I do give you a data source in the boot camp so if you need like open high L close volume data for minute five minute whatever you need whatever you need dude I show you exactly how to get it Robo what's up dog he said my dude I started taking Harvard cs50 python class and I've been dabbling with my own bots in Python and also Pine script you're a big Ino yo I'm so happy you're here dude thank you thank you for introducing yourself I keep showing up if you keep showing up dog this is uh this is a game we can play Forever get always always find new edges and as our boy Jim said the Market's constantly changing the Market's constantly changing so there will always be new edges that appear anomalies anomalies there we go let's look through that I think this is a really good place we got got to because now I got this tool that can number one split any data into multiple regimes based on the hmm I haven't done this yet not done but maybe I'll do that right now and then to get back testing done back testing equation this now I have a great starting place for hmm bt. py as it allows for buying selling in different regimes just update to only allow trading in certain regimes okay but let's do this real quick let's see what gbt can do update the above hmm so that it only looks for four different um States I want to see if the results change send back all code so let's do a uh seven states and then uh four states okay these are going to be four states I'm going say seven regimes forward James okay so below is a continuation oh snap I'm using GG free right now so we'll see how she does dang claw got me hooked B I might have to upgrade to uh chat gbt again are you running multiple python programs Robo ass um yeah like you can see all the ones like on the screen right now what up programs and they collect data they'll do things like that you just got gbt plus and it's worth it nice I used to have it but then I uh switched to Cloud cuz that's what all cool kids are doing these days okay let's go ahead and look through this code here n components okay this is good this is good let's see if it's really good though this is four it should be four components now so the the idea here was we have somebody that works uh at the Quan firm and they say that they use uh four regimes risk on risk off I should do two regimes too let's just past it in there let's start there yeah let's see what's going on down here though state names update the state names B what is going on how many has got how many uh components does it have here is that what it's called I don't even know seven components dog what so does she just not have memory like risk on risk off what eles they say do please pretty please send back full code I might have to upgrade right now levia what's up where do I start learning if I want to get into data science YouTube dude YouTube YouTube has it all any channels Moon Dev Moon Dev this a lot of code let's see what gbt 4 min's like here let's see if she can handle it can she do it oh she looking a little soft be honest she's looking a little soft I'm nervous for her that's why I'm looking through see where she done messed up we got risk on risk off high volatility low volatility all let's run it dog gbt 40 mini she can handle it she can do it too okay that's nice that's really nice here we go we got four different colors now do I get access to the uh thing I can't believe I'm over here I'm over here without GPT 40 she's the one who started it for us and I turn my back I turn my back that's so whack that's so whack but gb40 minis they're trying to take over the world give it to me for free then I I'll take it for free I'll take it for free yeah sure sure sure hey but now it's missing the uhoh okay but it worked which is good to see risk on risk off that's all I want to know yo use this use this as a templet you you lost me B it looks like you lost a little code use the below as a template but just with the four gem Simons RI features features or States ah this is tough dude a lot of new stuff coming at me send back auto code send back full code nothing less please pretty please got a ruit for the biscuit risk on baby Robo says 777 Eric good to see you d Robo we're about to Robo we about to start working through gbt virtually I'mma ask him stuff and he's going to put it into his gbt and then send it to me oh get him get him I would run up your bill though not your bill but I be run up your credits mad you see I already got logged out oh it's about that time though 11 minutes I can wait I can wait for it I might do a little meditation in between actually that sounds quite nice you've hit the free plan limit continue with other basic models or wow wow wow wow wow I didn't know they had a I did not know they had a upsell right there dude that got me that one got me they're goingon to they're choking me out here be they're choking me out I think I'm just going to meditate then because I own my Time come on B let's go let me upgrade Claud is there a is there a bigger plan you got a bigger plan Come on B I'll give you 10 more Claude pricing I'll give you a little bit more I'll give you 30 I'll give you 30 bro okay team per why you guys just always copy each other be creative come on throw this 69 on the price price build annually free learn about CLA team I learn about it don't know if gonna be about it but okay well that's too bad um I guess I could just get another membership but if I'm going to get another membership I'm I'm going back to open AI yeah that's what I'm feeling like that's how Doug said clly can have my babies shaking you down for that 20 Robo says I run mine up every day but I don't got experience like you Robo said well thank you man that's that's odd that's odd to think but I'm just a I'm a womb engineer I'm new to the game 77 7 KW what up bro much love to you much love much love okay so Claus got me Claus got me in a choke hole so I think I'm just going to meditate and come back later use hmm to predict four V regimes and then determine risk on or risk off okay let's just start there because I have these risk on risk off here and these are the four regimes right this is what we created let's go ahead and see what's Gucci with it so okay now it has four different regimes here and it says risk on riskof High Vol volatility low volatility but remember remember those don't really mean anything they're just four different things here people are getting licked right now um did we already test this let's see here execution complete I don't think we did the second part of this yet because we got cut off we got cut off train the hmm analyze States Save State changes let's go to our best one here so this is the B best hmm so far and the idea I wanted to tap into today was splitting up to four four different states maybe even less no I think four would be appropriate okay so here's the the best one here I just want to see the data that I get back I get this stuff back so hmm likelihood State prediction all that good stuff so that's what I need back here and if I go to the best here best is here and then I'm just going to say um State prediction okay I need all this code dog so I'm going to copy all this in to Claudia and say what's up hey there I have attached some [Music] um code of an hmm which is about 250 lines of code use that as a template to update this the other the other script it is missing a few prints and model saves send back full code right back to it CLA Cuts us off and it slows us down for a second I think I'm just gonna I think I'm just going to buy open AI as well I'm just careful with that subscription addiction careful I'm careful but we'll see today if it cuts us off I might have to just light it back up just run it back up yo what up Nicholas how are you 777 Aqua what's up bro he says yeah Aqua says yeah it's weird paying for something and still being cut off AI products are interesting bro drop your script PL says okay so I've dropped it it's dropped I dropped it on the floor I dropped it on every single YouTube video but let's keep it going B much love much love 777 all right so I understand you'd like me to update the second script from the past 2.txt using the first script from paste. txe interesting okay so they' have paste and then page two as a template adding some prints and model saves I'll do that and provide the full updated code added more print statements included additional analysis all right let's check it out copy there we go that's looking better it's looking a lot better and it's got the save yes there we go wow they really gave us too much power here with this AI dog too much power that's why they had to cut us off yeah duh I'm just going to I'm just going to get open AI back and I'll just I'll be running both cuz if it's going to come me off every day I wonder if it's based on times too like times of day but this is our data back or this is our chart back you can see it splits into four different states they call it um I often refer to as regimes and risk on risk off high volatility low volatility are the four regimes but these names don't matter remember that remember that remember that okay here we go this is the stuff we really wanted to see here it's going to take a second it's gonna take a second there it is Boom okay and it save the model as well so that is our analysis here and that's pretty good I mean the prediction at least let's go ahead and triple triple no no no let's put it I need to put it somewhere better I need to put it somewhere better uh read me [Music] um live ops. MD I love it dog all right so let's go ahead and say 821 okay this is the this is for risk on risk off copy relative path all right so that's that now I have that let's just open Live UPS to the right and then go ahead and go back and get the best model here so the best model model two best and then I want to run this against that and just see what the diffys are if there are diffys or not if there's not then we'll just I don't know move on keep keep it moving you know keep moving forward that's it next idea you see the list all right this is fitting the model so it takes a little bit of time time time time so this one predicts seven different states so the reason I went down to four was because I wanted to try four and I've heard Jim Simons say or I haven't heard him but I've heard it it's been said that he thinks there's nine states but I've also o heard somebody say that he thought there's four so I did seven cuz I couldn't think of nine right away remember these names don't matter all right so I'm just testing both just testing both copy path copy relative path here and say VV the heck this is supposed to be our best model best model I don't expect good but you know that's our best and let's see let's look at this one the Bic I think this is I mean that looks better not better but I can't remember that's the thing is I can't remember what the the good one was but you can see the bicc here on our best model it's supposed to be negative so it looks like this one's not very good either unfortunately our best model is still the volume change okay H what does that tell me I don't know but that's my prediction let's see if I've learned because if I've learned that's that's awesome we've been here and uh you've been here dog so let's go ahead and compare these two models here Claudia what's up above I have two two hmm models one with four states and the other with seven states um please walk me through the results and compare them which is better okay and that just had me think of something what are the inputs we're putting into this one that's the question because here you can see it volume change and that's taking up most of the I don't know I don't know what you call it dog the brain power of the hmm the volume change is that weighted the most it's the most important it's the most important so what am I using over here am I using the volume change I am perfect perfect perfect are these the same these should be the same let's go and look at it Lynn okay these should be the same they should be the same let's see it they're the same perfect perfect so it's a fair match it's a fair match four vers seven this one has seven this one has four regimes okay let's go ahead and see what shorty said certainly I'll walk you through the results of both models and compare them based on the metrics provided let's analyze each aspect State prediction accuracy for State model 91 that was the new one 7even state model 89 the four state model has a slightly higher accuracy in predicting the next state this suggests it might be more state in it State transitions the log likelihood of four state is - 29k 7 state is 14K the seven State model has a much higher log likelihood which typically indicates a better fit to the data a positive log likelihood as in the seven State model is generally better than a negative one [Music] ban information Bic four state 58 daow and 7 State negative 27 lower bic is better cross validation negative has a higher positive cross validation indicating better generalization on unseen data however both have high standard deviations suggesting some instability cross folds feature importance the four state volume change 97 but 94 on the seven State interesting a little bit more but did not make it better model fit comparison and conclusion the seven State model shows a significantly better fit to the data higher log likelihood lower Bic generalization the seven State model generalizes better to unseen data higher cross validation despite having more States the seven State model perform performs better according to Bic suggesting the additional complexity is justified by improved fit prediction stability the four state model is slightly more stable in predicting the next state but the difference is small overall the seven State model appears to be better choice it demonstrates Superior fit to the data better better generalization and more nuanced state representation the only metric where is slightly behind is the state prediction okay so the seven State still better seven State still better okay well I think we [Music] uh I'm going call this multi- regimes reim reimy multi- regimes okay risk on risk off I'm going say four here yo what happened to it four regimes okay I need to do this for all of them let's do it for two three two three turn this into just two regimes three regimes uh risk on risk off and frenzy send back all code send back all code frenzy that sounds fun a frenzy all right so three regimes py all I'm just going to start cooking up here I'm going say two regimes py you working for me today baby girl you working for me regimes. py 6 six regimes six there's got to be a better way to do this but oh my God I'm just going to use all my credits here I'm going to use all my credits let's run it three regimes okay okay risk on risk off frenzy risk on risk off and frenzy that's good 95% killer wow log likelihood though let's see it let's see our good log likelihood log likelihood this is like a test how do you how do you remember this how do you remember if it's supposed to be high or low just repetition be so let's go ahead and say uh this one is that and what is it though three regimes copy copy copy relative path you thought I was going to say path all right send me um now do one with only two regimes excuse me yo send me back all the code too don't play me don't play me not today don't play me dude I think this one's better no it's not cross validation not looking good that should be positive if you look here see our best one here um up here hey where are you dude it was positive all these others are negative this one's not good either but it's uh SS copy paste it run it do this the same do the same but with five regimes then do the same with six regimes then do the same with eight regimes you name the states send back full code one script at a time then please ask me if I would like to move to the next one yo you know what I was thinking just right now is Sonet 3.5 they got to change the name because 3.5 n they don't got to do anything honestly they can do whatever they want to do but 3.5 like degrades it lowkey because you know open AI over here 3.5 is not you know not popping like that we're on four now dos dude all right let's get this plot out of here actually want to look at it first risk on risk off risk on risk off all right so now you see it log likelihood okay okay so this one here then live Ops and paste it right there and this was two regimes right copy P we got AI working over there right she working right she working right I hope so yes there you go shorty that's how we're going to do it from here on out okay so that was I believe that was five let's see let's see let's see what their names are 1 2 3 4 five okay because the names don't matter remember that just remember that dog names don't matter we got to label them later I guess I guess that's how it goes I don't know to be honest I'm just out here in the ocean I swam out in the ocean there's no boat actually there is a boat Cloud but she left me yesterday that was so whack it was so so whack okay handle that Claudia Claudia Claudia all right so this one's looking bullish bearish sideways volatile and accumulation looks suitable to me 92% let's wait for this to finish up and then I'm gonna have oh yeah you're going to keep going yes copi you you're going to keep going yes let's go ahead and put six up in there let's go check out the uniqueness of the names they came up with strong ball weak bull sideways weak bear strong bear High volatility okay we got this this is a good flow right here nice dude don't drop the ball don't drop the ball in this nice float let's go let's keep going what was that five copy relative path yes and then that was the last one we wanted let's do a nine to why not why not oh because they cut you off yeah that's why not but that's it is what it is I'm just about to run it up I'm G get them both my goal is to get cut off from both open Ai and Claudia in one day that would be dope that means we're working that means we're going hard and I'm sorry I'm sorry I actually feel better bad I feel bad I feel bad like if I feel bad if somebody has any hard feelings about that's being able code so fast now God 1 two 3 4 five 6 1 2 3 4 five 6 7 8 all right let's do it let's run it don't drop the ball I said don't drop the ball keep it going did we get six on there I hope so six we didn't okay so let's run six first it's going to take a little bit longer I believe I don't know though how's this looking whoa whoa whoa whoa whoa you can print out 10 now get copy nine regimes. py okay and then please please print out 10 please print out 10 okay and we got I'm not going to show you the plot on this one sorry I'm just closing it so we're on six I believe so let me go ahead and grab the copy path here five and then let's just wait for it let's make 10 10 regimes. py now just wait on six so these look like they're getting longer every time I guess that makes sense I guess that makes sense 95 okay look at this cross validation score take shots dude so six six is dirty but I don't know I I don't trust me on that I have to ask a i because you know they know everything so this is eight and then we got to run eight let's run it and then we got 10 in the oven yep there it is I'm not going over 10 I'm not going over 10 that's just [Music] egregious there we go dang my computer working right now I can feel it I can smell it so this is eight we're running right now and then here we go what yo what's up Wishy anthropic um everything in Cosmos is up is Claude better than gbt 40 uh I don't know I like it right now but I don't I'm indifferent all right this is taking a long time huh this is 8 so we'll see if nine comes through I'm chilling Wishy how you doing bro how you doing there we go got it all right let's do nine now copy path relative path here hey hey hey copy relative path okay perfect perfect perfect all right now did we get eight in there yeah we did we did nine copy path okay now I'm just waiting for the data just waiting for this fitting of the machine learning model the hidden markof model um let's see you got any questions what do you always mean with 777 it's just a way to some love and some good vibes Aqua says 777 symbolizes spiritual Perfection and divine completion representing the ultimate alignment of mind body and spirit with Cosmic forces wow that's deep that is deep I love it so this is the thing but I'm just going to let it go through perfect perfect you can see this one's a little less here 88 uh how much resources does this kind of a I use oh that's a good question dude I don't know um I don't know I don't know about that everything in Cosmos is up Cosmos everything in Cosmos is up all right since I got a little time I'll go over there see what categories maybe is that how you do it oh clicked on an ad they got me already monetize me already um Adam yes so this is awkward I thought they had like a category or something maybe you just go to Adam or [Music] something similar coins Marcus how do you yeah how do you observe how do you observe everything's up in Cosmos that's my question so this is what number is this that was nine nice now we're just going to do 10 copy path copy relative path here and then I'm going to run this yo Derek what's up 77 bro 777 how are you how are you Daniel 777 to you as well much love peace all right so this one is training with 10 components 10 different components 10 states 10 regimes whatever you want to call them call them what you want to call them all right so here we go and um these are the 10 states now so what do you think about him what are your thoughts my thoughts are close it and let's get the data back and then let's get some a professional thoughts how about that that's a good idea right I think so we got all the data from our our file here compare all of these 10 models Maybe not maybe not I'm going to do it one by one yo Wishy yeah I'm working on a hidden markof model um Jim Simons tipped me off so I just wanted to dive into it SP some time it's been the last few days still Noob as usual just a womb engineer over here trying to find his way in this deep deep ocean without a boat but I got a boat until the boat rugs me and we'll see if Claudia rugs me today if she does then you know maybe it's time to go beg beg for forgiveness from open Ai and get back on chat gbt because I might need both how's Gemini in comparison to GPD 40 and Claude 3.5 are be curious cuz it you know I just be curious come on baby show yourself show yourself you're almost there so it has 10 states so it should be a little bit long longer Daniel says thanks I've been discovering and testing Alo trading thanks to your channel my goal for 2025 is to move from webdev to trading and Bot that's dope dude that's dope welcome to the world all right here we go glad you're here my brother let's go ahead and throw this 10 in there and compare all these just compare them all just compare them all except the first one because the first one we know is doooo berries above is [Music] are a bunch of different hmm test Iran please walk me [Music] through them what which did best why I mean that's that's you know it's good enough that's good enough it's good enough prompt you know we're deep deep deep in this game now and um you know it's been a journey this has been a journey I love I love these Journeys though I do I do love these Journeys [Music] bro uh oh are we back on mic talk is my mic messing up again everything clear for everybody [Music] dang I'm from the US West Coast where you where are you I think I think you've told me before Daniel the mic sounds good okay cool oh it's clear that's why I'm asking it's a um cool that's cool to hear because yesterday we took a pivot some uh there was an audio engineer here or somebody that sitting in a studio and he had some problems with the Bas so he helped me with like the I don't even know honestly it was like learning how to code again playing with the little meters and stuff anyways it's a ATR or it's audio tech audio tech sorry it's upside down audio tech Nika Cuban half Cuban half Spanish I'm from Spain dope dude great place it's a great place I would love to go back all right let's go ahead and see what they said so certainly I'll walk you through the results of the different hmm models and analyze their performance let's break this down by looking at various metrics State prediction accuracy so two regimes had 97 three regimes had 95 five regines 92 six regines okay this kind of makes sense I guess the models with fewer States tend to have higher prediction accuracy this is expected as fewer states are easier to predict but it doesn't necessarily mean they're better at capturing market dynamics log likelihood higher is better 10 regimes was the best nine regimes was second best Okay so this just go 10 n six all the way down straight up this is insightful because it seems like it's all in order the models with more states have higher log likelihood indicating better fit to the data BC basian information Criterion lower is better so 10 regimes is the lowest so it's just the more regimes the better the Bic penalizes model complexity yet we see the more complex models more States perform better cross validation score higher is better so the more States the better I mean that's what it's saying I ain't say it I mean we'll see we'll see we're not even done reading come on moon read the whole thing again models with more States for perform better in the cross validation F feature importance most models heavily rely on volume change except for the two regime model which gives more weight to the BB W width and volatility based on these metrics the models that perform best are 10 regime nine regime and eight regime and we had we weren't even going to test those dude give me something with 50 regimes B let's see see how that does the 10 regime model shows the best performance across most metrics highest log likelihood lowest Bic highest cross validation score why did why did these models perform best complexity matching market dynamics the cryptocurrency market especially Bitcoin is known for its complex and volatile Behavior models with more states can capture more Nuance market conditions balancing fit and generalization while the 10 regime model has the highest complexity it still outperforms simpler models in Cross validation suggesting it's not overfitting capturing diverse Market states with 10 regimes the model can distinguish between subtle market conditions easy neutral bullish and neutral vers neutral bearish which might be crucial for trading strategies feature utilization as the number of regimes increases the models start to make slightly more balanced use of BB width and volatility features potentially capturing more information from the data however it's worth noting the difference in performance between 89 and 10 seems relatively small the 8 regime model might be a good balance between performance and complexity get them out of here ldo whatever that is somebody just got Cho it off rest in peace don't trade by hand somebody else just got Chopper it off n they just got shot dang both ways what's going on over there that's always interesting when I see liquidations on both sides how did this guy get licked right there on BTC what is you doing dude right here what are you [Music] doing what are you doing how do you get licked right there how much Levy did you use maybe that's a mistake too I don't know ldo okay let's check it out let's see what ldo is sorry for the the switch up here but dang I don't get that maybe this data comes in wrong sometimes um I don't know but that would not be good somebody just got r wrecked for a mill on uni swap a hand Trader of course it's so sad to see may he rest in peace on Unis swap you were shorting Unis Swap and got licked let's see his decision okay so I mean shorting by hand is just a bad idea in general in my opinion all right so let's go ahead and say however is worth no noting the difference in performance between 89 and 10 the two regime model stands out for its unique feature importance distribution giving weight to BB width while it performs poorly on the other metrics 4,000 Chopper out of here every time you hear that copter that means somebody just lost all that cash B they're on tilt they'll probably lose some more boy don't go back there dog don't go back there as the number of regimes increases the state prediction AC accuracy decreases this trade-off might be important depending on your specific case in conclusion the 10 regime performs best overall but eight or nine may be the give me please please send me the full code for a 15 regime one yo civil what's up dog good to see you yo Ry bear why are so many people hyped over Tron and their memes are they that good I don't know dog I don't know I don't know but Jane hit me up about it you hit me up about it right there anybody else hearing things about tronny Tron I've had bad out a bad outlook on Tron since you know for a long time so I instantly faded but then I think it through and it's like okay if I instantly fade something I got to at least look at it I'm that biased against a whole chain I got to at least Peak but I feel like they always have this thing every year or so they've got some sort of bubble of something that goes on um I'd be curious if other people are hearing about Tron and all that stuff um I've never really been a fan but you know that makes me want to look into it I get bad vibes from Tron Derek says I agree always have probably always will unfortunately so I don't know enough just though I don't want to dog it or anything but 10 regimes now we got the 15 regime Mac Daddy let's see how this one does let's see it dance I know you're going to take a while it's okay I'm here I'm patient I'll be here for 60 years until I can't stand no more let's read out these names extreme bolt let's see one two strong bolt three 4 5 6 7 8 9 10 11 12 13 14 15 indeed so extreme bear strong bear moderate bear weak bear bearish consolidation neutral bearish slightly bearish neutral slightly bullish neutral bullish bullish consolidation weak bull moderately bull strong bull extreme bull I can see how this does well now that they've explained it to me thank you AI here we go it's plotting we're done with half of it you want to see it I sh you do let me just copy this over real quick seven state is still better not 8910 is best all right past it yo civil bro you are the you you are the man thank you dude got to hop in a meeting just wanted to say hope you all a blessed day dang so much love to you dude 777 from me and everybody in this chat that is so much love dog you didn't ever have to do that so kind of you all right so let's see this this here 15 I see how this could be good how many days are in the year 365 so what would this equate to 365 / 15 24 different 24 days 24 day regimes yeah 24 day regimes essentially let's let's not get ahead of ourselves looks cool colorful yeah yeah let's check it out though so okay okay okay so 10 regimes what's the highest right here log likelihood we need High 47,000 is the highest oh snap so what if Jim was capping not capping but you know like you got to keep his stuff a secret and I don't know where I heard him say nine regimes but somebody else said four regimes what if it's really like 18 regimes what if it's 20 regimes what if it's 30 regimes I don't know I'm going to keep testing this though yeah why wouldn't I why wouldn't I why wouldn't I dude so much love to you dog civil he said be for real you put you putting on Game for free 99 every day got to respect that you're a good dude man you're a good dude you got me over here shooting jumpers he just threw five at your boy for no reason I'm over here shooting jump air jumpers probably a couple air balls but you know appreciate you man that's love that's love I just know the pro the process of pulling out a credit card to to send a super like that is just like you're spending your time which is so kind so kind so kind good to see you my dude good to see you so this takes longer every time which is it makes sense right so it makes sense explain why explain why it takes longer to run the um man this stuff this part of run the code with more um more States I just want to get an ai's perspective on this because you know me I know nothing B I'm just here vibing all right so here we go this is good the increase in runtime for models with more States is due to several factors related to the computation computational complexity of hidden Markoff models hmm let me explain the main reasons parameter estimation in an hmm the end States we need to estimate n * n transition probabilities the transition Matrix and initial State probabilities probabilities and sets of emissions probabilities in this case means and covariances for each state as n increases the number of parameters to estimate grows rapidly requiring more computations expectation maximization em algorithm the bomb Welch algorithm a type of em is typically used to train hmm each iteration of this algorithm involves forward backward procedure complexity where T is the length of the sequence reestimation maybe I asked to I asked the wrong question reestimation of parameters also grows within more States mean more iterations are often needed for convergence State sequence prediction the vctor B algorithm has a Time complexity of N2 S as n increases the time to predict States grows quadratically Matrix operations operations like Matrix multiplication have a complexity of O to the N the 3 for n * n matrices larger State spaces okay so convergence time the likelihood landscape becomes more complex with more parameters this can lead to more iterations of the EM memory usage more states require more memory to store parameters and intermediate calculations if memory usage exams exceeds available Ram it can lead to dis swapping dramatically slowing down computations cross validation and model evaluation okay so it just takes longer sweet I'm going to get one send me one send me one one with uh I want two two per month 24 States two states per month cuz what if that's the case you know so this is the new one and it looks like this was 15 regime so more I don't know how to make my notes look good but 15 State cuz that's what she said no it's not what she said but I'm just going to have it there so Bic how's that looking so it's obviously not as good 86 83 so it just goes straight down that makes sense I don't know maybe I'm going overboard now but we'll see you know uh hmm it means hidden markof model ryar says I've always stayed away from Tron he did a fork of pump fun on Tron the volume looks interesting but Tron Tron yo what's the link it's called Sun pump or something that's just so sketch this is it okay um I'll look into this a little bit later though let me uh get this last one done here and then I think this is a pretty good tippity tip tip dude I think we learned something big here dude we learned something big today already dude all right so this is the 24 regimes let's see if it's truth I think it is let's run it dude all right so this is 24 regimes if you got questions let me know Sun pump are there any other resources for sun pump while I'm waiting on this Hmm this is so funny we're using machine learning models to predict Market regimes but also looking at some punp oh my God okay so I don't see much movement like the other one but what's Sun swap is that there that thing um one minute the thing is I don't know how to use pump fun I never I never used it and I can't build a bot for it I'm not that interested is this just because it just launched like if there's a ton of volume I don't know what is this where are you guys getting this volume from like uh not volume where you getting this uh this information from is like Twitter everybody's talking about on Twitter are there anybody yeah like I'm curious I'm curious anyways here's 24 this this is for Kobe so it should be the best yeah it is the best log likelihood so far I'm going to go ahead and say hey please tell me one or two things about my data please go ahead here and tell me this here's the 15 here's 15 regimes attached is 15 regimes and 24 regimes compare them to my 78 my 8 910 which is best okay so that's going to be done here in a few everybody's talking about this over here it seems super sketch already I mean I think pum fun seems sketch but so if you get to put it on Tron then it's like okay A little Sketcher but so sketch on sketch equals super sketch in my opinion a few seconds ago here we go I'm a pass for now I'm G pass on any uh any of this I'mma fade it I'mma fade it I think I'm G fade is this a bad idea I don't think so I think it's I think it's a good idea anything Tron I instantly fade I'm sorry it's sketch I don't know why but I got no proof I've got no proof but see they got API he's over here like they got API though yeah I'm good I'll keep an eye on it thank you for like like it for me but you know pump fun seems cool like I just never got into it I can't use right API whatever I build snipers I build snipers so I just kind of faded pump fun um probably for the better it seems like I'll keep an eye on it so Tron double fade I'mma stick get I'mma stay right here I'mma stay right here I'mma stay right here let's check out the see what people are saying though maybe maybe it will convince me while we're waiting ton dropped an R all right nothing yet nothing yet dang this is taking a long time but that's good let's see what my computer activity is looking like I haven't looked at that in a long time that's a luxury right there dude remember back in the day when I could barely keep the stream up we've come so far I could barely keep this stream going now we got every order coming in every order on binance flying through our screen all the time add a apple good job good job much love so much love so much love here we go this might be the best one so I mean what does that tell me I don't know 24 State let's just let's just go ahead and see what she says oh no no no no no did that oh my God copy pass no no no no paste it in paste it in 24 States copy P why do I just switch up the whole thing from here I don't get that okay okay now we're good we're good now that's the path this is the path every day dog every day every day dude every day just keep going just keep going Peace by piece makes a little more sense every day every day dude copy PA flag it that's a flag right that's what they call it I'm growing up in front of your eyes it's amazing thank you for all your help could not have done it without you you teach me things I don't even know I need to know attached is 15 regimes and 24 regimes compare them to the 89 10 and we just keep on going come on dog I'm still here where you at the screen's gone is what you're telling me I was just talking about this I was just giving you props Apple the screen's gone again so that means I'm gone again that's crazy dude essentially it takes the stream down all right this makes sense though I'm might have to cut some of these orders off I was just giving daps to Apple thank you for providing the results for the 15 regime and the 24 regime models let's compare these with your previous 89 and 10 regime models to determine which performs best we analyze each metric State prediction accuracy eight regimes not provided nine regimes 88 10 regimes 86 81 okay the accuracy decreases as the number of regimes increases which is expected the more state to make predictions more challenging log likelihood higher is better the log likelihood consistently increases 15,000 versus 68,000 log likelihood in consistently increases with more regimes indicating better fit to the data this keeps decreasing lower is better the Bic continues to decrease with more regimes suggesting that the increased complexity is Justified to improve to the improved fit cross validation score higher is better dude so this is just telling me that more regimes more better more regimes more better let's do 56 regimes no no no no come on now you're going crazy dog 24 is fine what you think I should do 56 throw a 56 and if I should do 56 it's going to take a minute I might have to shut some of this stuff down shut some of these order flows down should I do 56 that's how many years are months are in the year right weeks in a year 52 56 it's 52 isn't it it's 52 isn't it how many weeks in a year how many weeks in a year time is irrelevant to me I don't live by time I time I live I time my living 4 hours a day dog that's it I locked in I didn't know how to code I locked in for 4 hours a day and it was so hard it was so hard dog it was so hard boohoo me oh I have to tap my fingers for 4 hours a day on this little code and now we got AI this is so so hard 4 hours a day you kidding me dog I used to game for eight hours a day dude I used to game for eight hours a day four hours is easy feature importance interesting I like this on the 24 regimes it goes down a little bit it it passes it over to volatility analysis model fit the 24 regimes chose to fit best fit the data with the highest log likelihood and lowest bicc the 24 regime this is for Kobe this is for Kobe we got to use the 24 yeah I'm I'm sticking with it though maybe the maybe it's better maybe there's others that are better but I got to I got to kick it with the 24 I mean 50 is pretty pretty N I got to do 50 give me the code for a 50 state that's probably GNA be too many but but let's just get it anyways let's just get it anyways might take a while that's okay I'll just run it on a terminal dog that we can keep it coding duh come on moon you got to be thinking sharp all the time Come on B tap in based on the metrics the 24 model appears to be the best performing model overall it shows the best fit to the data the best generalization and the the most balanced features this is fun dog this is fun all right I'm going to do 150 and we're just going to let this run for a while over somewhere else I'm not going to worry about it but it's free it's free dog this is free a free it's a free beat dog if it's working 24 or 50 those are the two most important numbers to me so I would love that to be part of part of a system because then I get to smile every time I think about it cond activate T flow bro cond activate T flow bro python run it what cussing what is going on copy P python python python python python python python whoa dude what's going on here con activate T FL okay I'm there copy path LS CD oh maybe I got to do that no I don't have to CD in anything [Music] python what's going on dog what is this right here is that a thing it's always been there look at it I've never seen that yeah I guess so and does it run from here oh okay I'm not quite sure all right big guy love you bro I don't know if the 50 is going to work out for us but let's try it let's give let's give it a try give it a try let's give it a try 24 seems acceptable though that's what um every 10 days or something 4 days 14 days 17 n 365 ID 24 365 / by 24 so 15 days every 15 days main execution this is what we're changing dude this is what we're changing okay let's check it out then main execution no we want all of that out of here peace all right let's run it all right so here we go we got 50 for the big guy this is going to take a long time let's see if I can run it from my other thing here copy path okay um cond all right we're not doing 50 we're not doing 50 I'm deleting in it much love 24 is the best so if there's 24 regimes for Kobe that means that okay I got this now if there's 24 for Kobe I was pre-shock that was pre-shock for sure post Shack trained I want to get one of these back tests going BT okay the first paste let started new new little chesy the first paste of code above is a template template for you to follow for the hmm plus back test I want to test the trading idea of I mean what idea do we want to do is the question dude hmm models 24 regimes dog this is what my I want to make a script here to explore this data no dude I want to make a script to explore the above data with Matt plot live um I'd like to print out pie of it all right so I'm plotting it out now M plot Liv all right so so from all this I found the conclusion is 24 regimes volume change heavy there you go there's the alpha dude much love I don't know if it's actually Alpha but we'll see strong ball okay this is sketch AF dog I don't like it already I'm sorry so this is I'm glad we're looking at this because this is this shows me okay this is looking at the past right let's ask AI let me try to get this thought out here a sample of the state changes for the 24 regime is above My worry is that each row has a different regime but these are all decided after they happen so how do we know that the accuracy is 80% 80% or whatever is shown below I do that why am I like this it's not whatever below it's literally 83% dog dang that's funny I didn't even do it I didn't even get it I think it's up here though post Haw classification you're correct that these regimes or states are typically classified after the fact this is common in financial market analysis where patterns and Trends are often identified retrospectively accuracy measurement when someone claims an accuracy at 80% or any percentage for such a classification system they're referring to how well the system performs in predicting future Market Behavior based on these identified States here's how to think this might work historical analysis the system is first developed in tested on historical data where states are identified post HW forward testing the system is then applied to new unseen data either through out of sample testing or actual real time application to see how well it predicts the future Market Behavior accuracy calculation the accuracy percentage would typically be calculated based on how often the market behaves as expected given a particular State classification verification process to truly verify the claimed 80% you would need a clear definition of what constitutes correct prediction for each state a set of test data that wasn't used in developing the classification system that's what we need to do thank you I'm AI Alis Le okay I want to I want to go back here send me back the code that yo what up 777 Bailey uh that um okay so model regimes hmm models trained models 24 copy path where's the other 24 here somewhere copy P okay and then what was the best one before it was seven s s seven 6 10 3 where is seven we on a seven here all right we're going to do eight then we're going to do eight cuz eight was good too copy path oh sorry seven is two duh seven is two model two copy pass duh copy path okay so yeah I want to do that now so now let's go up here now back to or not now doing an oos test on model 2 7 States verse Model 24 States stats okay so that's going to be good dude it's going to be really good I like the idea how about you how about you dude do you like it I like it a lot o s o o s okay o s two No 7 verse 7 v24 py hi my name is Mev let's run it cousin all right there we go what's that looking like I don't know you don't you know I don't know I need shots I need to get shots up I need to get shots up till I can say anything I don't know anything in confidence I'm sorry um thank you but I also want to get my uh Bic 15 24 here's my 24 that I need because I didn't write it down and it's the best it's the best in the ass all right 24 past it in there yes let's do a little two-hand action here Bic cross Valley feature importance ball change BB with uhoh I'm lost to be there okay there we go now what this I'm going to actually run this again because I'm going get on over here and say hey I actually no this ain't it hmm did we get this script though this ain't it this ain't it oh no I lost it okay that's fine fine we're just going to run it from here then OSS o o OSS s s o s let's get it okay okay [Music] okay to throw extra couple KS on there because I was sketch boy sketch seven State versus 24 State please break down the uh OS test and the results which is better so I did a OS test here copy copy copy relative path okay live Ops here let's go ahead and check it out certainly let's break it down out of sample uh test for the seventh stay in the 24 Hidden stay applied to BTC price log likelihood 7 State 3890 24933 the log likelihood is a measure of how well the model fits the data a higher log likelihood indicates a better fit in this case 24 model has significantly higher suggesting it fits the out of sample data better than the seven State model visual interpretation the seven State model shows distinct broad State changes over time the 24 State model exhibits more frequent and granular State transitions the 24 State model appears to capture more Nuance market conditions potentially identifying s SU subtler shifts in the Bitcoin price Dynamics State utilization these seven State model seems to use all of its states regularly the 24 State model also appears to utilize most if not all the states indicating that the additional complexity is being put to use in describing the data responsiveness to price movements the 24 State model shows more frequent State changes which could indicate higher responsiveness to short-term price movements and market conditions the seven State model captures broader longer term Trends in the market complexity versus interpretability interpretability interpreting being able to interpret it the seven State model is simpler and potentially easier to interpret with each state possibly representing a more distinct Market condition the 24 State model offers more granularity but may be more challenging to interpret the differences between some states could be subtle which is better based on the information the 24 State model appears to perform better in terms of fitting out of sample data as evidenced by a higher log likelihood score it captures more nuanced Market Behavior and potentially offers a more detailed representation however better depends on your specific goals if you're aiming for the most accurate representation of Market States and don't mind dealing with increased complexity the 24 State model is superior if you prefer a simpler model that's easier to interpret and still captures major market trends the state seven State model might be more appropriate for trading purposes the 24 State model might provide more frequent signals which could be beneficial for shortterm trading strategies the seven State model might be more suitable to for longer term Trend following the 24 State models better fit suggests it might be more predictive but this should be verified with additional metrics like prediction accuracy on future data what other metrics could we print out to see which is better Jane what up 777 good to see you good to see you Jane says check your DMs I check them I'll check them I'll check them I appreciate you so so [Music] much lots of licks Happening Here excellent question to get more comprehensive comparison between the 7th State and 24th State models we can add several additional metrics here are some important metrics we could calculate and print out the basian the Bic the basian information Criterion this metric balances model fit with model complexity lower bic is better aaii information Criterion on the AIC similar to Bic but penalizes complexity less lower AIC is better State prediction accuracy how well the model predicts the next State Higher accuracy is better transition Matrix analysis entropy of the transition Matrix to to measure State stability feature importance to see which features are most influential in each model perplexity a measure of how well the model predicts a sample send back full C I'm getting the most out of you you're typing every letter I'm sorry I'm sorry I'm not sorry I'm going to run you up every single day until you cut me off and then I'm going to go get somebody else I'm sorry every single day dude CLA you know we're hitting we're hitting the limits every single day I'm not leaving till we hit the limits that's my new thing I'm not leaving till I hit the limits on both no let's start with Cloud let's start: no no but I get I be get the limits too early no way keep going keep going every single second for iby contents 10 messages look I'm already there that's funny bunny some funny bunny stuff think I'm just going to sign right back up I think I'm just going to sign right back up so this is going to give us a little bit more detail supposedly let's check it out though if that's truth or not oh man here we go we got some things though there we go that's what I like to see dude all code back if they're like limitting on messages number of messages then that's different have you experimented with any high frequency models for voting agents I'm new your streams introduced from Twitter don't know which AI to use if any to execute logic or if it's all Quant what are HF models not sure what that is like high frequency models as voting agents nah man you're ahead of me you're ahead of me bro let's go ahead and pop this in here thanks for stopping by though appreciate you much love let's go ahead and say run it up there we go okay this is more info here's more info oh and there's okay okay cool cool Co above is the prince out please walk me through it and how if and if it changes your outlook or things I should know about it I don't know dude I don't know dude I just be asking anything this is why I run out of chat so fast they want me to be more thoughtful and I'm not going to be I'm sorry certainly let's walk through the results for both models and analyze what they tell us about their performance on the out of sample data seven State Model results log likelihood of 3890 this is lower than the 24 State indicating a less precise fit viic minus 7,000 2 and the AIC isus 7, 601 this is on the 7th State lower values indicate better models these are higher less negative than the 24 State suggesting the 24 State model provides a better balance of fit let's go see it I mean okay so here I can see a little bigger here just so we can kind of understand 24 State you can see the accuracy is much lower though transition entropy what the heck does that mean dude what does that mean I don't know it's okay though cuz it's just a word feure importance perplexity I'm perplexed you would say perplexity on the print out average State duration 3.59 6.3 average State [Music] duration okay we looked at them well let's compare them now so State prediction is 84% on the 7 verse 72 transition entropy 3.96 verse 2.41 and then that stuff doesn't matter I mean it does matter but State prediction is 84 on the 24 indicating a better short term is that cap no no this is higher than the 24 all right transition entropy 2.41 lower than 24 State model suggesting stable State transitions feature importance the volume change of 94% perplexity is 0.0000 very low perplexity and indicates high confidence in predictions nice average State duration is 6.3 States persist no longer for longer periods compared to the 24 State model State usage 0 two 4 and six are most frequently used while state three is never used all right this is actually useful this is awful right here I mean maybe we'll see let's go check it out let's go check it out see what can do with this thing dude all right so State usage for the 24 I want to see that for seven too why because then I can I can do some thg dizzles over here and try some things all right for seven cool cool cool um now let's says model fit the 24 State model provides a better fit to the data higher log okay the seventh state has a a higher short-term prediction accuracy stability the seventh state shows more stable State transitions feature utilization the 24 State model makes more balanced use of all features unused States both models have unused States 1 and 7 and 2 and 24 suggesting some redundancy in the space State distribution the 24th state has a more even State Chopper him out of here get him out of here 400k off damn dude that hurts that hurts my soul that hurts my soul that hurts my soul rest in peace young cat changes in Outlook the the 24 State model shows stronger statistical performance better fit lower Bic AIC than initially thought despite its complexity the seven State model's higher prediction accuracy and stability could make it more practical for certain applications especially short-term predictions things to consider the trade-off between model complexity and interpretability is more pronounced than initially apparent the unused States in both models suggests that further optimization of number of states might be beneficial the high Reliance on volume change in both models indicates this feature is crucial for bitcoin price Dynamics the very low perplexity in both models suggests high confidence but this should be verified with additional out of sample testing to ensure it's not overfitting in conclusion while the 24 State model shows better statistical fit the seven State models higher prediction accuracy and stability make it a stronger Contender especially for practical applications where interpretability and short-term predictions are important the choice between them would depend on specific use case and whether the additional complexity of the set 24 State model provides actionable insights for your Bitcoin analysis or trading strategy in conclusion while the 24 State model shows better statistical fit the seven State model higher prediction accuracy and stability make it a stronger Contender especially for practical applications where inter predicability and short-term predictions are important the choice between them would depend on the specific use case and whether the additional complexity of the 24 State model provides actionable insights for your Bitcoin analysis or trading strategy all right so that's interesting so rip 24 much love my brother we miss you but I think we got to go with seven we got to hit the Jaylen Brown on them I think we got to hit the Jaylen Brown on them all right so that's all this is saying I mean not entirely but okay the top two the top two are 24 and S but it may be beneficial beneficial to use just the um just the seven so we can read States easier all right so that's that people are getting licked over here what's going on let's check it out oh yeah we had that huge Unis swap liquidation of 1 million that was earlier today um looks like BTC is up to this liquidation here I'm sure we have it what day was this August 20th at 14400 probably this right here yeah that's what I would guess it's my best guess but not much liquidity huh not much liquidation going on for this move right here to be real with you cuz what is this two hours of time so I'm looking for like a group of two hours of liquidation 7 to 11 that's 4 hours dude this is it right here 9:30 to 11 so that only that move right here what does that tell you dog I don't know I don't know but that move right there 3% move was this much in liquidations it's interesting what does it tell you though I don't know let's go look at the seven regimes it's funny that we are back on model number two after we tested 82 models not 82 but I want to go ahead and I want to go get this back test going again I know I just asked AI for some sort of back test but I don't think I went through with it so multi- regimes here and then BT I want to go see what this back test looking like dude copy why did I copy it I don't know save close save close close close close close close close all right we closed them all and now I'm going to look at the back test because I want to see what the back test was doing was this the seven 34 weeks train scaler model 2 okay okay okay okay so I want to say I don't want to say anything yet but I want to run this model two again and then try to identify some of the yeah that's what we F to do right now I want to identify some of the um the things identify the seven regimes and label them um okay that's still a lot though but maybe I can like zoom in or something this gets a little bit of everything right still be oh I'll do that I'll do that I got these seven regimes right here dagn where I put those suckers right here we got them right here so this is um 24 seven do you see seven in here I don't see it no [Music] volume yeah it's just that one okay so it's not what I wanted though sheesh I think it's this right here let's double check it you can see here that says bullish trending bearish trending so sideways consolidation I see it bearish trending I see it sideways consolidation I see it bearish trending I see it sideways consolidation I see it bearish trending I see it sideways consolidation I see it downward consolidation I see it what's the objective here I wish they stayed longer but I guess that's what the uh like three states does let's go look at the three states then dude how are we back to three states open reveal and finder okay why is this like this though that's weird because it says green here H let's check out three states here with um L frenzy risk on frenzy risk on frenzy risk on frenzy risk on [Music] [Music] H this only says risk on I'm not sure about this how do it only show Once this entire data from 2018 and just keep switching this is a good exercise though strong neutral bullish it's just looping through these dude strong bear strong bear neutral bullish but these again these aren't these aren't real the names aren't real so it keeps tricking me in my mind they're not real all right I'm going to say build a back test here what are the [Music] uh the state uses zero two four and six okay so I have these states use the below as a template back test in order to send me back a fullback test code that test all possible buy and sell combos like buy at zero sell at next zero then test then test sell at two then four then six all different iterations of this back test and I think this is different what I did before It's tricky to for my brand to get around it but we'll see um I optimized it before but I was trying let me just get this out we'll see um all different iterations on this back test so after so test buying on two selling on zero then two then four Etc below is the model we will use get them all right so this we like deep up in here we're so deep copy path okay and then model two here scaler two okay by the way what is the model 2. job lib verse the scaler to okay let's paste in the BT code the BT code okay uh BT the back has code somewhere this is so intimidating bro I tell you that much so I'm sorry I'm really sorry send back pull code okay there we go and um let's watch her work trying to cut me off crazy crazy stuff hey man can I make a suggestion for this CLA rate limit issue please please please do please please please Josiah says oh I already WR that one go for it subp Jane hope you ate on time today uh not at all brother thank you for all you do my pleasure bro my pleasure agnan G says I've been trailing this VC Fork code Fork cursor AI that implements its own C- pilot you get 500 CLA uses per month and cursor expands the context to your entire code base it's like 20 a month and if you have an anthropic API key different than the base Cloud plan you can use Claud at Cost instead I've been trailing this VC code Fork what do you mean no way so you're just watching okay you saw somebody get funded and you've watching their code that's dope smart smart man that imple that implements its own co-pilot you get 500 claw uses per month and cursor expands the context to your entire code base it's like a $20 a month or if you have an anthropic API different than base you can um use claw instead dang I'm G write that down appreciate you sharing that how did you get started I just started on YouTube and just started learning it's all there um Josiah Valentine says don't want to distract you from the session here's a link that runs through how to leverage multiple models to complete task you never have to shut up bro stop it how do you get started how's it going I'm doing good man thank you was thanks do Uber scalping in one minute candle um something about scalping the one minute I'm not sure what do you think about SC in one minute it's harder down there Zeus what's up bro 777 much love hope you're doing great he said thank you 777 ditto Jane yo ditto even ditto ditto Zeus you you on Discord Discord question Mar we have a Discord Zeus please one sec finally I've joined the Discord zo says by the way I'm working on a project for a SAS course at my University I am extremely impressed with the Claud Dev plugin for vs code it's like Claw on steroids that's the second time I've heard that that's the second time I've heard it Claude Dev e eest appreciate your shares Fellers okay so thank you thank you both for that plug in by sa rzan there a few YouTube videos on it if you search something like claw Dev that might help at first to get a hang of it I gotta go peace see you Jane 777 to be honest they got you for this 20 this it's super valuable nice I'll just check out the YouTube videos or something why YouTube video okay there we go it's called cursor AI essentially essentially actually for real so test States here 024 four and six by stay okay let's see how this do do dude seven regimes multi- regimes o o testing let's call these uh testing where that back test at you come up in here the testing dude testing okay let's add it um what do we do here we say this is uh BT seven states I don't know dog I don't know I know nothing except I'mma keep going I'm keep going and going and then I'm going keep going I'm going get rxed I'mma keep going I'm going be stuck I'mma keep going every day dude every day all right look at that uh-oh what did I just show you what did I just show you dude win rate 60% 1.1 sharp return 23 okay I mean exposure times 11% that's what I just showed you is that worth it to you I like it I would give up a few percentage points for that okay let's go ahead and copy this error here man I love this game dog love this game so much it's so hard I love it so much it's so hard boohoo let's get it every day dog come on come on that's just a warning cousin that's great to see let's see what it said though those warnings are related to the Boke Library used for plotting back testing package they don't affect the functionality of your back test but let's address them to clean up the output clean up the output for me please clean up the output for me please cuz you know we're just getting started Oh I thought we've been going a while now no way no way dude why why would I slow down I'm sorry I'm sorry every time you see me here I'm working on something new my bad my bad I'mma keep going I'll show you everything though I'll show you everything bro every day what I find you find that's the game replace the original plotting car with this okay let's go to the plot plot plot plot BT plot oh is that the one nah yeah it is all right all right let's go ahead and run in now Warning free beautiful beautiful beautiful okay dude what else can I do there's a lot I can do I like this a lot all right so I'm just going to save this then dog stop playing stop playing dog that's out of sample data too I like this dude 23% return is not better than Buy and Hold but you're only holding 12% of the time dog 8 months 9 months 8 months stop it stop it stop playing with your boy I don't know how many hours this has been dog but it's been a lot of hours been here a long time feel like this is just a start though [Music] Max trade time two days expectancy how L is your boy [Music] Nate 2.3 okay this doesn't mean anything though all it means is we're just getting started seven states let's do 24 States cousin I got mad I got mad things I need to do now okay live Ops copy path copy path copy path Cy relative path read me live Ops where you at Dog there you are okay so I'm just going to keep vibing out here until I'm done I don't know when I'm done I'm never going to be done dog I'm never going to be done there is no done in this game that's why I love it hey mm this is a great looking back test with 11% exposure and 1.1 sharp it's not the highest but you know it's a good start now I want to test with uh eth hourly data let's grab that eth data right now let's just grab it right now eth data how many weeks we want is the question do I already got it no I'm going to get it I'm GNA get it from here how many weeks we want is the question let's do 100 weeks let's do 104 weeks let's do 104 weeks that's two years dog maybe I should go more though 10 weeks what do I got in here already I got 34 weeks I got a th weeks of BTC data okay I should try it on that for sure copy path this we probably just break it but it's okay there's a th weeks of data and I'mma put this janky little comma there because no wait wait wait wait needs to be in the data Boulder one no janky comma no janky comma no comma no comma I'm just going to get all the data I can yeah I'm just going to do like 500 weeks of e data I can't imagine it's going to disrupt things but what's going to happen is it's going to not be as good I mean but it wasn't as good as buying hole anyways that's fine how many weeks let's do 200 weeks so that's like four years just I got to be conscious of the time it takes to download all the data but now it's downloading [Music] um one of the many things I show you in the boot camp the boot camp shows you step by step how to automate your trading 100% money back guarantee always always always if if you got questions about any of this stuff it's probably not because you can't do it it's just you haven't started yet so this will help you out a lot I wish I had this when I got started step by step how to automate your trading how to back test how to build your own Edge and um yeah I'm flying bro I'm flying I'm out of here I'm gonna keep going all right so see what happens can I just pop this in let's see if I can just pop it in yeah this is out of sample data copy relative path here and back test seven states this is for that okay and now we'll try it with this thousand I don't expect it to be better than Buy and Hold because this this is all Bitcoin data and there's uh that's a pretty good strategy in retrospect to just buy and hold from the start of Bitcoin but that's not what we're looking at all right so copy path okay I'm just going to plop in that data now let's just say uh let's mark this one out and then maybe my AI can just do it for me yes thanks shorty look at that I don't even have to code I just move things around playing Legos all right let's see if it can do it with a thousand then it looks like it's working so I bet we have our information here in a little bit of time there it is dude okay how's that look to you I don't know it's 10 years of data 2015 wowers bro cool stuff cool stuff all right so the profit factor is 1.6 the expectancy is 1.83 sqn is 1.47 280 trades sharp ratio kind of low the return 56% [Music] annualized but like I said it's the Buy and Hold is like obviously goat here exposure time 68% % so over 3,000 days it's not the goat but you can't go back 3,000 days in Bitcoin that's the that's the tricky part about back testing against a lot of Bitcoin data so I like to look at other things like the expectancy that's nice The Profit factor is nice win rate is solid draw down is not solid but it's you know that return [Music] um oh cool I got the uh eth data I'm going to do the same thing for uh soul but all of these are just up over the years which which is makes it tricky tricky tricky I'm going to do it for sa as well so let's go ahead here and try it out eth 1 hour okay this is the 1 hour data for 200 weeks it says and I haven't tried a different um choer him out of here get get them ghost it's time to go home buddy quit the game if you're going to play the game the same way you always play quit the game you're going to get liquidated over and over and over and over again till you can't even play the game no more so quit the game now put the game down put the game down dude come on stop playing stop playing stop playing like that so you can see it's a 380 per return I like this one though exposure is only 33% so to get that close of return versus Buy and Hold that's like a lot of capital free that's a lot less risk Sharp's not great 48 228 this is cool though this is a cool new um new back test here so yeah that's pretty neat it's pretty neat it's got me a lot it got me thinking a lot like um where' it go from here with this cuz this back test is really just like a filter it's a regime Creator it's a regime Creator so I'm going to grab this one as well I'll just leave this here for now 23% Buy and Hold 35 this is a I mean as a filter and now layer strategies in could be interesting we'll see we shall see copy this down here I already did that dum dummy come on baby boy come on copy relative path okay so that's that and this is not as good return as Buy and Hold if you would have bought on 1023 of this date to this date four years but a lot of free capital and this is just a filter so I'm actually super interested in this now man I need to write some stuff down so eth was like four 380 return verse 500 Buy and Hold with 30 3% whole time but that's pretty biased like I just wrote that down there's another one in there um what I look forward to what I look forward to here is that these are simply regimes or filters so being able to find a profitable in the past not guaranteed the Prof in the future Strat off regime changes is dope but also the market just goes straight up so I'd be interested to see how this looks on other data yo rigar what's up dude Samir says is the strategy finding the hidden states on the entire data set once or is hmn predicting future States based on a window of past data [Music] um I uh it's it's past data past data it's trained on past data and predicts with that model Jamie says hey man I see your streams appear on my YouTube what is it that you're doing hey dude welcome much love to you I am uh I'm coding I'm coding a uh machine learning model it's a hidden markof model and we've been playing around with it for a while and we got the back testing it today it's a pretty big uh pretty big moment actually pretty big moment but I still think it's just a start I still think it's just the start CU there's so many ways we can slice and dice this now so live Ops I wanted to write that okay I want to do the same thing I want to do the same concept here for the 24 um State model I want to do the same type of back test where we test all variables again uh but for this these many more states listed above which appear most in the 24 model okay and then we got to get that code because I don't know if she remembers I don't even remember so you know how's she F to remember if I can't even remember cuz she's smarter than you Moon she can definitely remember that she can defin I didn't know it was the last one it's just been a while we've been over here executing what a great teammate she is huh it's crazy it's crazy stuff here dude trained model let's go to 24 copy path she probably doesn't even need this to be honest she's so smart send back full code for for what up for for e e e all right so this is the 24 back test just curious you know 24 rest in peace King all right R it all right so looks uh oh this is on the that data too wait what dat is on worst trade good expectancy good profit Factor beats the Buy and Hold uhoh what does that tell me dog that tells me good dang dude another one H so 240 days all right so that's that and what data was that on dude 31 we 34 weeks okay let's try on some other data then this is this one's almost impossible to beat but might as well try it the uh data from BTC 10 years 10 year BTC data let's see it let's see it dude let's see you dance all right it's going to take a little bit of time there it is expan See's nuts 8,000% verse um 23,000 so no didn't beat it 53 88 win rate is good though 54% expectancy is good draw down how's that draw down looking 74% so not so hot but I think that's pretty standard or this 70% exposure time and volatile asset I don't know what to conclude I have no conclusion except low s s Lo s all right let's go ahead paste it here B it up here okay so it's cool it's really cool I like this idea I like this idea of cutting things up into regimes and having AI mess with it so here I'm going to say this is the out of sample out of sample data here this is the first one okay and now this one here is the BTC 1 hour all right sick so let's try that eth here let's say copy path um I don't know why I'm up in there but copy path and say data path let's go ahead and delete this one be out dog data path okay that's cool with me hey guess what else we got some more data too so oh you tickling me oh you're hungry again you hungry I'm hungry too we both hungry shorty we both hungry I'mma feed you right now you grow and grow and grow um you stay on track just for a few more seconds here and then I'm G get you something I want to get this Soul one hour in here all right I got it in there um this here is the detailed results for that Strat 24 States this one was for the eth okay it looks solid 2.54 sharp a little better 7 beats the Buy and Hold it's cool to see exposure time is 2/3 of the time okay so what was that that was the e 1 hour I believe let's try something new dude let's try something new let's say data path equals data path data path data path data path data path data path data path copy path copy path paste it in okay now we're going to say up here okay copy run it all right that one's going to take a minute but this is it this is that oh snap what is this though okay so this was eth copy relative path here ET States I just want to make sure that it's here so I know what I'm doing data okay what did the what did that chart looked kind of so this is the draw Down August 20 22 interesting profit factor is 10 only six trades interesting and it beat b and hold but not very many trades so you know so quick look six trades this one had 150 trades a little more statistically significant better than Buy and Hold but then the BTC one from 10 years is not what about BTC for like I don't have it I could get it pretty easily though I could do other time frames but you know I think I don't want to over test this just it is what it is it's good information to have what you going to do with that information though that's a cute dude that's the queue so this is a good place this is a good place to now look at this and see how do we layer it how do we take it from here I'm always constantly just following this if you're ever wondering what I'm doing it's this every day every day for the next 60 Years bro RBI system for AO trading it's just my simplified version of this the process of automating your trading comes down to researching trading strategies watching videos listening to podcasts reading books I didn't love reading growing up but now you're just searching searching for ideas one idea can change the world then back test those ideas strategies to see if they actually work in the past so the idea today was the hmm a markof model to predict the different states in the market and then have a back test that trades based off of the changes of those States see if it works in past data and then if that's Prof profitable in the past it might work in the future so I always start with tiny size but most of the work is in the r and the B and everybody wants to build a bot including myself just take that lesson from my experience seriously and don't just go build a bot don't buy a bot on the internet none of that that dude please if everybody's running the same algo it's going to converge to zero profits over time pretty much guaranteed it's just math and I think we went over this a few times in this this uh this stretch uh this journey this Marathon this journey that's what it's been it's been a journey that's why machine Lear seems to be hard for trading because if everybody can predict the price then the price is not going to be the same anymore because everybody's going to trade off that price it's not like if everybody knows the prediction of the weather it's going to change the weather no it's different it's a fun Journey I like this journey a lot