when you first dive into the field of interpretable machine learning you'll notice some very similar terms flying around we're going to discuss one definition and hopefully clarify some things that is the difference between an interpretable model and an explainable model although I should warn you there is no consensus part of the problem is that IML is a new field definitions are still being proposed and debated you can't even seem to decide on the name for the field is it interpretable machine learning IML or explainable AI X AI so we'll focus on one potential definition we will learn how to classify a model as either interpretable or explainable using this definition we'll also discuss the related concept of interpretability to end we'll discuss the problems with this definition and why it's actually probably not necessary to classify models using it if you miss some details or want the references for the information in this video make sure to check out the article Linked In the description also if you're interested in IML make sure to wait till the end of the video where I explain how you can get access to a python shap course so we say something is interpretable if it is capable of being understood with that in mind we say a model is interpretable if it is capable people have been understood by humans on its own we can look at the model parameters or model summary and understand exactly how a prediction was made another term you may have seen for these types of models is an intrinsically or an inherently interpretable model a decision tree is a good example to understand how a prediction was made we simply have to transverse down the nodes of the tree another example is linear regression the model you see here gives the predicted loan size for a customer based on their age and income we can immediately see why someone aged 29 with three thousand dollars monthly income is predicted to have a maximum loan of 33 000 and 100 it is also easy to see the general Trends captured by the model that is the loan size will increase by a hundred dollars for every year a person ages and ten dollars for every additional dollar of income so we can look directly at these models to understand how they make predictions this is because they are simple the decision tree only has a few nodes and the regression only has three parameters as our models get more complicated we can no longer understand them in this way a model is a function the model features are the input and predictions are the output an explainable model is a function that is too complicated for a human to understand another name for this is a black box model we need an additional method or technique to be able to peer into the black box and understand how it works an example is a random Forest which is made of many decision trees to understand how the random Forest works we need to simultaneously understand each of the individual decision trees even with a small number of trees this is not possible for a human things get even more complicated when we start to look at algorithms like neural networks to put it in perspective alexnet convolutional neural network used for image recognition has over 62 million parameters in comparison our regression had three parameters it's simply not possible to understand how a model like Alex net works by looking at the parameter weights alone so we need some additional techniques to understand how these models work this includes methods created for specific types of models such as deep lift which was created to explain neural networks they also include model agnostic approaches which can be used to explain any model these include methods like lime shop pdps and Ice plots we should always use these methods with a level of caution they all come with their own assumptions and limitations and really they only provide approximations for how the model makes predictions up to this point we have discussed models as either being interpretable or explainable yet it may not always make sense to apply this binary flag this is because interpretability is on a spectrum or in other words interpretability is the degree to which a model can be understood a convolutional neural network is less interpretable than a random Forest Which is less interpretable than a decision tree most models can generally be classified as either interpretable or explainable however there is a gray area where people would disagree on the classification this gray area is where we find our first issue with the definition we may agree that a random forest with two trees is interpretable but a random forest with a hundred trees is not at what point does the model go from being interpretable to explainable even a decision tree with many nodes or a regression with many parameters can become too complicated for a human to understand without additional techniques the issue is that we are trying to classify a model based on human comprehension and there's no formal way to measure this your ability to understand a model will depend on your technical skills and professional experience even amongst professionals there will be disagreement another issue is what we Define as additional techniques even with the simplest models we often see kulp from additional methods it is common to use a correlation Matrix when explaining the weights of linear regression does this mean regression is now an explainable model this leads to the question do we even need this definition the goal of IML is to understand and explain our models we do not need to classify them as interpretable or explainable to do this the methods we choose will ultimately depend on the type of model and the specific questions we seek to answer but like I said in the beginning there's no consensus so what do you think do you agree with this definition or do you think there's a better way to classify our models if you're interested in IML and want to get started then check out my course on Sharp it's the most powerful python package for understanding and debugging our models and from the theory to application my course will teach you everything you need to get started and also for a limited time if you sign up to the newsletter in the description you will get free access