Freak AI Introduction and Configuration Notes

General Overview

• Freak AI: Software for automating tasks in training predictive machine learning models for market forecasts.
• Main Features:
  • Self-productive retraining to adapt to market conditions.
  • Ability to create a large number of features from market data.

Market Data Overview

• Essential market data includes:
  • Open, High, Low, Close prices.
  • Date and Volume.
• Feature Generation: Can create numerous features using data transformations (e.g., percentage change, moving averages).

Key Features of Freak AI

1. Self-Adaptive Retraining:
   • Automatically retrains models during live deployments to adapt to market changes.
2. Rapid Feature Engineering:
   • Allows creation of extensive feature sets based on user strategies.
3. High Performance:
   • Utilizes high-end GPUs for model training and rapid inferencing.
4. Realistic Backtesting:
   • Emulates training on historical data.
5. Extensibility:
   • Supports various ML libraries (e.g., TensorFlow, PyTorch, XGBoost).
6. Outlier Detection:
   • Removes outliers from datasets to maintain data integrity.
7. Crash Resilience:
   • Stores trained models to disk for quick reloads after crashes.
8. Data Normalization:
   • Normalizes incoming data using statistical methods.
9. Automatic Data Download:
   • Updates historical data automatically in live deployments.
10. Cleaning Incoming Data:
    • Handles NaN values before training and inferencing.
11. Dimensionality Reduction:
    • Reduces dataset size via Principal Component Analysis (PCA).
12. Deploying Botnets:
    • Utilizes one bot to trade while others infer and handle trades.

Getting Started with Freak AI

• Initial Testing: Run in dry mode using command line: freek trade --config <config_file> --strategy <strategy_file> --model <model_type>
• Examples Provided: Includes classifiers, regressors, and convolutional neural networks.

Data Processing Pipeline

1. Strategy Creation: Pass market data, including candlestick and additional features.
2. Feature Engineering: Clean and prepare features, including outlier detection and normalization.
3. Adaptive Learning:
   • Continuous training on a separate thread.
4. Model Inferencing: Use the trained model for predictions based on new data.
5. Entry/Exit Logic: Define trading parameters such as stop loss and entry points based on predictions.

Important Concepts

• Features: Parameters based on historical data for model training.
• Labels: Target values associated with features for training.
• Training: Process of teaching the model to match features to labels.
• Inferencing: Feeding new data into a trained model to get predictions.

Installation and Setup Considerations

• Ensure system compatibility (e.g., CatBoost not available on ARM devices).
• Use Docker for deploying Freak AI if needed.

Pitfalls to Avoid

• Avoid dynamic volume pair lists; use a static pair list for performance reasons.
• Focus on pairs with higher trading volumes for better market moves.

Conclusion

• The Freak AI system provides a robust framework for integrating machine learning with automated trading. The next steps involve configuring the system for individual trading strategies.