AFML: Advances in Financial Machine Learning - Refactored Library

This repository contains a refactored Python library based on the concepts and code from the book "Advances in Financial Machine Learning" by Marcos López de Prado. The library provides a modular and extensible framework for backtesting, portfolio optimization, and financial data analysis.

All example scripts have been tested and are confirmed to run correctly with the refactored library.

Installation

To install the package in development mode, run:

pip install -e .

This will install the package in editable mode, allowing you to make changes to the source code and have them immediately reflected without reinstalling.

Key Features

• Data Structures: Advanced financial data structures like tick, volume, and dollar bars.
• Labeling Techniques: Sophisticated methods for labeling financial data, such as the Triple Barrier Method.
• Feature Analysis: Tools for feature importance analysis, including Mean Decrease Impurity (MDI) and Mean Decrease Accuracy (MDA).
• Cross-Validation: Specialized cross-validation techniques for financial time series, like Purged K-Fold CV.
• Vectorized Backtesting: Fast and efficient backtesting of trading strategies.
• Portfolio Optimization: A suite of tools for constructing optimal portfolios.
• Performance Analytics: In-depth analysis of portfolio performance and risk.
• Model Validation: Tools for checking the robustness and validity of financial models.

Example Scripts

This repository includes several example scripts that demonstrate how to use the refactored library, corresponding to chapters in the book.

• scripts/chap2_bars.py: Demonstrates the creation of standard and alternative time series bars (e.g., tick, volume, dollar bars).
• scripts/chap3_labeling.py: Implements the Triple Barrier Method for labeling financial data for supervised learning.
• scripts/chap4_sample_weights.py: Shows how to compute sample weights to account for overlapping observations in financial time series.
• scripts/chap5_fractional_differentiation.py: Illustrates the concept of fractional differentiation to achieve stationarity while preserving memory.
• scripts/chap8_feature_importance.py: Provides examples of various feature importance techniques like MDI, MDA, and Single Feature Importance.
• scripts/chap7_cross_validation.py: Demonstrates the use of Purged K-Fold Cross-Validation to prevent data leakage in backtesting.
• scripts/chap8_feature_importance_analysis.py: Expands on feature importance by analyzing clustering and pairwise importance of features.
• scripts/chap14_backtest_statistics.py: Shows how to compute various backtest performance statistics, such as Sharpe ratio, drawdown, and precision.
• scripts/chap10_bet_sizing.py: Implements different bet sizing techniques based on model predictions and confidence.
• scripts/chap11_dangers_of_backtesting.py: Demonstrates how to use the DangerDetector and RobustnessChecker to identify common backtesting pitfalls.
• scripts/chap12_efficient_backtesting.py: Showcases efficient backtesting techniques, including vectorized and streaming backtesters.
• scripts/chap19_microstructural_features.py: Provides an example of how to process high-frequency data and analyze market microstructure features.
• scripts/chap16_machine_learning_asset_allocation.py: Demonstrates portfolio optimization using various techniques and performance analysis.
• scripts/refactored_library_usage.py: A comprehensive script that ties together multiple library components.

To run an example, execute the following command:

python3 scripts/chap2_bars.py

Project Structure

AFML_ver6/
├── scripts/
│   ├── chap2_bars.py
│   ├── chap3_labeling.py
│   ├── chap4_sample_weights.py
│   ├── chap5_fractional_differentiation.py
│   ├── chap7_cross_validation.py
│   ├── chap8_feature_importance_analysis.py
│   ├── chap8_feature_importance.py
│   ├── chap10_bet_sizing.py
│   ├── chap11_dangers_of_backtesting.py
│   ├── chap12_efficient_backtesting.py
│   ├── chap14_backtest_statistics.py
│   ├── chap16_machine_learning_asset_allocation.py
│   ├── chap19_microstructural_features.py
│   └── refactored_library_usage.py
├── refactored_library/
│   ├── core/
│   ├── data/
│   ├── features/
│   ├── machine_learning/
│   ├── microstructure/
│   ├── portfolio/
│   ├── utils/
│   └── validation/

Core Modules

The library is organized into several core modules, each providing a specific set of functionalities:

• core: Contains the main backtesting engines.

  • backtester: Implements VectorizedBacktester and other backtesting strategies.
  • streaming: Provides StreamingBacktester for event-driven simulations.

• data: Handles financial data generation, simulation, and structuring.

  • bars: Functions for creating various bar types.
  • labeling: Implements the Triple Barrier Method.
  • weights: Computes sample weights for financial data.
  • simulation: Generates synthetic financial data for testing.

• features: Contains feature engineering tools.

  • fractional_differentiation: Implements fractional differentiation for time series stationarity.

• machine_learning: Implements ML algorithms for finance.

  • feature_importance: Provides MDI, MDA, and other feature importance methods.
  • ensembling: Tools for creating ensemble models.

• portfolio: Includes tools for portfolio construction, optimization, and analysis.

  • optimizer: Implements various portfolio optimization techniques.
  • bet_sizing: Contains functions for determining position sizes.
  • analysis: Provides in-depth performance and risk analytics.

• validation: Provides tools to test the robustness and reliability of models.

  • cross_validation: Implements Purged K-Fold CV.
  • statistics: Computes various backtest performance metrics.
  • dangers: Includes functions to detect common pitfalls in backtesting.
  • efficiency: Helps in analyzing the efficiency of the backtesting process.

• microstructure: Includes tools for analyzing market microstructure features.

  • processor: Processes raw tick data into various bar formats.
  • analysis: Provides tools to analyze features from high-frequency data.

• utils: Contains various helper functions and utility classes.

  • performance: Monitors the performance of trading strategies.
  • visualization: A set of plotting functions to visualize results.