How solalbaudoincs/ifcob works
Project type and purpose
• This is a research‐oriented, modular analytics framework (not a web application).
• Purpose: process high‐frequency crypto order‐book data, extract predictive features, train multiple models, and evaluate custom trading strategies via event‐driven backtesting.
File structure
• Top‐level orchestration scripts: main.py, backtesting_example.py, scripts/, found_documentation.md
• Modules:
• preprocessing/: raw CSV → cleaned, deduplicated, wide‐format parquet
• feature_extraction/: compute order‐book features (imbalance, slope, volatility, etc.)
• prediction_model/: notebook‐driven model experiments + model_manager.py
• predictors/: serialized model artifacts (.joblib + perf.json)
• backtesting/: event‐driven engine, Strategy base class, portfolio & order processing
• perf_analysis/: post‐backtest performance analysis
• strategies/: strategy implementations (trend_following, linear_regression, rf, etc.)
• Support: requirements.txt, LICENSE, README files per module
README for additional insights
• Describes a pipeline: data preprocessing → feature extraction → model training & selection → backtesting → performance metrics.
• Highlights modular design, custom strategy support, and realistic transaction cost modeling.
1. Raw Data Ingestion
2. Preprocessing Module
3. Feature Extraction Module
4. Model Training & Management
5. Model Repository (predictors/)
6. Backtesting Engine
7. Strategies Library
8. Performance Analysis
9. Orchestration Scripts / CLI
• Raw Data → Preprocessing → Preprocessed Data
• Preprocessed Data → Feature Extraction → Feature Store (e.g., parquet, JSON)
• Feature Store → Model Training → Trained Models → Model Repository
• Model Repository + Feature Store → Backtesting Engine ← Strategies Library
• Backtesting Engine → Backtest Results → Performance Analysis → Reports/Plots
• Orchestration Scripts invoke each step in sequence or individually
c. Architectural patterns & principles:
• Pipeline / ETL pattern for data flow
• Plugin architecture: custom strategy classes conform to Strategy interface
• Separation of concerns: data prep, feature gen, modeling, backtesting, and analysis are distinct modules
• Artifact management: serialized models + perf metadata in predictors/
d. Key technologies & frameworks:
• Python modules & scripts, Jupyter notebooks for exploration
• pandas, numpy for data manipulation
• scikit‐learn / XGBoost / AdaBoost for modeling
• joblib for model serialization
• Custom backtesting engine in backtesting/
• JSON for feature metadata and perf stats