MLBT — ML Backtest Framework
Python research framework for ML-driven trading strategies.
The problem it solved
Quant strategies that look great in a notebook leak future information and overfit — they fail when shipped. MLBT enforces time-aware cross-validation and a reproducible experiment loop so only strategies that survive proper backtest reach production.
Similar problems it can solve
Any supervised-ML pipeline on time-series data (demand forecasting, churn prediction, anomaly detection), reproducible-research infrastructure for quant teams, and ML-evaluation harnesses for finance.
A Python research framework that takes the ML-quant trading workflow seriously: a typed ETL pipeline, target-generation helpers, time-aware cross-validation, an experiment tracker, model training and persistence, and a backtest harness — the research bench that complements the production BLZN engine.
- ETL → target generation → quant cross-validation → backtest pipeline
- Experiment tracker + model persistence for reproducible research
- Python research bench complementing the .NET production trading engine
Overview
MLBT is the research-side counterpart to the BLZN production trading engine — a Python framework for designing, validating and backtesting ML-driven trading strategies before they graduate into the live system.
Engineering depth
It structures the quant ML loop into orthogonal modules: an ETL stage to assemble features, a target-generation stage for supervised labels, quant cross-validation that respects time and avoids leakage, models and save_model for training and persistence, an experiment tracker to record runs, and a backtest harness that scores strategies on out-of-sample data.
Technology
Python, scikit-learn-style modelling, time-series cross-validation, and a research-grade experiment tracker.