Skip to content

ML Feature Engineering: Stability, Parity & Tiny Model

Canonical example for correct usage of QuantWave's ML feature toolkit (part of the quantwave-4ps work).

What this notebook demonstrates

  • Building rich feature matrices using HurstFeatureExtractor, CyberCycleFeatureExtractor, Trendflex, InstantaneousTrendline, etc.
  • Proving zero-lookahead and batch vs streaming parity
  • Regime-conditional stability analysis on synthetic data with known shifts
  • End-to-end tiny model training with per-regime metrics

Run the notebook

pip install "quantwave[all]" marimo polars numpy
marimo edit docs/examples/notebooks/ml_feature_stability.py

(Or use maturin develop -p quantwave-python --release if working from source.)

Sources & Context

This is the living reference for the feature engineering surface and validation methodology.

View source

Raw notebook on GitHub

Note for the docs site: Full interactivity requires a local installation because the feature extractors are implemented in Rust. This page provides context and the exact commands to run the real notebook.