BayesWitnesses/m2cgen

Supports scikit-learn, XGBoost, LightGBM, and StatsModels through a unified Python API that analyzes model internals and generates functionally equivalent code without runtime dependencies. Uses an AST-based code generation approach that preserves numerical precision and classification semantics (decision functions vs. probabilities) across target languages. Includes CLI tooling for batch transpilation of pickled models and handles edge cases like ensemble recursion limits and multi-class decision strategies.

2,964 stars and 15,885 monthly downloads. Used by 1 other package. No commits in the last 6 months. Available on PyPI.