How bayeswitnesses/m2cgen works
Project type and purpose
• m2cgen is an open‐source code‐generation library/CLI tool written in Python.
• It takes a trained statistical or ML model (e.g., scikit‐learn, XGBoost, LightGBM, StatsModels, Lightning) and transpiles it into native code in one of many target languages (C, Java, Go, JavaScript, etc.).
High‐level file‐structure analysis
• Top‐level:
• generated_code_examples/ – sample outputs in all supported languages (not part of core runtime).
• m2cgen/ – the core Python package.
• CLI entrypoint files: __main__.py, cli.py.
• setup files & CI configs (Dockerfile, Makefile, GitHub workflows, tests/).
• m2cgen/ subpackages:
• ast.py – definition of in‐memory model representation (AST).
• assemblers/ – logic to convert AST into an intermediate “assembly” for different model families (linear, tree, ensemble, SVM).
• interpreters/ – language‐specific code generators that traverse assembly/AST and output source code.
• exporters.py – public API functions (export_to_python, export_to_java, …) wiring together model → AST → assembler → interpreter → code.
• cli.py & __main__.py – parse command‐line arguments & invoke exporters.
Important architectural patterns & principles
• Pipeline: model → AST → assembler → interpreter → code.
• Strategy/Factory: each target language has its own interpreter/ code_generator class that implements a common interface.
• Visitor: interpreters walk over AST/assembly to emit code.
• Separation of concerns:
• AST encapsulates model math in a neutral form.
• Assemblers transform AST into language‐agnostic operations.
• Interpreters handle language syntax, formatting, imports.
Main components & relationships
• User Interfaces
• Python API (m2cgen.exporters.export_to_<lang>(model))
• CLI (m2cgen/__main__.py, m2cgen/cli.py)
• Model Loader (external): accepts a fitted sklearn/lightgbm/etc. model.
• AST Generator (m2cgen.ast)
• Assembler (m2cgen.assemblers.*)
• Interpreter / Code Generator (m2cgen.interpreters.<lang>)
• Output Writer: returns source code string or writes to file.
• Tests: ensure correct assembly & interpretation for each language & model type.
User → CLI/API → AST Generator → Assembler → Interpreter → Generated Code → User