Skip to content

Classic SciML

The phaethon.ml module provides deep integration with Scikit-Learn, bridging classical machine learning with strict dimensional algebra.

While Neural PDEs handle infinite-dimensional spaces, Classic SciML equips traditional, heavily optimized tabular algorithms (like Random Forests, XGBoost, and Support Vector Machines) with physics-aware meta-estimators, automated Buckingham Pi feature engineering, and dimensionally safe evaluation metrics.

Dependency Note

The Classic SciML module strictly requires the Scikit-Learn backend. Install via: pip install 'phaethon[sklearn]'

Classic SciML Architecture

  • Meta-Estimators & Workflows

    The DimensionalEstimator and AxiomValidator. Wrappers that safely strip and resurrect physical dimensions, alongside physics-aware train-test splitting.

  • Physics-Aware Transformers

    Secure scaling pipelines (DimensionalTransformer) and automated SVD null-space extraction (BuckinghamTransformer) to synthesize pure dimensionless features on-the-fly.

  • Physics-Aware Metrics

    Dimensionally strict evaluation engines for MAE, MSE, and R-Squared. Accurately validate models and automatically synthesize complex error dimensions (e.g., converting Distance errors into Area).