pyGAM
pyGAM is an open-source Python library for building Generalized Additive Models.
pyGAM is an open-source Python library for building Generalized Additive Models (GAMs) with a strong emphasis on modularity, performance, and interpretability. Its API is designed to feel familiar to users of scikit-learn and SciPy, making GAMs accessible without sacrificing statistical rigor.
The library provides a flexible framework for regression, classification, and distributional modeling using smooth, interpretable functions.
Purpose
Generalized Additive Models offer a powerful middle ground between linear models and fully black-box machine learning approaches. pyGAM was created to make GAMs easy to use in practice while retaining their core strengths: interpretability, flexibility, and solid statistical foundations.
It is suitable for exploratory data analysis, applied modeling, and research-focused workflows.
Core Features
Generalized Additive Models for regression and classification
Familiar, scikit-learn–like API
Modular design for terms, penalties, and distributions
Automatic smoothing and model tuning
High-performance numerical optimization
Strong focus on model interpretability
Generalized Additive Models
Additive Modeling
pyGAM models relationships as a sum of smooth functions, allowing non-linear effects while keeping each feature’s influence interpretable.
Terms and Interactions
The library supports:
Smooth terms for continuous variables
Factor terms for categorical variables
Interaction terms between features
Custom model specifications
This enables flexible model construction without opaque representations.
Supported Modeling Tasks
Regression
pyGAM supports regression with multiple distributions, including Gaussian and Poisson, making it suitable for continuous, count, and rate data.
Classification
Binary and multi-class classification are supported through appropriate link functions and distributions.
Distributional Modeling
Advanced use cases include:
Poisson regression and histogram smoothing
Expectile regression
Custom likelihoods and link functions
Optimization and Constraints
Penalties and Constraints
pyGAM includes built-in support for smoothness penalties and constraints, allowing users to:
Control model complexity
Enforce monotonicity or shape constraints
Improve generalization and stability
Automatic Tuning
Models can automatically tune smoothing parameters, reducing manual trial-and-error and improving usability.
Performance and Dependencies
pyGAM is optimized for performance using NumPy and SciPy and supports optional acceleration through sparse linear algebra libraries for large or constrained models.
The library is tested on modern Python versions and integrates cleanly into scientific Python environments.
Use Cases
Interpretable regression and classification
Exploratory data analysis
Modeling non-linear relationships with transparency
Statistical learning in applied research
Alternatives to black-box machine learning models
Scientific Foundation
pyGAM is academically grounded and citable via Zenodo. It is used in research and applied settings where model interpretability and statistical validity are essential.
Open Source
pyGAM is released under the Apache License 2.0 and developed openly on GitHub. The project welcomes contributions, bug reports, and feature ideas from the community.
GC.OS supports pyGAM as an open-source project that enables transparent, interpretable, and statistically sound machine learning.
