og-learn

Overfit-to-Generalization Framework for Equitable Spatiotemporal Modeling

Overview

og-learn is a Python library that implements the Overfit-to-Generalization (OG) framework, designed to address local overfitting in spatiotemporal machine learning models. The framework redistributes predictive capacity from data-rich to data-poor regions through a two-stage approach.

The Problem: Local Overfitting

Models trained on non-uniformly distributed observations tend to:

• Overfit to data-dense regions (achieving high accuracy)
• Underperform in sparse areas (poor generalization)
• Show inconsistent optimal hyperparameters across different validation groups

This disparity is often invisible when using conventional global metrics.

The Solution: OG Framework

The OG framework exploits complementary properties of different model architectures:

1. Stage 1 (Overfit): A high-variance model (e.g., LightGBM) is deliberately overfitted to capture local patterns and generate pseudo-labels.

2. Stage 2 (Generalization): A low-variance model (e.g., MLP, Transformer) learns from the pseudo-labels with density-aware sampling and noise injection.

Installation

pip install og-learn

Quick Start

from og_learn import OGFramework
from og_learn.models import LightGBMOverfitter, TransformerRegressor

# Initialize the OG framework
og = OGFramework(
    hv_model=LightGBMOverfitter(),  # High-variance model for Stage 1
    lv_model=TransformerRegressor(), # Low-variance model for Stage 2
)

# Fit the model
og.fit(X_train, y_train, coords=coordinates)

# Predict
predictions = og.predict(X_test)

Key Features

• Density-Aware Sampling: Prioritizes sparse regions and rare covariate patterns
• Unlimited Synthetic Target Generation: Generates high-fidelity pseudo-labels via spatial perturbation
• Flexible Model Pairing: Supports various HV/LV model combinations
• Cross-Validation Strategies: Built-in time-wise, site-wise, region-wise, and spatiotemporal-wise CV

Supported Models

High-Variance (HV) Models

• LightGBM
• XGBoost
• CatBoost

Low-Variance (LV) Models

• MLP (Multi-Layer Perceptron)
• Transformer Regressor
• ResNet

Documentation

📖 Full documentation: https://px39n.github.io/og-learn/

Data

Example datasets are available on Zenodo: https://doi.org/10.5281/zenodo.17915696

Direct download: og-learn.zip

Citation

If you use this package in your research, please cite:

@article{liang2025og,
  title={Countering Local Overfitting for Equitable Spatiotemporal Modeling},
  author={Liang, Zhehao and Castruccio, Stefano and Crippa, Paola},
  journal={},
  year={2025}
}

License

This project is licensed under the MIT License - see the LICENSE file for details.

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.