Geographically weighted modeling based on scikit-learn.

The aim of the package is to provide implementations of spatially-explicit modelling.

Current development status is early beta. API of the package can change without a warning. Use with caution.

gwlearn provides a framework for prototyping geographically weighted extensions of regression and classification models based on scikit-learn and libpysal.graph and a subset of models implemented on top of this framework. For example, you can run geographically weighted linear regression in a following manner.

import geopandas as gpd
from geodatasets import get_path

from gwlearn.linear_model import GWLinearRegression


gdf = gpd.read_file(get_path('geoda.guerry'))

adaptive = GWLinearRegression(
    geometry=gdf.representative_point(),
    bandwidth=25,
    fixed=False,
    kernel='tricube'
)
adaptive.fit(
    gdf[['Crm_prp', 'Litercy', 'Donatns', 'Lottery']],
    gdf["Suicids"],
)

For details, see the documentation.

The package is currently not released, so you will need to install it from source.

You can either clone the repository:

git clone https://github.com/pysal/gwlearn.git
cd gwlearn
pip install .

Or install directly from Github:

pip install git+https://github.com/pysal/gwlearn.git

The package depends on:

geopandas>=1.0.0
joblib>=1.4.0
libpysal>=4.12
numpy>=1.26.0
scipy>=1.12.0
scikit-learn>=1.4.0
pandas>=2.1.0

To search for or report bugs, please see the Github issue tracker.

If you have a question regarding gwlearn, feel free to open an issue or join a chat on Discord.

The package is licensed under BSD 3-Clause License (Copyright (c) 2025, Martin Fleischmann & PySAL Developers)