gwlearn

Geographically weighted modeling based on scikit-learn.

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

Features

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

For details, see the documentation.

Status

Current development status is beta. The core API of the package should not change without a warning and a proper deprecation cycle. However, minor breaking changes may still occur.

Installation

You can install gwlearn from PyPI or from conda-forge using the tool of your choice:

pip install gwlearn

Or from conda-forge:

conda install gwlearn -c conda-forge

Bug reports

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

Get in touch

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

License

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

Funding

Charles University’s Primus programme through the project "Influence of Socioeconomic and Cultural Factors on Urban Structure in Central Europe", project reference PRIMUS/24/SCI/023.