ensemble-integration (or eipy) leverages multi-modal data to build classifiers using a late fusion approach. In eipy, base predictors are trained on each modality before being ensembled at the late stage.
This implementation of eipy can utilize sklearn-like models only, therefore, for unstructured data, e.g. images, it is recommended to perform feature selection prior to using eipy. We hope to allow for a wider range of base predictors, i.e. deep learning methods, in future releases. A key feature of eipy is its built-in nested cross-validation approach, allowing for a fair comparison of a collection of user-defined ensemble methods.
Documentation including tutorials are available at https://eipy.readthedocs.io/en/latest/.
As usual it is recommended to set up a virtual environment prior to installation. You can install ensemble-integration with pip:
pip install ensemble-integration
If you use ensemble-integration in a scientific publication please cite the following:
Jamie J. R. Bennett, Yan Chak Li and Gaurav Pandey. An Open-Source Python Package for Multi-modal Data Integration using Heterogeneous Ensembles, https://doi.org/10.48550/arXiv.2401.09582.
Yan Chak Li, Linhua Wang, Jeffrey N Law, T M Murali, Gaurav Pandey. Integrating multimodal data through interpretable heterogeneous ensembles, Bioinformatics Advances, Volume 2, Issue 1, 2022, vbac065, https://doi.org/10.1093/bioadv/vbac065.