A Python package containing ML tools for use with TileDB-SOMA.
Docs: single-cell-data.github.io/TileDB-SOMA-ML.
NOTE: this is a pre-release package, and may be subject to breaking API changes prior to first release.
The package contains a prototype PyTorch IterableDataset, ExperimentDataset, for use with the torch.utils.data.DataLoader API.
notebooks/ contains tutorials and examples that use this repo to train toy models. For a general introduction to PyTorch data loading, see this tutorial. Additional information on the DataLoader/Dataset pattern can be found here.
Defects and feature requests should be filed as a GitHub issue in this repo. Please include a reproducible test case in all bug reports.
Install from PyPI:
pip install tiledbsoma-mlDevelopers may install editable, from source, in the usual manner -- clone the repo and execute:
pip install -e .Documentation can be found at single-cell-data.github.io/TileDB-SOMA-ML, and in the notebooks directory.
This is a pure Python package. To build a wheel, ensure you have the build package installed, and then:
python -m build .See the CHANGELOG.md file.
This project is licensed under the MIT License.
The SOMA team is grateful to the Chan Zuckerberg Initiative Foundation CELLxGENE Census team for their initial contribution.