Torch-TDA

A package used to do automatic differentiation in Pytorch for topological data analysis.

This package provides utilities for using constructions in topological data analysis with automatic differentiation. It wraps functionality from

BATS for Basic Applied Topology computation.
persim for computations comparing persistence diagrams.
hera a library for fast calculation of bottleneck distance and Wasserstein distance on persistence diagrams.
topologylayer for polynomial features of barcodes.

The design is heavily inspired by topologylayer. Key differences are that we use bats for faster topological computations.

Here is the documentation and examples.

Reference

If you want to reference this project, please consider referencing one of these two papers or both because it is a side-product of the two papers:

Use

Package installation provides a package under the torch_tda namespace. Functionality is primarily under torch_tda.nn, which provides several PyTorch layers.

import torch_tda

Installation

Attension: please use Linux OS to install for now and the support for Mac OS will come soon.

First, it is recommended to set up a conda environment

conda create -n bats
conda activate bats

If you are installing torch_tda from the development version, please install the development version of BATS.py, install it from source.

Otherwise, you can use the latest release of BATS.py

pip install bats-tda

Now, to use torch_tda, you will need Pytorch pre-installed and then

pip install torch-tda

Hera

If you want to compute bottleneck distance, you will need to install hera by 1. following their installation description to build from source and install the Python binding 2. (Optional if you can put the Python package in the correct position) write a simple setup.py file to install the package into environment path.

Build from source

If you want to install from source, after cloning the repository, you can setup with setup.py

python setup.py install

Attention:

Note that installing by pip install torch-tda in the cloned repository will probably install the older version from remote, see the explanation. So, DO NOT use pip if you want to build from the latest source.
If you have installed an older version of torch-tda and want to update to the newest version, please uninstall it first by pip uninstall torch-tda.

Documentation

If you want to contribute to the documenation, you can add some jupyter notebooks to the docs/examples folder and then generate documentation using Sphinx

cd docs
pip install -r requirements.txt
make html
xdg-open _build/html/index.html # or just open this file