WARNING: this is legacy code, which is no longer maintained!! The TensorFlow layers of this repo are now improved and part of the GUDHI library (https://gudhi.inria.fr/python/latest/).

Description

The notebook illustrations.ipynb implements the experiments presented in the article "Optimizing persistent homology based functions". The experiments on filter selection can be run with the code in optim_filter/ folder. This code was meant to be run on cluster, so if you do not use a cluster, please remove the oarsub -S and the python specific Python environment loadings (module load ..., eval ... and conda activate ...) in the .sh files, and update the variable path (at the beginning to the .py files) to the local Github repository path on your machine.

Dependencies

Our code is based on the Gudhi library (http://gudhi.gforge.inria.fr/python/latest/), which can be installed, e.g., by running conda install gudhi in an Anaconda environment. Our code also depends on Tensorflow 2.4.1.

Data sets

Most external data sets are available in the data repository. The MNIST data set is available in Tensorflow 2.4.1. Graph data sets have to be extracted from graphs.zip.