Dataset can be found at :
https://zenodo.org/records/2603256
I used the following libraries for a highly efficient and memory-saving representation and computation of jet features (pre-processing) and for the computation of Persistent Homology as a wrapper for the ParticleNet model.
For the computation of Persistent Homology of the jet clouds, I use Pyrivet which is a Python API for the RIVET library that allows multiparameter persistence models to be fitted on discrete data, describing it in terms of a continuous shape - an Abstract Simplicial Complex.
To initially make single-parameter estimators such as that which has been currently implemented in 1d_persistence.py I have used the topological methods in gtda and also Ripser library.
The following figure describes the architecture for the conventional ParticleNet model that uses the EdgeConv operation on the k-NN graph :
We replace the k-NN graph step by the Persistence step.
The workflow for the project is :
