Use of VietorisRipsPersistence Module on Big Data and parallelization of code

[ShreyamMishra]

Hello all, I've been using giotto's VietorisRipsPersistence module for generating Persistence diagrams from Vietoris-Rips filtrations. However, a key issue I'm facing is with respect to the size of my point cloud. My point cloud database has patient time-series data, with the point cloud of each patient having 12K+ points. Performing the computation of the homology groups on my local system (Intel i9 with 64GB DDR4 RAM), for even a subsampled a pointcloud of 1000 from 12K+ points takes an extremely long time, so doing it for the entire pointcloud for even a single patient has been impossible.

Regarding the above, I had a few questions:

1. Is there a parallelized version of the VietorisRipsPersistence module available? If not, is there any other TDA library that has parallelized versions available?
2. Is there a simpler, faster serialized way to perform the above that I am missing?

I'm extremely new to TDA and parallel computing, so apologies incase I'm missing something obvious. Thanks in advance!

[ulupo]

Hi! You should check out giotto-ph, which provides a parallelized version of the Vietoris-Rips persistence calculation. giotto-ph is actually a dependency of giotto-tda so you should have it installed already. Of course, another standard trick is to use a threshold (max_edge_length parameter). Furthermore, if your point cloud is in low dimension (say < 6) you should consider an alpha or alpha-type persistence instead of Vietoris-Rips. The true alpha complex is not implemented in giotto-tda, but an approximate version is implemented as WeakAlphaPersistence. You could check out the GUDHI library (not our project) for a true alpha persistence calculation.