Integrating MDANCE functionality in CPPTRAJ

[ramirandaq]

We finally have C++ versions of some MDANCE functionality (NANI and HELM for now, DIVINE coming soon). Maybe this could be a good starting point to get some of these clustering methods in CPPTRAJ? This is the current repo: https://github.com/mqcomplab/CPP-MDANCE

[drroe]

This is great! I'm hoping to merge the code from here to make CPPTRAJ v7.0 (https://github.com/drroe/cpptraj/tree/prepareforleap.remote.buildatom), and after that I will try to make this a top priority. Thanks!

[ramirandaq]

Quick update: DIVINE is done, unit-tested, and already merged in the main branch of CPP-MDANCE. These should be the 3 key methods to include in CPPTRAJ. In summary:
1- NANI: Proposed to become the "default" k-means, since it guarantees 100% reproducibility and the last implementation was already faster than the k-means++ implementation of scikit-learn.
2- HELM: agglomerative hierarchical method reducing time and (in our opinion, more important) memory bottlenecks by operating at k-means levels of efficiency.
3- DIVINE: divisive hierarchical method.
TODO for upcoming integrations: multi-trajectory analysis (SHINE, PRISM algorithms); online clustering (mdBIRCH).