mordred descriptors

[JochenSiegWork]

Mol2Any element for calculating mordred descriptors using https://github.com/JacksonBurns/mordred-community

The original mordred repo is not maintained anymore and has some dependency issues like numpy<2. See https://github.com/mordred-descriptor/mordred

• Check where to put the extra mordredcommunity dependency in the extra deps
• Check where to put the unittest. Wait for Restructure tests #255
• A significant portion of all mordred 2d descriptors just fail for standard molecules. Check why and perhaps remove the descriptors or improve configuration.
• Mordred may oversubscribes CPUs. We need to add an mechanism to control CPU cores. Either setting BLAS, OpenMPI env vars etc manually or probably better use joblib's threadpoolctl
  • Check that implementation doesn't oversubscribes cores on HPC env

[JacksonBurns]

@JochenSiegWork you can set nproc in the call to Calculator.map (see here) to control the number of CPU cores used for multiprocessing. Not sure about the underlying parallelization of numpy operations, but I don't think scikit-learn typically controls those, do they?