immediate MLIP (MACE) evaluation of sampled MD configurations.

[williamdavie]

It will be useful to have an option e.g. named evaluate_committee=True to automatically call mace.cli.eval_configs on configuration generated via scripts from DataCollection.write_MD_submission_scripts().

A possible issue with this is how the jobs are distributed on the respective cpu/gpu. Though this could be dealt with by say, separate scripts with an automatic dependancy.

Edit:

I actually think it's better to do this the other way round. When 'ActiveLearn.evaluate_committee' is called there is an option to 'fetch configs' and this will use a provided lammps base directory and variables to sample the configurations.

[williamdavie]

The issue now facing all of MLIPTS is too many input parameters, a lot of these are repeated so I will create a sub issue to fix this.