_format_experiment_name() unusable

[MarshallYan]

The function _format_experiment_name() in training.py is unusable if ever the name assigned by user is not {potential_name}_{dataset_name}, since it is hard-coded for this naming convention.

This function should be changed to be able to handle any parameters parsed in configuration.

[MarshallYan]

This function should be limited to only handle simplest cases (like {potential_name}_{dataset_name}). Documentations should be added to specify this case and any more complicated cases should be handled by users outside of the modelforge core project.

[chrisiacovella]

As we discussed, making the formatting function generally accept any key in the .toml files is probably not really the best way to go, since those keys are not necessarily unique (we could have something like a parameter named "max" in various places in the .toml files, like just made up things here, but something like potential.core_parameters.max potential.embedding.max, or training.scheduler.max). A better approach to being able to automatically format the name based on some set of experiments we are performing would be to do this outside of model forge. i.e., write a script that defines the experiment_name in a way that makes sense to the parameters being modified.

As we discussed for the experiments we are working on, writing some scripts using Jinja (https://jinja.palletsprojects.com/en/stable/) would probably be a good approach.

[chrisiacovella]

But I agree, this needs to be clear in the documentation. We actually do not discuss the "runtime.toml" parameters in their own section (it's in the getting started), so we should probably create a new .rst file that specifically defines that information.