support optional data_path args

[PythonFZ]

• tests

[M-R-Schaefer]

thanks! does this just allow training without an AddData node or does it also fix the issue of copying IPSuite Atoms to the node directory?

[PythonFZ]

thanks! does this just allow training without an AddData node or does it also fix the issue of copying IPSuite Atoms to the node directory?

It would also fix the issue with copying the data, by telling the apax node the path to the existing file. If the general concept of the Node is fine by you, I'll add some tests.

There is one alternative to this, I want to mention. We could disallow passing list[ase.Atoms] to the node and require a pre-training node, which saves the configurations to a file somewhere that apax then uses. This would simplify the node but would require two nodes then.

file = apax.nodes.FramesToFile(data: list[ase.Atoms] = ...)
apax.nodes.Apax(data_file = file.path)

I am not entirely sure which way I'd prefer. Benefits here would also be, given you have a selection from a very large file which might take long to read will be cahced in the FramesToFile node and changing model parameters would not affect it.

[M-R-Schaefer]

lgtm