non-urgent todos regarding the MLIPs in autoplex

[QuantumChemist]

just want to note that the following changes should be done to improve the user experience with autoplex

• automatic fit of separated and combined dataset for all MLIPs (now it's just implemented for GAP)
• add default hyperparameter json file

[vlderinger]

Is this a pure enhancement, or are there questions about how we want to separate the datasets? (I'm asking so we can assign the right "label" to this issue)

[JaGeo]

I think @QuantumChemist is referring to separation into single displaced supercells and random ones. This is a feature that might be useful for the phonon part

[QuantumChemist]

I think @QuantumChemist is referring to separation into single displaced supercells and random ones. This is a feature that might be useful for the phonon part

I mean this here. For now, one can only do the fit separately for single-atom-displaced, rattled supercells and the combined dataset for GAP: All the other MLIP can only use the combined dataset.

[vlderinger]

Ah! Yes it seems to me to make sense to let the user combine datasets, and to choose their MLIP framework, and to do both of these things independently from one another. (But not needed urgently I think)