search

ppdebreuck / modnet

MODNet: a framework for machine learning materials properties
MIT License
81 stars 34 forks source link

Backwards compatibility of test data with pymatgen #206

Closed ml-evs closed 8 months ago

ml-evs commented 8 months ago

As described in #204, our test data is losing compatibility with newer versions, as we used pickle rather than to_dict to serialize it. In the longer term, we will probably have to regenerate this data or refactor our tests, but for now this PR hot patches the loaded pickles with the required attributes for latest pymatgen versions.

This (somewhat bizarrely) allows more matminer featurizers to run successfully, so I had to also allow new features to be added on top of the reference data. This PR also allows the matminer ignore_errors arg to be controlled at the level of a featurizer, which is useful for debugging.

ml-evs commented 8 months ago

Hi @ppdebreuck, I was able to get everything working with the latest pymatgen, so think this is ready to release now.