Make the MACE fit more flexible and build a finetuning workflow #182
Conversation
Update pyproject
docs/user/fitting/fitting.md
Outdated
| Currently, this can only be done by cloning the git-repo and installing it from there: []() | ||
|
|
||
| It is now important that you switch off the default settings for the fitting procedure (use_defaults_fitting=False). | ||
| If one wants to perform very low-data fine-tuning (less than 4 additional structures), please also think about switching off `preprocessing_data` |
There was a problem hiding this comment.
@QuantumChemist can you confirm that this is enough to skip the train, test split and simply take all data?
There was a problem hiding this comment.
But then the MLIPFitMaker will only run if train.extxyz and test.extxyz files are present in the database_dir
autoplex/autoplex/fitting/common/flows.py
Line 182 in fdbee8e
There was a problem hiding this comment.
We should think about a way to circumvent this in the future. I will implement a basic implementation first.
There was a problem hiding this comment.
It was implemented this way by @YuanbinLiu for the RSS part, so we might be careful with the changes until the rest of the RSS part is fully merged.
There was a problem hiding this comment.
Yeah. I will only add the option not change the default. This is what I have done for the rest of the implementations here as well. The defaults have not changed.
|
@YuanbinLiu FYI, I will merge this and then push a few tiny changes to #203 to fix the merge conflicts. I have prepared everything in #213 already and this should not cause any problems for your pull request. |
Closes #178