❓ [QUESTION] How to perform transfer training with nequip? #245
-
|
I have trained a nequip model M with a large dataset A. What can I do to make M also applicatable to a small but different dataset B? (except training a new model with A+B) |
Beta Was this translation helpful? Give feedback.
Replies: 2 comments
-
|
See #235 for discussions on how to do fine-tuning (which seems to be what you are asking?) in Note that if A and B are very different this might not work very well... in general I believe this is still an open questions for machine learning atomic systems. Please let us know if you find anything interesting! In particular, do note that B cannot contain any species that were not defined in the original model. (Technically it is possible to define a species in the original model that is not present in the original data A, though I don't know how well that would work in terms of accuracy.) |
Beta Was this translation helpful? Give feedback.
-
|
Thank you very much! That is what I want. |
Beta Was this translation helpful? Give feedback.
See #235 for discussions on how to do fine-tuning (which seems to be what you are asking?) in
nequip.Note that if A and B are very different this might not work very well... in general I believe this is still an open questions for machine learning atomic systems. Please let us know if you find anything interesting!
In particular, do note that B cannot contain any species that were not defined in the original model. (Technically it is possible to define a species in the original model that is not present in the original data A, though I don't know how well that would work in terms of accuracy.)