search

gomes-lab / H-CLMP

MIT License
5 stars 6 forks source link

Where is the implementation of CrabNetMP and CrabNetMP(T) from the paper? #1

Open sgbaird opened 2 years ago

sgbaird commented 2 years ago

@sk2299, a search for "CrabNet" within the repository returns no results.

sk2299 commented 2 years ago

Hey, thanks for the question. As mentioned in the "baseline models" section of the paper, we extended CrabNet to multi-property prediction by changing the last layers of the DNNs to produce the desired output dimension. This should be easy to implement upon CrabNet's codes.

sgbaird commented 2 years ago

@sk2299, thanks for the quick reply! As I was digging a bit deeper, I found the relevant place in CrabNet, and while the change looks simple, it might be a bit involved to actually get it working. Do you still have a copy of your implementation? It would be nice to have a working example to compare/debug against. Also, what about for CrabNetMP(T) (transfer learning)? From the paper, I see:

Specifically, we concatenated each generated DOS vector to the respective latent vector from the model’s encoder, creating a latent representation with MPDOS transfer learning that served as the input to the model’s decoder.

Do you have a copy of your implementation for this one, too?