[Question] Do input feature is fixed?

[feyhong1112]

Sorry to ask the question. I am the beginner on this. I have very big gab in knowledge in here. Thank you for visit my post.
Base on paper do the feature input is fixed with int like one-hot encoding?
My input feature is the embedding atom encode from Bert model, the data inside is arbitrary float with positive and negative with continues numeric type.

[yilunliao]

Hi @feyhong1112

The inputs are as here: https://github.com/atomicarchitects/equiformer/blob/master/nets/equiformer_md17_dens.py#L238-L242
The above is for MD17 but can be generalized to other datasets.

One of the inputs is atomic numbers, which uses long int to represent atom types. This long int can be converted to a vector of floating points (same as the "atom encode from Bert model" in your question) by (1) first converting long int into one-hot vectors and (2) then applying a standard linear layer to the one-hot vectors.