"Distances" attribute in OC20 dataset IS2RE task

1

Hi,
I am attempting to predict Bader charges from the OC20 dataset. However, the ‘distances’ attribute in the OC20 IS2RE dataset, which takes the form of a 1D array, is not defined in the READ_ME File. What does each value in the array correspond to? Is it the distance between two atoms, and if so, how can I determine which two atoms?

Thanks so much!

2

Hi -

Sorry for the delay here. Yes, the distances corresponds to the distance between two atoms. Specifically, edge_idx tells you which atom indices. As an example:

dataset[0].edge_index[:, 0]
--> tensor([55,  0])
dataset[0].distances[0]
--> 3.4

Suggesting the distance between atom index 55 and 0 is 3.4.