Hi, thanks for sharing the code. With the code, I trained the model locally using FreeSolv dataset. But with many trials, I can't reproduce the result in the NJP paper Fig. 3.
I acknowledge that the random split of training and validation set contribute to the variation of the model and subsequently the OOD prediction. I used the default setting, is there anything that I can do to improve the OOD performance?
Hi, thanks for sharing the code. With the code, I trained the model locally using FreeSolv dataset. But with many trials, I can't reproduce the result in the NJP paper Fig. 3.
I acknowledge that the random split of training and validation set contribute to the variation of the model and subsequently the OOD prediction. I used the default setting, is there anything that I can do to improve the OOD performance?