search

torchmd / torchmd-net

Training neural network potentials
MIT License
335 stars 75 forks source link

Update TensorNet-QM9.yaml #230

Closed guillemsimeon closed 1 year ago

guillemsimeon commented 1 year ago

I updated the optimal hyperparameters for TensorNet on QM9 to obtain the results of the camera-ready paper, which give an improvement over previous accuracy.

RaulPPelaez commented 1 year ago

You changed the seed, but I guess this is just a side effect, right? Like it should not be changing the accuracy noticeably. If this is the seed you used for the paper then its ok for reproducibility.

guillemsimeon commented 1 year ago

Yes, the seed for the paper, the learning rate, and the batch size.

guillemsimeon commented 1 year ago

I tried multiple seeds, and the effect of changing the LR and the batch size is statistically significant compared to previous results.