search

atomistic-machine-learning / schnetpack

SchNetPack - Deep Neural Networks for Atomistic Systems
Other
774 stars 214 forks source link

Adjust learning rate layerwise #601

Closed ThorrenKirschbaum closed 7 months ago

ThorrenKirschbaum commented 7 months ago

Dear Schnetpack Team,

is it possible to adjust the learning rate of PaiNN to have different learning rates at each layer of the network? If yes, how can I do it?

Many thanks in advance Thorren

stefaanhessmann commented 7 months ago

Hey Thorren,

Yes this should be possible. But I think this question is more about pytorch than about schnetpack. Since we use pytorch to build our models, everything that is possible with pytorch should also work in schnetpack. However, we have not implemented a solution for this. I assume, you would need to build your own script and manually define how to change the learning rates. From this post in the pytorch forum it seems to be an easy fix.

Hope this helps to answer your question!

Best, Stefaan

ThorrenKirschbaum commented 7 months ago

thank you!