how to train casanovo v4 on huge dataset like massive-kb from scratch? about casanovo HOT 3 CLOSED

1. Rather than a specific number of epochs (or iterations for such a large training dataset), I recommend that you determine how long to train based on a sequence-disjoint validation set. That's also what we did when training on MassIVE-KB, and based on the validation performance we empirically noticed that the model converged already prior to having gone through a single full epoch.
2. No, these are the settings that we consistently use to train the various models. No special change needed.

from casanovo.

Yes, that's right.

from casanovo.

Thanks a lot for your advice!

If I would like to fine-tune your massivekb pretrained model for my specific data, I simply need to continue train casanovo with your model and my own data (my data might be slightly different from massivekb) until it performs best on my own validation set?

from casanovo.