search

jasonkyuyim / se3_diffusion

Implementation for SE(3) diffusion model with application to protein backbone generation
https://arxiv.org/abs/2302.02277
MIT License
333 stars 54 forks source link

Small issue about warm start in training code #34

Closed Immortals-33 closed 1 year ago

Immortals-33 commented 1 year ago

Hi Jason & Brian, thanks for this inspiring work!

When I look through the training code, a potential issue occured:
https://github.com/jasonkyuyim/se3_diffusion/blob/53359d71cfabc819ffaa571abd2cef736c871a5d/experiments/train_se3_diffusion.py#L97-L103
Seems that the ckpt_dir is referred as a boolean value here. Was it intended to set to ckpt_dir = conf.experiment.ckpt_dir ? https://github.com/jasonkyuyim/se3_diffusion/blob/53359d71cfabc819ffaa571abd2cef736c871a5d/config/base.yaml#L99-L102
Btw running the training code without setting the warm_start option is OK.

jasonkyuyim commented 1 year ago

Hi and thanks for the kind words. warm_start is the checkpoint directory of the weights we want to start training from. If warm_start is null then the weights are initialized from scratch.ckpt_dir is the directory the current experiment will saved checkpoints to. So the logic shuld be correct. I should've named the variables differently, my apologies.