search

FabianFuchsML / se3-transformer-public

code for the SE3 Transformers paper: https://arxiv.org/abs/2006.10503
475 stars 69 forks source link

Is caching is different for pre/post the update 2 month ago? #19

Open hiranumn opened 3 years ago

hiranumn commented 3 years ago

First of all, thank you for this great work!

I had a model trained before the update you made 2 months ago. Since then I updated everything to match the new version except that the cache folder stayed constant (accidentally).

I was experiencing any problem until today when I tried to run the trained model with the newly generated cache folder.

So my questions are, 1) Is the cache generation process any different between pre/post update? 2) Is there any concern if you train a new model with the post-update package with the old cache? Particularly, could it invalidate some of the nice assumptions that this model has?

Thank you so much in advance! Nao

FabianFuchsML commented 3 years ago

  1. I don't think we changed anything about the caching for a long time (if ever since the first release); but if you suspect a change somewhere, git diff allows you to compare different commits, so maybe that might help

  2. as my answer to 1. is no, I don't suspect that using 'old caches' would lead to problems

I hope this helps :) Fabian