Open zhonge opened 1 year ago
Summarizing the above discussion on a different pull request. Some cells in the jupyter notebooks assume poses include both rotations and translations, and that there are CTF parameters... which is obviously the case for real datasets, but our toy dataset in testing
currently only has rotations and no CTF. The software still runs, but the jupyter notebooks currently assume real data (which just prevents testing at the moment).
I think we should support both idealized and real imaging, and
Based on #355 we have users who get stuck on trying to run the notebooks after trying training without a CTF — +1 from me for fixing the notebooks to support cases without a CTF, and to clean up and document the test case datasets a bit too, see e.g. #375.
Ah yes, I think we should open a new ticket for updating the toy datasets. Right now the toy dataset doesn't have any translations (just rotations since the images are perfectly centered). There is also no CTF applied on this dataset. I'm pretty sure this idealized setting breaks some of the assumptions in the notebooks, and we can discuss whether we need to support these idealized settings in the code and in the jupyter notebooks.
Originally posted by @zhonge in https://github.com/zhonge/cryodrgn/issues/240#issuecomment-1484144739