Closed MarcoGorelli closed 3 years ago
Seems that removing papermill
from the metadata does the trick
Thanks for reporting. Looks like metadata created by papermill is not being handled well. We're going to dig some deeper and see if we can fix this behaviour.
Cool, thanks! To be clear, I just needed to remove papermill
from the notebook's metadata, not from each cell's metadata
We face the same problem but we have no papermill metadata, only long outputs which may throw a wrench in the diff. Do hiding cell outputs within ReviewNB performs a new diff without considering them? (I suspect not). Our PR at https://app.reviewnb.com/cfoisy-osisoft/OSI-Samples-OCS/pull/5/files/
@MarcoGorelli We've deployed a fix to handle notebook level metadata more carefully. You shouldn't see any issues due to papermill metadata going forward. The fix can be seen on this commit: https://app.reviewnb.com/pymc-devs/pymc3/commit/cf435e1560f81c5e3d61d0666c3bbd7fbe12290b/
@cfoisy-osisoft If there are a lot of cell level changes (cells being split, merged, deleted, added) in a given PR/commit then our algorithm might not be able to compute line by line diff. In such cases we resort to simple diff which is what has happened in your case.
Do hiding cell outputs within ReviewNB performs a new diff without considering them? (I suspect not).
No. "Hide Output" is just a UI switch to hide outputs, it doesn't recompute the diff.
Here's the PR - https://github.com/pymc-devs/pymc3/pull/4215
Rather than showing something like
it is instead showing
and
Any ideas why?
For the second pair of screenshots, I ran the notebook on Kaggle - does any of the metadata they add make it look like all the cells are completely different?