✨ Docker settings, BatchEnsemble utilities rework + minor fixes #141
Conversation
Co-authored-by: Anton <anton.zamyatin@student.tuwien.ac.at>
🐛 Fix MNIST test dataloader for shifted data Also fast-tracked another PR from @tonyzamyatin that got lost with the documentation due to a scheduling mistake.
- now the input is repeated during training depending on `repeat_training_inputs` argument - with `convert_layers==True` all `nn.Linear` and `nn.Conv2d` layers are replaced by `BatchLinear` and `BatchConv2d`
✨ LeNet BatchEnsemble and Deep Ensemble + minor bugfixes
o-laurent
added
bug
Codecov ReportAll modified and coverable lines are covered by tests ✅
✅ All tests successful. No failed tests found. Additional details and impacted files@@ Coverage Diff @@
## main #141 +/- ##
==========================================
+ Coverage 99.15% 99.19% +0.03%
==========================================
Files 143 144 +1
Lines 7136 7192 +56
Branches 910 925 +15
==========================================
+ Hits 7076 7134 +58
+ Misses 27 26 -1
+ Partials 33 32 -1
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Sentry. |
|
|
||
|
|
||
| class BatchEnsemble(nn.Module): | ||
| """Wrap a BatchEnsemble model to ensure correct batch replication. |
There was a problem hiding this comment.
We decided to put all class docstrings below the __init__ for now, hence we should move it.
There was a problem hiding this comment.
Alright, I just thought it would be cool to be able to view the docstring with your editor's "inline quick documentation" keyboard shortcut :)
There was a problem hiding this comment.
Ideally (unless mistaken), we should have docstrings split into two parts for all classes, but this would need a lot of time. If we don't do this, I tend to prefer having the arguments available when I instantiate a class so that I know what they mean instead of having them when hovering over the class itself. It would work like this, at least on VSCode; I don't know about PyCharm, which you seem to be using?
Co-authored-by: o-laurent <olivier.laurent@ensta-paris.fr>
| raise ValueError("`num_estimators` must be greater than 0.") | ||
|
|
||
|
|
||
| def batch_ensemble( |
There was a problem hiding this comment.
Does it make sense to have this function and the class since they have exactly the same arguments?
The scope of this comment is actually larger because we have this kind of structure in many different files, even though we sometimes discourage using the classes by putting underscores.
|
🥳👏 |
Also:
WandbykorniaFixes #136
Fixes #138