- Pull request to
torchvision
for TinyImageNet - Redesign TransformationSet, make it just a list of transformations, remove the parameters and valid_transformation methods. Move parameters to tmexp.
- Add documentation to classes and functions
- Generate html docs from docstrings
- Fix experiments on other repo to use new API
- Fix non-deep names in AutoActivations
- Label font size for visualization must depend on layer name. Else, use last part of layer name or use type.
- Fix errors propagating correctly when a thread crashes
- Add labels to TQDM bars to indicate the measure being computed (use abbreviation)
- Re-add Distance implementations
- Add Numpy-Pytorch adapter.
- Check Numpy and Pytorch implementations match in their results
- Add graph or tree visualization (or at least a hierarchical visualization) using colors to indicate invariance