Arseny Kravchenko

Users may want to use custom metrics. At the same time, current metrics set may be excessive: e.g. L2 norm and mean value and correlated a lot.

``` In [1]: import torch In [2]: from hookandlook.watch import Wrapper In [3]: model = Wrapper.wrap_model(torch.nn.Linear(10, 10)) In [4]: torch.save(model, '/tmp/model.pth') --------------------------------------------------------------------------- AttributeError Traceback (most recent call last) in ---->...

correction for soft targets loss may be needed

1

Quoting original paper (block 2): > Since the magnitudes of the gradients produced by the soft targets scale as 1/T**2 > it is important to multiply them by T**2 when...

add Dockerfile

1

This PR adds an option to run the script in Docker, which simplifies usage for people who don't have PHP installed.

Hi AugLy team, Thanks for the great package! I have seen your comparison with other libs in the paper and have to highlight a missing point: in the table below,...

Add tests on CI

1

There is a test file https://github.com/arsenyinfo/qudida/blob/main/test_transforms.py, but it can be run manually only at the moment. Add a Github Actions check that will run this test using pytest for every...

http://www.inference.vc/mixup-data-dependent-data-augmentation/