[WIP] Starting point for batch-level data augmentation #254

cedricrommel · 2021-06-21T09:40:17Z

No description provided.

braindecode/augmentation/base.py

robintibor · 2021-06-21T11:26:09Z

Hi, very interesting thanks, seems a lot of work, will need to have a closer look! This works fine with what was done in https://github.com/braindecode/braindecode/pull/198/files ? Note the discussion there as well

tomMoral

A first batch of comments

braindecode/augmentation/base.py

braindecode/augmentation/functionals.py

braindecode/augmentation/transforms.py

cedricrommel · 2021-06-21T12:44:56Z

Hey @robintibor, thanks a lot ! Glad it looks interesting. @sbbrandt will soon add his nice MixUp implementation as well and we are still working on polishing things a little with him and @tomMoral. Concerning @hubertjb's PR, yes we are aware. We've discussed that when the PR was opened and have a few ideas on how to ensure both are compatible (since both logics of batch-level and sample-level data augmentation have their own pros and cons). We will get to that as soon as we are done polishing :)

tomMoral

Few more comments on transform.

braindecode/augmentation/transforms.py

Co-authored-by: Thomas Moreau <thomas.moreau.2010@gmail.com>

tomMoral

Last few comments for the day. I will do a second review tomorrow.
But this looks good overall.

test/unit_tests/augmentation/test_base.py

test/unit_tests/augmentation/test_transforms.py

Co-authored-by: Thomas Moreau <thomas.moreau.2010@gmail.com>

robintibor · 2021-06-21T14:40:36Z

Definitely looks like this is growing to something nice! Make sure to also create some examples, an add to API docs etc. .

-added mixup functional -adapted base to handle tuple as return from transform -added mixup loss -added mixup tests

braindecode/augmentation/base.py

robintibor · 2021-06-21T16:13:49Z

I would still also argue to write transforms that only transform X on a per example basis using the existing https://github.com/braindecode/braindecode/pull/198/files solution. And keep the dataloader way for those transformations that need access to either X and y or to batch-level examples? Or what do you think? Otherwise it becomes confusing when to use which one? You could try to recreate your Transform Superclass as an "ExampleTransform" superclass it would make this transitition easier.

robintibor · 2021-06-21T16:14:00Z

also @hubertjb let me know what you think

braindecode/augmentation/_functionals.py

braindecode/augmentation/base.py

braindecode/augmentation/transforms.py

braindecode/training/losses.py

braindecode/augmentation/_functionals.py

Co-authored-by: Alexandre Gramfort <alexandre.gramfort@m4x.org>

agramfort · 2021-06-25T10:10:13Z

grrrr ok ! @cedricrommel you are on the example now then?

…

robintibor · 2021-06-25T10:21:05Z

yes then he can change to [MRG] when example is there so I see it as well.

examples/plot_data_augmentation.py

cedricrommel · 2021-06-25T18:00:21Z

I've fixed things. Will push soon. Just making a final check at the html :)

cedricrommel · 2021-06-25T18:21:34Z

Ok Alles gut for me ! 😃 I just need confirmation from @sbbrandt that he is ok with my commit where I put his email address in the file to add him as contributor and then I'm ok for merging 💪
(fyi @robintibor @agramfort @tomMoral )

agramfort · 2021-06-25T18:42:24Z

@robintibor green button is yours !

robintibor · 2021-06-25T19:21:13Z

Great work, merged!
@sbbrandt mail is already in other parts of the repo, so should be fine.
Thanks so much for all the effort that went into this! Really great to have so much new functionality implemented in a clean way!

Here a celebratory gif:

agramfort · 2021-06-25T19:26:30Z

🎉🎉🎉🎉🎉

cedricrommel · 2021-06-25T22:30:55Z

Niiiice ! 👏 👏 👏 👏
Thank you both for the really nice event, as well as @hubertjb 💪

sbbrandt · 2021-06-26T07:15:03Z

🎉🎉🎉

Thank you guys! Also for critics and correction of errors.

From me also thanks for the event and thank you @cedricrommel for the nice augmentation implementation :)

Starting point for batch-level data augmentation

3427ec5

cedricrommel marked this pull request as draft June 21, 2021 09:41

sbbrandt reviewed Jun 21, 2021

View reviewed changes

braindecode/augmentation/base.py Show resolved Hide resolved

cedricrommel added 3 commits June 21, 2021 11:55

Update augmentation __init__ with all transforms

6b80411

Add first batch of augmentations tests

c7cec4d

Fix flake8 incompatibilities

6e8d7a2

tomMoral reviewed Jun 21, 2021

View reviewed changes

Improve Time Reverse docstring wording.

b9954ca

Co-authored-by: Thomas Moreau <thomas.moreau.2010@gmail.com>

tomMoral reviewed Jun 21, 2021

View reviewed changes

test/unit_tests/augmentation/test_base.py Outdated Show resolved Hide resolved

test/unit_tests/augmentation/test_transforms.py Outdated Show resolved Hide resolved

cedricrommel and others added 2 commits June 21, 2021 15:39

Fix typos in DownsamplingShift docstring

4cc427c

Co-authored-by: Thomas Moreau <thomas.moreau.2010@gmail.com>

Fixes and improvements in docstrings

a78228b

Co-authored-by: Thomas Moreau <thomas.moreau.2010@gmail.com>

-added Mixup transform

3911191

-added mixup functional -adapted base to handle tuple as return from transform -added mixup loss -added mixup tests

robintibor reviewed Jun 21, 2021

View reviewed changes

braindecode/augmentation/base.py Outdated Show resolved Hide resolved

braindecode/augmentation/base.py Outdated Show resolved Hide resolved

braindecode/augmentation/base.py Outdated Show resolved Hide resolved

braindecode/augmentation/base.py Outdated Show resolved Hide resolved

sbbrandt and others added 2 commits June 21, 2021 18:20

added underscore to cite references

ec1bfe7

Improve many things related to data augmentation

5b53329

tomMoral reviewed Jun 21, 2021

View reviewed changes

braindecode/augmentation/_functionals.py Outdated Show resolved Hide resolved

agramfort reviewed Jun 22, 2021

View reviewed changes

cedricrommel and others added 7 commits June 22, 2021 10:45

Remove useless part of get_mask and better wording

37a15c5

Fix quotes in docstrings (PEP127)

589f69b

Replace absolute by relative imports

4baee72

Make make_collateable function private

1ce153b

Co-authored-by: Alexandre Gramfort <alexandre.gramfort@m4x.org>

Fix a few typos

43e586f

Co-authored-by: Alexandre Gramfort <alexandre.gramfort@m4x.org>

Fix rebasing error (mixup was not being imported)

40e6c27

Small fixes in MixUp docstring (references mainly)

8d249c2

agramfort changed the title ~~[WIP] Starting point for batch-level data augmentation~~ [MRG] Starting point for batch-level data augmentation Jun 25, 2021

robintibor changed the title ~~[MRG] Starting point for batch-level data augmentation~~ [WIP] Starting point for batch-level data augmentation Jun 25, 2021

robintibor mentioned this pull request Jun 25, 2021

some nitpicks on SequenceSampler PR #280

Merged

sbbrandt and others added 4 commits June 25, 2021 17:52

added example for data augmentation

165c96b

Fix important bug with the augmentation masks

6f4ea20

removed whitespace

337892a

Add a few tests with probability<1.0

72696f2

agramfort reviewed Jun 25, 2021

View reviewed changes

examples/plot_data_augmentation.py Outdated Show resolved Hide resolved

examples/plot_data_augmentation.py Outdated Show resolved Hide resolved

examples/plot_data_augmentation.py Outdated Show resolved Hide resolved

examples/plot_data_augmentation.py Outdated Show resolved Hide resolved

Merge remote-tracking branch 'upstream/master' into augmentation

68b8ae8

robintibor reviewed Jun 25, 2021

View reviewed changes

examples/plot_data_augmentation.py Outdated Show resolved Hide resolved

cedricrommel added 7 commits June 25, 2021 20:02

Few fixes and additions to augmentation example

4237a51

Force converted tensor from numpy to hold floats

64cd470

Fix typos

7752551

Correct cedric email

eeb9a5f

Fix typos in example and add clarifications

7e3e43c

Flake8

2f51732

Fix naming

ea636e2

Last tiny fix :)

b7d5c72

robintibor approved these changes Jun 25, 2021

View reviewed changes

robintibor merged commit e5b3d73 into braindecode:master Jun 25, 2021

cedricrommel mentioned this pull request Sep 21, 2022

Augmentation Enhancements #408

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[WIP] Starting point for batch-level data augmentation #254

[WIP] Starting point for batch-level data augmentation #254

cedricrommel commented Jun 21, 2021

robintibor commented Jun 21, 2021 •

edited

Loading

tomMoral left a comment

cedricrommel commented Jun 21, 2021

tomMoral left a comment

tomMoral left a comment

robintibor commented Jun 21, 2021

robintibor commented Jun 21, 2021

robintibor commented Jun 21, 2021

agramfort commented Jun 25, 2021 via email

robintibor commented Jun 25, 2021

cedricrommel commented Jun 25, 2021

cedricrommel commented Jun 25, 2021 •

edited

Loading

agramfort commented Jun 25, 2021

robintibor commented Jun 25, 2021 •

edited

Loading

agramfort commented Jun 25, 2021 via email

cedricrommel commented Jun 25, 2021

sbbrandt commented Jun 26, 2021

[WIP] Starting point for batch-level data augmentation #254

[WIP] Starting point for batch-level data augmentation #254

Conversation

cedricrommel commented Jun 21, 2021

robintibor commented Jun 21, 2021 • edited Loading

tomMoral left a comment

Choose a reason for hiding this comment

cedricrommel commented Jun 21, 2021

tomMoral left a comment

Choose a reason for hiding this comment

tomMoral left a comment

Choose a reason for hiding this comment

robintibor commented Jun 21, 2021

robintibor commented Jun 21, 2021

robintibor commented Jun 21, 2021

agramfort commented Jun 25, 2021 via email

robintibor commented Jun 25, 2021

cedricrommel commented Jun 25, 2021

cedricrommel commented Jun 25, 2021 • edited Loading

agramfort commented Jun 25, 2021

robintibor commented Jun 25, 2021 • edited Loading

agramfort commented Jun 25, 2021 via email

cedricrommel commented Jun 25, 2021

sbbrandt commented Jun 26, 2021

robintibor commented Jun 21, 2021 •

edited

Loading

cedricrommel commented Jun 25, 2021 •

edited

Loading

robintibor commented Jun 25, 2021 •

edited

Loading