Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Pseudo labeling #138

Open
wants to merge 71 commits into
base: main
Choose a base branch
from
Open

Pseudo labeling #138

wants to merge 71 commits into from

Conversation

benjamin-cates
Copy link
Contributor

No description provided.

@benjamin-cates benjamin-cates changed the title Pseudo labelling within epochs Pseudo labelling between epochs Aug 6, 2023
@benjamin-cates
Copy link
Contributor Author

Something in this branch is causing logger statements to be printed twice and I'm not sure why. It doesn't happen on main 😭

@benjamin-cates
Copy link
Contributor Author

benjamin-cates commented Aug 6, 2023 •
edited
Loading

The most recent commit has a fix for the problem. Calling config.Config() causes the logger to duplicate 😭 . Should prolly store that in the spaghetti code tupperware for later.

@benjamin-cates
Copy link
Contributor Author

benjamin-cates commented Aug 6, 2023 •
edited
Loading

FIXED: Note for later: breaks if class list is None

benjamin-cates and others added 21 commits August 6, 2023 09:46
@benjamin-cates
Fix borked wandb logging
f69cebc
@benjamin-cates
Move pseudolabel init from run_raw to main
bf4c816
@benjamin-cates
Fix class_list bug
982d279
@benjamin-cates
Only read audio files in pseudo directory
8da2ddd
@benjamin-cates
Fix get_test_map checkpoint
567f555
@mbazzani
Merge remote-tracking branch 'origin/main' into nutella
2a125f8
@mbazzani
Merge branch 'nutella' into pseudo
74d721d
@benjamin-cates
Pylint fixes 🥰
0cc5508
@mbazzani
Change pseudolabeling to finetuning
97f0204
@mbazzani
Merge remote-tracking branch 'refs/remotes/origin/pseudo' into pseudo
9b7b1ec
@benjamin-cates
Notela project
ae6b73d
@mbazzani
Add epsilon to mixup
d08c782
@mbazzani
Bug fixes
0c8654f
@mbazzani
Merge remote-tracking branch 'refs/remotes/origin/pseudo' into pseudo
c59c68a
@mbazzani
Add pseudo labeling parameters to config
1f6ad35
@mbazzani
Linter happy!
6041cbf
@mbazzani
Add wandb logging to pseudo
e1b87b8
@mbazzani
Enforce consistent naming
e87470a
@mbazzani
Remove old comments
830a76a
@mbazzani
Fix pyright
f10e5d7
@benjamin-cates
Fix pseudo label progress bar
9af622f
@benjamin-cates
Copy link
Contributor Author

Ok one issue I have with this is that if you're just running on pseudo labels it loads the training dataset multiple times even though it's not being used. We need a command line argument or something to specify the kind of fine tuning like "pseudo", "nutella" or "none"

benjamin-cates and others added 2 commits August 10, 2023 14:41
@benjamin-cates
Fix clips/s metric
7036419
@benjamin-cates
Debias pseudo labeling (#145)
b018e22 
* Initial debias testing

* Fix labeled sample DL crashing

* Add infer map

* Add PCEN

* Oops

* Oops2

* Merge fixes (oops)

* Move debiasPL to own file

* Improve resampling of train set on debiasPL

* Remove logging from pseudolabel.py

* Pylint 🥰
@benjamin-cates benjamin-cates changed the title Pseudo labelling between epochs Pseudo labeling Aug 14, 2023
benjamin-cates and others added 22 commits August 14, 2023 11:10
@benjamin-cates
Fix debias clips/sec bug
a772e0f
@mbazzani
Initial implementation
03d177c
@mbazzani
project_suffix option to wandb_init
1bc82e3
@mbazzani
Bug fixes
0f3f076
@mbazzani
Add eca_nfnet_l0 features
5ec63e6
@mbazzani
Remove superfluous print statements
68a5f27
@mbazzani
Add error message for missing pseudolabels
7c51c0e
@mbazzani
Linting fixes
26742ee
@mbazzani
Fix bug with label existence edge cases
6787a4c
@mbazzani
Fix type issues
23d2f6d
@mbazzani
Resolve merge conflicts
1291a1c
@mbazzani
Move pseudolabel loading into nutella
ef554cc
@mbazzani
Switch to using scipy distance
2115c23
@mbazzani
Set multiprocessing method to spawn
03de22e
@mbazzani
Restore usage of all files for pseudolabeling
bd3c304
@mbazzani
Add options to choose between flattening and pooling features
86ef35f
@mbazzani
Add Notela parameters
bb31add
@mbazzani
Fix max method call
9bca3ea
@mbazzani
Various bug fixes
b705aae
@mbazzani
Pylint happy 🥰
73d5c10
@mbazzani
Stop pyright being a silly little goose
c34d927
@mbazzani
Remove excess print statement
9d1ba3f
@Sean1572
Copy link
Collaborator

I'm concerned that this maybe too big, or may conflict with notela changes as well

This was linked to issues Aug 22, 2023
Dataloader does not raise error if files exist but loading fails #144
Open
Add sparse storage to NOTELA #140
Open
Allow for f-strings in logging? #127
Open
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@sprestrelski sprestrelski sprestrelski left review comments

@mbazzani mbazzani mbazzani left review comments

At least 1 approving review is required to merge this pull request.

Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
4 participants
@benjamin-cates @Sean1572 @sprestrelski @mbazzani