Skip to content

xmc-aalto/missing-labels-tmlr

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

1 Commit
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Unbiased Loss Functions for Multilabel Classification with Missing Labels

https://openreview.net/forum?id=hMq1hUhLqp

  • eval contains the code for generating results evaluting recall-at-k on fully artificial data.
  • train contains the code for generating the semi-artificial dataset, where train/keep_top.py is used to remove all but the top-k labels from an XMC dataset, and train/runner.py runs a sweep over different regularizations for training with unbiased/upper-bound losses. The calculations of the losses themselves are defined in train/losses.py. Note that missing labels are induced inside the runner script, instead of during data-preprocessing, because we want to calculate both original loss function on original data, and ubiased/upper-bound loss on missing-label data.
  • yahoo contains the code for the yahoo-music-r3 experiment. It follows the same structure as train, but uses yahoo/prep.py to generate the training data from the original music rating data.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors