Skip to content

Latest commit

 

History

4 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 

Repository files navigation

gl-recommender-sgd-issue

Illustration of a problem I've encountered with graphlab's factorization recommender

When working with rating data from Goodreads, I noticed some strange behavior in the graphlab recommender I was using. Some more tests showed that this was due to the learning rate decay schedule, which does not seem to be fully under the user's control, and was too fast for good convergence. The demo notebook in this repo reproduces this behavior on toy data, and shows that a recommender using vanilla SGD can do much better on this data (as it could on my Goodreads data).

About

Illustration of a problem I've encountered with graphlab's factorization recommender

Resources

Stars

0 stars

Watchers

1 watching

Forks

Releases

Packages

Contributors

Languages