-
Notifications
You must be signed in to change notification settings - Fork 1.9k
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.
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
matrix factorization #22
Conversation
…eventually blows up
…ll inputs (learning rate decayed before examples were processed)
Merge branch 'master' of git://github.com/JohnLangford/vowpal_wabbit Conflicts: Makefile parse_regressor.cc sparse_dense.cc vw.cc
Do we really want vw.R? And what does it do exactly? Also, it would be cool if you could add an example to the wike about how to use MF. |
vw.R just facilitates parameter sweeps for cross-validation, e.g.: args <- list("b"=18, results <- vw_param_sweep("cat movielens_train.dat", we can probably do w/o it, it's quite rough around the edges and probably not appropriate for general use at the moment anyway. as for an example on the wiki, we can use the small movielens data set (100K ratings): wget 'http://www.grouplens.org/system/files/ml-data.tar__0.gz' just let me know how to proceed. |
Ok, I'm incorporating it. I removed vw.R, but perhaps you can add it into the example. Can you add an example on the wiki, and then make a post to the mailing list? We should probably either bring LDA back in or make MF run as a separate executable for consistency. |
wiki example is here: https://github.com/JohnLangford/vowpal_wabbit/wiki/Matrix-factorization-example have you accepted the pull request and pushed the updated code to your github repo? (i don't see gd_mf.{h,cc} in the repo.) just let me know and i'll send out a message to the mailing list. |
Oops, right. I actually did everything local on my laptop. It's checked in -John On Tue, Jul 19, 2011 at 2:31 PM, jhofman <
|
…of epsilon greedy. Also fixed a bug in ActionSet::Get that was returning incorrect Action due to 0-based index assumption. This is fixed by adding methods to MWTAction class to make id 1-based or 0-based. This is not ideal but a good temporary solution to keep track of related code.
added simple matrix factorization (with regularization) in gd_mf.cc.
need to add mf to test suite.