Recommenders: Review SVD results

Created on 22 Nov 2018  路  3Comments  路  Source: microsoft/recommenders

https://github.com/Microsoft/Recommenders/blob/staging/notebooks/02_modeling/surprise_svd_deep_dive.ipynb

bug high priority
Source
picture miguelgfierro

Most helpful comment

There is a bug in the Surprise SVD notebook. See cells 4 and 5.
In cell 5, train is assigned to the full data Movielens100K data set. So, if you do train.n_ratings you get 100K. Then SVD is trained on the whole data set and hence the results are too good, because the test set is a subset of the training set.

picture anargyri on 22 Nov 2018
👍3

All 3 comments

There is a bug in the Surprise SVD notebook. See cells 4 and 5.
In cell 5, train is assigned to the full data Movielens100K data set. So, if you do train.n_ratings you get 100K. Then SVD is trained on the whole data set and hence the results are too good, because the test set is a subset of the training set.

anargyri picture anargyri on 22 Nov 2018
👍3

Great catch Andreas!

miguelgfierro picture miguelgfierro on 22 Nov 2018

Resolved in https://github.com/Microsoft/Recommenders/pull/284

anargyri picture anargyri on 23 Nov 2018
Was this page helpful?
0 / 5 - 0 ratings

Related issues

seancaodo picture seancaodo  路  3Comments
tqtg picture tqtg  路  5Comments
miguelgfierro picture miguelgfierro  路  5Comments
miguelgfierro picture miguelgfierro  路  4Comments
yueguoguo picture yueguoguo  路  4Comments