I've finished a prototype and the test and I think it'll be nice if we can do a code review. Given 2 categorical variables(e.g. user-song, user-artist, etc), the estimator should estimate their relation(e.g. rating(regression), like/dislike(classification)). I'll need to do hyper parameter search to get more idea about its performance. We never did code review before but I like it a lot so please don't hesitate to leave some comments on the files changed. The tensorflow and skflow parts are pretty dense and I'll go over it when we meet. Thanks!
The branch name is wrong and please disregard it - I haven't test the estimator on any movie dataset. However, as we might test the estimator on more than one datasets, I think it's better to put the dataset specific descriptions(e.g. data categories, embedding sizes, hidden units formation) into separate specs files, instead of creating a separate test for each dataset.
I've finished a prototype and the test and I think it'll be nice if we can do a code review. Given 2 categorical variables(e.g. user-song, user-artist, etc), the estimator should estimate their relation(e.g. rating(regression), like/dislike(classification)). I'll need to do hyper parameter search to get more idea about its performance. We never did code review before but I like it a lot so please don't hesitate to leave some comments on the files changed. The tensorflow and skflow parts are pretty dense and I'll go over it when we meet. Thanks!
The branch name is wrong and please disregard it - I haven't test the estimator on any movie dataset. However, as we might test the estimator on more than one datasets, I think it's better to put the dataset specific descriptions(e.g. data categories, embedding sizes, hidden units formation) into separate specs files, instead of creating a separate test for each dataset.