Open FuriouslyCurious opened 8 years ago
@FuriouslyCurious as for original LibFM - definitely not. It is quite useless in practice, since it officially doesn't support storing/loading trained model. It's a pity, since original implementation is quite reliable compared to other implementations (for any real-world application this is a show-stopper).
LibFFM requires passing additional information (fields), which can't be provided in scikit-learn format. Without it, LibFFM is not different from LibFM.
I wanted to compare ALS implementations of LibFM and FastFM. If FastFM ALS turns out to be good enough, we'll integrate it. (BTW, help in this comparison is welcome)
Sure I am happy to help Alex - I will do a comparo on my data and share some results here next month.
libFM and libFFM are used a lot in Kaggle.
I will buy the whole team a round of beers if you can include a wrapper for either one of those in REP.
Thanks guys!