search

dselivanov / rsparse

Fast and accurate machine learning on sparse matrices - matrix factorizations, regression, classification, top-N recommendations.
https://www.slideshare.net/DmitriySelivanov/matrix-factorizations-for-recommender-systems
170 stars 31 forks source link

PureSVD #8

Closed dselivanov closed 6 years ago

dselivanov commented 7 years ago

From Performance of Recommender Algorithms on Top-N Recommendation Tasks it seems like vanilla truncated SVD (which they call "PureSVD") is a good baseline for OC-CF settings. So worth to implement separate class for that.

dselivanov commented 6 years ago

I've added fast implementation of soft-svd - truncated SVD with regularization on nuclear norm (adopted from softImpute package). Now need to expose R6 class with cross-validation with warm-starts.