search

takuti / flurs

:ocean: FluRS: A Python library for streaming recommendation algorithms
https://flurs.readthedocs.io/
MIT License
108 stars 17 forks source link

regularization for FM #1

Closed Sandy4321 closed 6 years ago

Sandy4321 commented 6 years ago

data matrix usually spars, so may you clarify if you use regularization Thanks

Sandy4321 commented 6 years ago

video 30:07 regularization to use with SGD for FM
https://www.youtube.com/watch?v=LV4JLTIZxNU

takuti commented 6 years ago

Hi, I cannot understand what you expect. FM already uses L2 regularization here. Do you have any suggestions related to the code?

Sandy4321 commented 6 years ago

yes it is my question, so you do have L2 regularization <- very good. May you clarify , 1 do you have L1 regularization?
2 do you have regularization not only for regression coefficients but for satisfaction parameters V as well? 3 May you share some code example for simple recommendation system from 0 to end for example movies recommendations?

takuti commented 6 years ago
  1. No. The implementation simply follows formulation in the FM paper which only utilizes L2 regularization.
  2. Yes, as the name of variable l2_reg_V suggests.
  3. Thank you for your suggestion. I'll prepare it as soon as I can.