fm_im.md

Factorlization Machine Implementation Survey

Most of all implement using C++ or Cython due to the speed

Though, I just need a education version

Fastway to pick up all of the facotrlization machines

search factorlization machine in github

check:

1. knowing how to update the interaction part
2. support spaerse input

doesn't need in this phase:

1. cypthon-based, C, C++ based
2. complex wrapper
3. ranking supported

candidate

1. from scratch

factorization machine for prediction keras star 95+

FluRS: A Python library for streaming recommendation algorithms star 95+

tensorflow implementation, 700+ star

lightfm - for ranking and classification

2. api

PyFM (most straight forward) 800+ stars using python and cython

fastFM 800+ using python and C, support binary classification, regression, ranking

Pick :

1. tffm(for understanding)
2. fastFM - for explaination(can access w and v), and other task (ranking)
3. keras-fm - for study purpose, eligant Embedding layer to make problem very easy.
4. lightfm - for stduy ranking optimization

Discussion

Although now is 2020. The implementation if FM is still few. That's why we have fastFM as a paper. There is still no package like scikit-learn in recommandation system.

Yes, there is a pacakge called surprise , but it's too basic.