search

jfkirk / tensorrec

A TensorFlow recommendation algorithm and framework in Python.
Apache License 2.0
1.28k stars 222 forks source link

what's the different between LightFM and tensorrec? #92

Closed letotefrank closed 6 years ago

letotefrank commented 6 years ago

hi,jfkirk！Can you answer this question in detail? I have used LightFM to build a recommendation system. Is it necessary to use this(tensorrec)?

jfkirk commented 6 years ago

Hey @letotefrank

LightFM and TensorRec are similar packages. I used LightFM quite a bit before writing TensorRec.

LightFM is a great system for learning linear embeddings from user/item metadata. The project is mature, and the package is quite effective at problems where a linear embedding is sufficient.

TensorRec gives you the option to customize your recommender systems embedding (aka representation) functions. This allows you to apply deep nets, non-linear transforms, or anything else you can implement in TensorFlow as embedding functions. By default, TensorRec is very similar to LightFM.

If you don't have a particular need to use non-linear representation functions, custom prediction functions, or custom loss functions, then you're probably best served using LightFM for now.

Hope this helps!