sadighian / recommendation-gym Goto Github PK

Hello, thank you for such a great repo

I've trained the model and accuracy only achieve 30% ?

Could I ask you some questions:

1. How much is your model's accuracy on MovieLens dataset?
2. How to run the inference for recommendation ? For example, I want to run the inference for one user_id ? How can I do that ? Thank you

Thank you for your contribution to the community! Personally, I found your neat implementation interesting and helpful!
By the way, is this current implementation following the Markov Property?
For instance, in MDPs, there is a state transition showing P(s' | s, a), right? And in RecSys setting, the state space consists of static and dynamic properties of users such as Static(User profile, eg., gender/age etc) and Dynamic(User's past activity, eg., click/non-click or view/purchase etc) so that we can correlate the states based on the action taken by an agent.

Looking at this implementation, it literally calculates the average of the ratings which I think it isn't dynamic over time and irrelevant to the occurrence of actions. could you help understand this part...?

Code

https://github.com/sadighian/recommendation-gym/blob/master/gym_recommendation/envs/reco_env.py#L207-L208