search

tensorflow / recommenders

TensorFlow Recommenders is a library for building recommender system models using TensorFlow.
Apache License 2.0
1.83k stars 274 forks source link

Ranking model predicting same value for all #377

Open harry7171 opened 3 years ago

harry7171 commented 3 years ago

Hi,

I am training a Ranking model with some basic numeric features and the target is either 0 or some value between 0-1. So the model is training and just in 1 epoch it comes to a minimum loss and doesnot decrease more. The main issue is observed when i predict with test data , i get all the predictions as 0. tried some changes in Layers but all leads to almost 99% values as 0 Where as same architecture when i trained using sequential model . its learning and getting atleast some values in prediction . Not sure why with tfrs Ranking its some issue .

maciejkula commented 3 years ago

There is not a whole lot that I can say based on this information.

There could be something wrong with your data, or your loss, or your model architecture, or your training hyperparameters.