atomu2014 / product-nets-distributed Goto Github PK

We are trying the author's PIN code on Criteo and Avazu. We are able to reproduce the AUC score of 78.72% on Avazu. But we can only achieve an AUC score of 80.18% on Criteo. However, if we use a different embedding size for each field, we are able to get an AUC score of 80.21% on Criteo. But this is not the setting claimed the paper. Could the authors clarify on this issue?

Could you share the reqirements.txt of the Python environments? That can provide information like Python and TensorFlow version.

I do some experiments use this repo following the paper, but the auc and loss have a gap between my experiments and the paper, maybe my paramaters are wrong, so can you provide the command lines the paper used. Thanks

' Product-based Neural Networks for User Response Prediction over Multi-field Categorical Data' (TOIS'17)

In section 5.1.1Datasets of this paper, there says "We randomly split the public dataset into training and test sets at 4:1, and remove categories appearing less than 20 times to reduce dimensionality.",
but when i preprocessing the raw avazu dataset by my self, i found that if #categories=6*10^5 in avazu dataset, the threshold need to be 10 , not 20.
when i use a threshold 20, #categories< 4*10^5

Is it an incorrect threshold number in section 5.1.1 ?