search

NVIDIA-Merlin / NVTabular

NVTabular is a feature engineering and preprocessing library for tabular data designed to quickly and easily manipulate terabyte scale datasets used to train deep learning based recommender systems.
Apache License 2.0
1.04k stars 143 forks source link

[BUG] Criteo TensorFlow examples does not use numerical features #1315

Closed bschifferer closed 2 years ago

bschifferer commented 2 years ago

Describe the bug Numeric Inputs in TensorFlow example for scaling criteo are not used in the next layers

EvenOldridge commented 2 years ago

@bschifferer is this still a valid issue? Can you provide more details?

rnyak commented 2 years ago

@EvenOldridge @viswa-nvidia what Benedikt refers here that in the scaling-criteo 03-Training-with-TF.ipynb notebook, we define numerical layers for numerical features, but they are never fed in the model architecture, therefore it is not used. The issue is still valid, yes.

bschifferer commented 2 years ago

@rvk007 created a PR to update the TensorFlow model with Merlin Models. I need to review it, but I think we will fix the issue soon

bschifferer commented 2 years ago

We updated the example to use merlin models (in Merlin/Merlin repo)