/IBNet

Optimizing Feature Interaction via Information Bottleneck for CTR Prediction

Primary LanguagePython

IBNet: Optimizing Feature Interaction via Information Bottleneck for CTR Prediction

Model Overview

image

Requirements

python==3.8.18
torch==1.12.0+cu116
fuxictr

Others

pip install -r requirement.txt

Experiment results

image

Datasets

Get the datasets from https://github.com/openbenchmark/BARS/tree/main/datasets

Hyperparameter settings and logs

Get the result from https://github.com/Acer888/IBNet/tree/master/IBNet_torch/checkpoints

Get Mice(PBMish) activation function from https://github.com/Acer888/IBNet/blob/master/fuxictr/pytorch/layers/activations.py

Acknowledgement

This implementation is based on FuxiCTR and BARS. Thanks for their sharing and contribution.
BARS: https://github.com/openbenchmark
FuxiCTR: https://github.com/xue-pai/FuxiCTR