This repository contains PyTorch Implementation of ICDE 2022 paper:
Memorize, factorize, or be naive: Learning optimal feature interaction methods for CTR Prediction.
You can find our paper here.
Following the steps below to run our codes:
pip install -r requirements.txt
Please Download the Criteo and Avazu datasets.
- For Criteo dataset, copy the
train.txt
file underdatasets/Criteo
and rename it tofull.txt
. - For Avazu dataset, copy the
train.csv
file underdatasets/Avazu
and rename it tofull.csv
.
- For Criteo dataset, run
python preprocess/criteo.py
- For Avazu dataset, run
python preprocess/avazu.py
- For Criteo dataset, run
python learn/CriteoSearch.py
- For Avazu dataset, run
python learn/AvazuSearch.py
- For Criteo dataset, run
python learn/CriteoTrain.py --load XXX --model DNN_cart --alpha_mode 0
- For Avazu dataset, run
python learn/AvazuTrain.py --load XXX --model DNN_cart --alpha_mode 0
Here XXX
indicates the logs dictionary generated during the search stage.
- For Criteo dataset, run
python learn/CriteoTrain.py --model YYY
- For Avazu dataset, run
python learn/AvazuTrain.py --model YYY
Here YYY
could be {LR, FM, FNN, IPNN, DeepFM, PIN, Poly2, DNN_cart}
@inproceedings{OptInter,
author = {Fuyuan Lyu and
Xing Tang and
Huifeng Guo and
Ruiming Tang and
Xiuqiang He and
Rui Zhang and
Xue Liu},
title = {Memorize, Factorize, or be Naive: Learning Optimal Feature Interaction
Methods for {CTR} Prediction},
booktitle = {38th {IEEE} International Conference on Data Engineering, {ICDE} 2022},
pages = {1450--1462},
address = {Kuala Lumpur, Malaysia},
publisher = {{IEEE}},
year = {2022},
url = {https://doi.org/10.1109/ICDE53745.2022.00113},
doi = {10.1109/ICDE53745.2022.00113},
timestamp = {Sun, 30 Jul 2023 12:39:38 +0200},
biburl = {https://dblp.org/rec/conf/icde/LyuTGTHZL22.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}