Final prediction is ensembled by only 5 models, which are DCNv2, HardShareSEnet, PLE, PLEMLPSEnet, PLESEnet respectively.
run analyze/process.ipynb, data/split_sample.ipynb and feature_engineer/weekday.ipynb to generate *.csv
fix url in RecStudio/recstudio/data/config/*.yaml
download DCNv2 and save *.ckpt in saved/DCNv2/
other checkpoints have already been kept in saved/
run predict.sh to generate the final prediction.
Please refer to RecStudio, which is a unified, highly-modularized and recommendation-efficient recommendation library based on PyTorch. We welcome more effient models from you.