ZhaoyueCheng / SASRec.pytorch

PyTorch(1.6+) implementation of https://github.com/kang205/SASRec

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

update, with few lines of manually initialization code added, it converges as fast as tf version. BTW, I strongly recommend checking issues for the repo from time to time for knowing new updates and details :)

update: a pretrained model added, pls run the command as below to test its performance(current perf still not as good as paper's reported results after trained more epochs, maybe due to leaky causual attention weights issue got fixed by using PyTorch 1.6's MultiHeadAttention, pls help identifying the root cause if you are interested):

python main.py --device=cuda --dataset=ml-1m --train_dir=default --state_dict_path='ml-1m_default/SASRec.epoch=601.lr=0.001.layer=2.head=1.hidden=50.maxlen=200.pth' --inference_only=true --maxlen=200

modified based on paper author's tensorflow implementation, switching to PyTorch(v1.6) for simplicity, executable by:

python main.py --dataset=ml-1m --train_dir=default --maxlen=200 --dropout_rate=0.2 --device=cuda

pls check paper author's repo for detailed intro and more complete README, and here's paper bib FYI :)

  title={Self-attentive sequential recommendation},
  author={Kang, Wang-Cheng and McAuley, Julian},
  booktitle={2018 IEEE International Conference on Data Mining (ICDM)},
ezoic increase your site revenue


PyTorch(1.6+) implementation of https://github.com/kang205/SASRec


Language:Python 100.0%