Open paulpig opened 2 years ago
I also have some questions about the experiment results of SASRec. I just copied sasrec.py from recbole and put sasrec.py into duorec. Then I run SASRec based on this code repo and got around 0.0326 for NDCG@5, which is significantly better than the NDCG@5 result 0.0236 of SASRec reported in the paper.
10 Mar 11:25 INFO Loading model structure and parameters from /home/leiwang/project_2021/RecSys/2022/xai_cl/modification/DuoRec/log/SASRec/Amazon_Beauty/bs256-lmd0.1-sem0.1-us_x-Mar-10-2022_11-20-05-lr0.001-l20-tau1-dot-DPh0.5-DPa0.5/model.pth
Evaluate : 100%|█████████████████████████████| 88/88 [00:00<00:00, 282.81it/s]
10 Mar 11:25 INFO best valid : {'recall@5': 0.0691, 'recall@10': 0.1027, 'recall@20': 0.1442, 'ndcg@5': 0.0409, 'ndcg@10': 0.0517, 'ndcg@20': 0.0622}
10 Mar 11:25 INFO test result: {'recall@5': 0.0555, 'recall@10': 0.083, 'recall@20': 0.1191, 'ndcg@5': 0.0326, 'ndcg@10': 0.0415, 'ndcg@20': 0.0506}
Same here
same here
same here
Hello, I find that the number of negative items is set to 1 on SASRec. However, DuoRec set the number to all items. Besides, I experiment on SASRec when the number of negative items is all items, and I find that the performance is comparable with DuoRec. Look forward to your reply.