您好，我尝试运行CL4SRec模型，想要使用pop100来评估模型，按照RecBole中的文档在seq.yaml修改eval_setting

eval_setting: TO_LS, pop100

但得到了一处错误

Traceback (most recent call last):
  File "/code/DuoRec/run_seq.py", line 15, in <module>
    run_recbole(model=args.model, dataset=args.dataset, config_file_list=config_file_list)
  File "/code/DuoRec/recbole/quick_start/quick_start.py", line 48, in run_recbole
    train_data, valid_data, test_data = data_preparation(config, dataset)
  File "/code/DuoRec/recbole/data/utils.py", line 170, in data_preparation
    valid_data = dataloader(**valid_kwargs)
TypeError: __init__() got an unexpected keyword argument 'phase'

我看到~data/dataloader/sequential_dataloader.py/SequentialNegSampleDataLoader 并没有phase 作为参数，官方repo关于dataloader的地方差别较大，可能您的部分代码使用的版本较早，出现了兼容性问题？

应该只是一个小小的问题。

In the augmentation phase of CL4SREC, which includes three tasks ,i.e. item_crop、item_mask、item_reorder, the code of item_mask and item_reorder unexpectedly changes the source seq, resulting in the augmented sequence being identical to the original sequence. The main reason is that the slicing operation is performed on the tensor instead of the list.

I use the source code provided here. Download the ml-1m dataset, just run sh duorec.sh I can't get the result as the paper.

More interesting, the baseline use BPR loss as the main objective which takes 1 positive and 1 negative item each time while this work introduces CE loss which takes 1 positive and all of the rest as negative items. It is unfair.

When I change the main objective to BPR loss which is the common used loss for former papers, the performance is even lower than CL4SRec.

In the paper you report 0.2946 recall@10 for ML-1M, however when I run duorec.sh command, I'm getting result around 0.25-026. Could you please share the run configuration which I can use to reproduce paper results?

Hi, thanks for your great work! But I have a question. In section 5.1 of the paper, you said "Following baselines [19, 41, 51, 58], the widely used Amazon dataset is chosen in our experiments with three sub-categories." But the 19[ Self-Attentive Sequential Recommendation, CIKM2018] and 58[ Sˆ3-Rec: Self-Supervised Learning for Sequential Recommendation with Mutual Information Maximization, CIKM 2020] have different datasets. The former uses the Amazon ratings only datasets, while the latter uses the 5-cores datasets. And model performance varies widely across datasets. So what kind of datasets are you using? Looking forward your reply！

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.