[🐛BUG] Benchmark preset error with sequential dataset

Hi, I am trying to use a sequential benchmark dataset, but I get following error:

Traceback (most recent call last):
  File "C:/Users/apein/Documents/projects/humrecsys/seq_rec/run.py", line 68, in <module>
    main(args=args)
  File "C:/Users/apein/Documents/projects/humrecsys/seq_rec/run.py", line 42, in main
    dataset = create_dataset(config)
  File "C:\Users\apein\.virtualenvs\humrecsys-Aky0O99T\lib\site-packages\recbole\data\utils.py", line 41, in create_dataset
    return SequentialDataset(config)
  File "C:\Users\apein\.virtualenvs\humrecsys-Aky0O99T\lib\site-packages\recbole\data\dataset\sequential_dataset.py", line 38, in __init__
    self._benchmark_presets()
  File "C:\Users\apein\.virtualenvs\humrecsys-Aky0O99T\lib\site-packages\recbole\data\dataset\sequential_dataset.py", line 146, in _benchmark_presets
    self.inter_feat[self.item_list_length_field] = self.inter_feat[self.item_id_list_field].agg(len)
AttributeError: 'SequentialDataset' object has no attribute 'item_id_list_field'

This error only appears while using a sequential model. General models work fine with the benchmark dataset.

[hyp1231]

Hi, I understand that you want to ensure that session id doesn’t overlap between train set, validation set and test set. Currently, we suggest to pre-split the dataset into xxx.train.inter, xxx.valid.inter and xxx.test.inter and load it via benchmark_filename (one example, session_based_rec_example, you can also download the processed diginetica-session dataset [download] to check it detailly.)

The 1. group by user and 2. split pipeline is mainly designed for general collaborative filtering methods or sequential recommendation methods (especially leave one out evaluation). For session-based recommendation, we recommend following the way in the above example.

[pintonos]

Sorry I have to reopen this. I was wondering about the data splitting: According to the documentation the group_by evaluation parameter, if set to user, groups the dataset by the user (or session when used sequentially) before splitting into train, valid and test set. This would imply for me, that e.g. a session id from the valid set is not part of the train set. However, this is not the case! In the split_by_ratio function of Dataset each session gets equally split into the 3 sets. Is this intended? Whats the point of grouping then? How can I ensure to split the data session-wise only?