how to read my own .csv files of train/evaluation/test

[dyoflut]

Once I generated train/evaluation/test interactions and saved them as .csv files, then how can I load them into RecBole for reuse?

[Wicknight]

Hello! The data formats we support include six atomic files. For details, see doc. So you need to convert your raw data into Atomic Files format before data loading, see doc. And after the conversion, you can use your own dataset by setting parameter 'benchmark_filename' in config, see doc](https://recbole.io/docs/user_guide/config/data_settings.html) and issue #1114

[dyoflut]

Thanks a lot~I'm confident it'll work out. But beforehand I also tried another seemingly roundabout strategy:

1. train/evaluation/test .csv files → DataFrame → Dataset, one by one through pandas, Interaction() and dataset.copy().

2. copy and modify the original 'data_preparation' function as

built_datasets = [train_data_Dataset, valid_data_Dataset, test_data_Dataset]
train_sampler, valid_sampler, test_sampler = create_samplers(config, dataset, built_datasets)
train_data = get_dataloader(config, 'train')(config, train_data_Dataset, train_sampler, shuffle=True)
valid_data = get_dataloader(config, 'evaluation')(config, valid_data_Dataset, valid_sampler, shuffle=False)
test_data = get_dataloader(config, 'evaluation')(config, test_data_Dataset, test_sampler, shuffle=False)
dataloaders = (train_data, valid_data, test_data)

3. Finally save such dataloaders as .pth files.

I think the above process works. But it meanwhile generates .pth files slightly different from the original .csv files about negative item_ids, because these ids have been recreated through '*_sampler' of the above process.

Could u plz help me check the process? Thanks a lot~

[Wicknight]

Hello! I think there should be no problem with your handling. Could you please provide more information about the difference about negative item_ids?