About the dataset preprocessing part. I think the index of items and users should start at 1 not 0

[Furyton]

def densify_index(self, df):
    print('Densifying index')
    umap = {u: i for i, u in enumerate(set(df['uid']))}
    smap = {s: i for i, s in enumerate(set(df['sid']))}
    df['uid'] = df['uid'].map(umap)
    df['sid'] = df['sid'].map(smap)
    return df, umap, smap

Is the 'umap' and 'smap' both beginning with 0? If so, then it is the same index with the label when the token was masked.

[yuanninesuns]

I think so, too

[thomalm]

Yes I agree since he set tokens = tokens = [0] * mask_len + tokens. 0 should be reserved for padding (1, num_items) for the item indexes and num_items + 1 for the CLOZE_MASK_TOKEN here: https://github.com/jaywonchung/BERT4Rec-VAE-Pytorch/blob/master/dataloaders/bert.py#L14

[jaywonchung]

I am very sorry for the delayed response. I'm recently getting little free time to actively maintain this repository. Similar issues have arisen quite frequently, and a PR is welcome. Thanks.

[federicosiciliano]

Hi everyone, I am also using the code, and I too ran into this issue. It's been a while, but it doesn't seem to have been solved. I think the problem originates here: https://github.com/jaywonchung/BERT4Rec-VAE-Pytorch/blob/master/datasets/base.py#L134 You should put i+1 (in general, I would suggest doing the same with UserIDs). The problem is that, as far as I know, this variation affects most of the code. Does anyone have a simple solution or is it necessary to check all the code?

[redhated]

I agree, I think it's enough to generate smap from index 1, there is no need to modify umap

smap = {s: i+1 for i, s in enumerate(set(df['sid']))}