search

facebookresearch / FiD

Fusion-in-Decoder
Other
551 stars 107 forks source link

Question about data loading #12

Open PlusRoss opened 3 years ago

PlusRoss commented 3 years ago

Hi,

In the src/data.py file: https://github.com/facebookresearch/FiD/blob/baf533c3f7a26c1cac624ee9252ce5ccf344a935/src/data.py#L139 May I ask why you set the score value as 1/(k+1), where k is the index of data? Shouldn't k be the index of ctx passage?

Thanks.

cryingjin commented 2 years ago

hi, I think so, too. So I fixed it like this. 

    for i,c in enumerate(example['ctxs']):
        if not 'score' in c:
            c['score'] = 1.0 / (i + 1)

If there's a better way, please let me know. Thanks.