Function comments_collate_fn

code/run.py:252–293

(batch)

    }

def comments_collate_fn(batch): 
    num_comments = 23 

    comments_like = [item['comments_like'] for item in batch]
    comments_inputid = [item['comments_inputid'] for item in batch]
    comments_mask = [item['comments_mask'] for item in batch]

    comments_inputid_resorted = [] 
    comments_mask_resorted = []
    comments_like_resorted = []

    for idx in range(len(comments_like)):
        comments_like_one = comments_like[idx]
        comments_inputid_one = comments_inputid[idx]
        comments_mask_one = comments_mask[idx]
        if comments_like_one.shape != torch.Size([0]):
            comments_inputid_one, comments_mask_one, comments_like_one = (list(t) for t in zip(*sorted(zip(comments_inputid_one, comments_mask_one, comments_like_one), key=lambda s: s[2], reverse=True)))
        comments_inputid_resorted.append(comments_inputid_one)
        comments_mask_resorted.append(comments_mask_one)
        comments_like_resorted.append(comments_like_one)
    
    comments_inputid = pad_sequence(num_comments,comments_inputid_resorted,250)
    comments_mask = pad_sequence(num_comments,comments_mask_resorted,250)
    comments_like=[]
    for idx in range(len(comments_like_resorted)):
        comments_like_resorted_one = comments_like_resorted[idx]
        if len(comments_like_resorted_one)>=num_comments:
            comments_like.append(torch.tensor(comments_like_resorted_one[:num_comments]))
        else:
            if isinstance(comments_like_resorted_one, list):
                comments_like.append(torch.tensor(comments_like_resorted_one+[0]*(num_comments-len(comments_like_resorted_one))))
            else:
                comments_like.append(torch.tensor(comments_like_resorted_one.tolist()+[0]*(num_comments-len(comments_like_resorted_one))))

    label = [item['label'] for item in batch]

    return {
        'label': torch.stack(label),
        'comments_inputid': comments_inputid,
        'comments_mask': comments_mask,
        'comments_like': torch.stack(comments_like),
    }

def title_w2v_collate_fn(batch):
    length_title = 128