File size: 2,106 Bytes
bf1497a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
import os
from tqdm import tqdm
import json

meta_file = '/etc/ssd1/jiangzhongtao/baai_embedding_tune/data/task_meta/casual/all.json'
with open(meta_file, 'r') as f:
    meta_info = json.load(f)


task_file = "/etc/ssd1/jiangzhongtao/baai_embedding_tune/data/all_collect/task_files.txt"
unsample_to_be_divisible_by = None
with open(task_file, 'r') as f:
    for line in f:
        file, _ = line.strip().split('\t')
        try:
            assert os.path.exists(file)
        except Exception:
            print(file)

tasktype2batchsize = {
    'classification': 2048,
    'clustering': 2048,
    'duplication': 512,
    'nli': 512,
    'retrieval': 512,
    'sts': 256,
    'super-ni': 2048,
    'unk': 2048
}


cnt = 0
type2cnt = dict()
steps = 0
with open(task_file, 'r') as f:
    for line in tqdm(f):
        file, train_size = line.strip().split('\t')
        _, task_name, _ = os.path.split(file)[-1].split('_')
        task_type = meta_info[task_name]['task_type']
        if task_type != 'retrieval':
            total_size = 0
            with open(file, 'r') as f_file:
                for line in f_file:
                    total_size += 1
        else:
            queries = set()
            with open(file, 'r') as f_file:
                for line in f_file:
                    queries.add(json.loads(line)['query'])
            total_size = len(queries)
        train_size = int(train_size)

        if train_size > 0:
            if total_size < train_size:
                raise ValueError(file)
        else:
            train_size = total_size
        batch_size = tasktype2batchsize[task_type]
        train_size = train_size // batch_size * batch_size
        steps += train_size // batch_size
        # print(file, train_size)
        cnt += train_size
        type2cnt[task_type] = type2cnt.get(task_type, 0) + train_size

print('steps', steps)
print(type2cnt)
# {
#     'retrieval': 2489655, 
#     'sts': 140819, 
#     'nli': 324951, 
#     'super-ni': 180000, 
#     'classification': 677510, 
#     'clustering': 452224, 
#     'duplication': 31409
# }
print(cnt) # 3825445