SKA-Bench/process_dataset.py at main · zjukg/SKA-Bench · GitHub

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
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
import json
import os
import random
import argparse
### python process_dataset.py --type Table+Text --sequence random --scale 1k
parser = argparse.ArgumentParser(add_help=False)
parser.add_argument('--type', default="KG", type=str)
parser.add_argument('--sequence', default="random", type=str)
parser.add_argument('--seed', default=42, type=int)
parser.add_argument('--scale', default="1k", type=str, choices=["1k", "4k", "12k", "24k", "64k"])
parser.add_argument('--negative_rejection', default=None, type=str)

args = parser.parse_args()
os.makedirs('./dataset', exist_ok=True)
if args.negative_rejection:
    output_dir = f'./dataset/{args.type}_{args.sequence}_{args.seed}_{args.scale}_negative_rejection.json'
else:
    output_dir = f'./dataset/{args.type}_{args.sequence}_{args.seed}_{args.scale}.json'
random.seed(args.seed)

if args.type=="KG":
    with open('./KG.json')as f:
        samples = json.load(f)
    scale2num = {"1k": 30, "4k": 150, "12k": 630, "24k": 1270, "64k": 2550}
    output = []
    for sample in samples:
        id = sample['id']
        question = sample['question']
        answer = sample['answer']
        data = sample["positive_triples"]
        noisy_data = sample["noisy_triples"][:scale2num[args.scale]]
        noisy_data = [item for item in noisy_data if item not in data] # filter positive triple
        if args.sequence == "random":
            data = data + noisy_data
            if args.negative_rejection:
                data = noisy_data
            random.shuffle(data)
        elif args.sequence == 'prefix':
            data = data + noisy_data
        elif args.sequence == 'suffix':
            data = noisy_data + data
        output.append({
            "id": id,
            "question": question,
            "answer": answer,
            "data": data
        })
    with open(f'{output_dir}', "w") as f:
        json.dump(output, f, indent=4, ensure_ascii=False)
elif args.type=="Table":
    with open('./Table.json')as f:
        samples = json.load(f)
    scale2num = {"1k": 10, "4k": 110, "12k": 470, "24k": 950}
    output = []
    for sample in samples:
        id = sample['id']
        question = sample['question']
        answer = sample['answer']
        header = sample["header"]
        data = sample["positive_rows"]
        if args.negative_rejection:
            data = []
        noisy_data1 = [line for line in sample["original_rows"] if line not in data]
        try:noisy_data2 = sample["noisy_rows"][:scale2num[args.scale]]
        except:noisy_data2 = sample["noisy_rows"]
        noisy_data = noisy_data1 + noisy_data2
        if args.sequence == "original":
            data = sample["original_rows"] + noisy_data2
            if args.negative_rejection:
                data = noisy_data2
        elif args.sequence == "random":
            data = data + noisy_data
            random.shuffle(data)
        elif args.sequence == 'prefix':
            data = data + noisy_data
        elif args.sequence == 'suffix':
            data = noisy_data + data
        output.append({
            "id": id,
            "question": question,
            "answer": answer,
            "data": [header]+data
        })
    with open(f'{output_dir}', "w") as f:
        json.dump(output, f, indent=4, ensure_ascii=False)
elif args.type=="Table+Text":
    with open('./Table+Text.json')as f:
        samples = json.load(f)
    scale2num_row = {"4k": 9, "12k": 59, "24k": 149}
    scale2num_text = {"4k": 14, "12k": 119, "24k": 369}
    output = []
    for sample in samples:
        id = sample['id']
        question = sample['question']
        answer = sample['answer']
        Table_header = sample["header"]
        Table_data = sample["positive_rows"]
        Table_noisy_data = sample["original_rows"]+sample["noisy_rows"][:scale2num_row[args.scale]]
        Table_noisy_data = [item for item in Table_noisy_data if item not in Table_data] # filter positive triple
        Text_data = sample["positive_texts"]
        Text_noisy_data = (sample["original_texts"]+sample["noisy_texts"])[:scale2num_text[args.scale]]
        Text_noisy_data = [item for item in Text_noisy_data if item not in Text_data] # filter positive text
        if args.sequence == "random":
            Table_data = Table_data + Table_noisy_data
            random.shuffle(Table_data)
            Text_data = Text_data + Text_noisy_data
            random.shuffle(Text_data)
        elif args.sequence == 'prefix':
            Table_data = Table_data + Table_noisy_data
            Text_data = Text_data + Text_noisy_data
        elif args.sequence == 'suffix':
            Table_data =  Table_noisy_data + Table_data
            Text_data = Text_noisy_data + Text_data
        elif args.sequence == 'original':
            Table_data = sample["original_rows"] + sample["noisy_rows"][:scale2num_row[args.scale]]
            if args.negative_rejection:
                Table_data = sample["noisy_rows"][:scale2num_row[args.scale]]
            Text_data = Text_data + Text_noisy_data
            if args.negative_rejection:
                Text_data = Text_noisy_data
        output.append({
            "id": id,
            "question": question,
            "answer": answer,
            "Table_data": [Table_header]+Table_data,
            "Text_data": Text_data,
        })
    with open(f'{output_dir}', "w") as f:
        json.dump(output, f, indent=4, ensure_ascii=False)
elif args.type=="KG+Text":
    with open('./KG+Text.json')as f:
        samples = json.load(f)
    scale2num_triple = {"4k": 9, "12k": 39, "24k": 159, "64k": 319}
    scale2num_text = {"4k": 1, "12k": 5, "24k": 18, "64k": 35}
    output = []
    for sample in samples:
        id = sample['id']
        question = sample['question']
        answer = sample['answer']
        KG_data = sample["positive_triples"]
        KG_noisy_data = sample["noisy_triples"][:scale2num_triple[args.scale]]
        KG_noisy_data = [item for item in KG_noisy_data if item not in KG_data] # filter positive triple
        Text_data = sample["positive_texts"]
        Text_noisy_data = sample["noisy_texts"][:scale2num_text[args.scale]]
        Text_noisy_data = [item for item in Text_noisy_data if item not in Text_data] # filter positive text
        if args.sequence == "random":
            KG_data = KG_data + KG_noisy_data
            if args.negative_rejection:
                KG_data = KG_noisy_data
            random.shuffle(KG_data)
            Text_data = Text_data + Text_noisy_data
            if args.negative_rejection:
                Text_data = Text_noisy_data
            random.shuffle(Text_data)
        elif args.sequence == 'prefix':
            KG_data = KG_data + KG_noisy_data
            Text_data = Text_data + Text_noisy_data
        elif args.sequence == 'suffix':
            KG_data =  KG_noisy_data + KG_data
            Text_data = Text_noisy_data + Text_data
        output.append({
            "id": id,
            "question": question,
            "answer": answer,
            "KG_data": KG_data,
            "Text_data": Text_data,
        })
    with open(f'{output_dir}', "w") as f:
        json.dump(output, f, indent=4, ensure_ascii=False)