-
Notifications
You must be signed in to change notification settings - Fork 5
/
Copy pathbenchmark_sampling.py
152 lines (134 loc) · 6.83 KB
/
benchmark_sampling.py
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
from collections import defaultdict
import os.path as osp
import argparse
import os
import json
from shutil import rmtree
from multiprocessing import Pool
from tqdm import tqdm
import pandas as pd
from fol.foq_v2 import (DeMorgan_replacement, concate_iu_chains, parse_formula,
to_d, to_D, decompose_D, copy_query)
from formula_generation import convert_to_dnf
from utils.util import load_data_with_indexing
parser = argparse.ArgumentParser()
parser.add_argument("--benchmark_name", type=str, default="benchmark")
parser.add_argument("--input_formula_file", type=str, default="outputs/test_generated_formula_anchor_node=3.csv")
parser.add_argument("--sample_size", default=5, type=int)
parser.add_argument("--knowledge_graph", action="append")
parser.add_argument("--ncpus", type=int, default=1)
parser.add_argument("--num_samples", type=int, default=5000)
parser.add_argument("--meaningful_difference_setting", type=str, default='mixed')
def normal_forms_transformation(query):
result = {}
# proj, rproj = load_graph()
# query.backward_sample()
result["original"] = query
result["DeMorgan"] = DeMorgan_replacement(copy_query(result["original"], True))
result['DeMorgan+MultiI'] = concate_iu_chains(copy_query(result["DeMorgan"], True))
result["DNF"] = convert_to_dnf(copy_query(result["original"], True))
result["diff"] = to_d(copy_query(result["original"], True))
result["DNF+diff"] = to_d(copy_query(result["DNF"], True))
result["DNF+MultiIU"] = concate_iu_chains(copy_query(result["DNF"], True))
result['DNF+MultiIU'].sort_sub()
result["DNF+MultiIUD"] = to_D(copy_query(result["DNF+MultiIU"], True))
result["DNF+MultiIUd"] = decompose_D(copy_query(result["DNF+MultiIUD"], True))
return result
def sample_by_row(row, easy_proj, easy_rproj, hard_proj, meaningful_difference: bool = False):
query_instance = parse_formula(row.original)
easy_answers = query_instance.backward_sample(easy_proj, easy_rproj, meaningful_difference=meaningful_difference)
full_answers = query_instance.deterministic_query(hard_proj)
hard_answers = full_answers.difference(easy_answers)
results = normal_forms_transformation(query_instance)
for k in results:
assert results[k].formula == row[k]
_full_answer = results[k].deterministic_query(hard_proj)
assert _full_answer == full_answers
_easy_answer = results[k].deterministic_query(easy_proj)
assert _easy_answer == easy_answers
return list(easy_answers), list(hard_answers), results
def sample_by_row_final(row, easy_proj, hard_proj, hard_rproj, meaningful_difference_setting: str = 'mixed'):
while True:
query_instance = parse_formula(row.original)
if meaningful_difference_setting == 'mixed':
formula = query_instance.formula
meaningful_difference = ('d' in formula or 'D' in formula or 'n' in formula)
elif meaningful_difference_setting == 'fixed_True':
meaningful_difference = True
elif meaningful_difference_setting == 'fixed_False':
meaningful_difference = False
else:
assert False, 'Invalid setting!'
full_answers = query_instance.backward_sample(hard_proj, hard_rproj,
meaningful_difference=meaningful_difference)
assert full_answers == query_instance.deterministic_query(hard_proj)
easy_answers = query_instance.deterministic_query(easy_proj)
hard_answers = full_answers.difference(easy_answers)
results = normal_forms_transformation(query_instance)
if 0 < len(hard_answers) <= 100:
break
# for key in results:
# parse_formula(row[key]).additive_ground(json.loads(results[key].dumps))
return list(easy_answers), list(hard_answers), results
if __name__ == "__main__":
args = parser.parse_args()
print(args)
df = pd.read_csv(args.input_formula_file)
beta_data_folders = {"FB15k-237": "data/FB15k-237-betae",
"FB15k": "data/FB15k-betae",
"NELL": "data/NELL-betae"}
for kg in args.knowledge_graph:
data_path = beta_data_folders[kg]
ent2id, rel2id, \
proj_train, reverse_train, \
proj_valid, reverse_valid, \
proj_test, reverse_test = load_data_with_indexing(data_path)
kg_name = osp.basename(data_path).replace("-betae", "")
out_folder = osp.join("data", args.benchmark_name, kg_name)
os.makedirs(out_folder, exist_ok=True)
for i, row in tqdm(df.iterrows(), total=len(df)):
fid = row.formula_id
data = defaultdict(list)
if args.ncpus > 1:
def sampler_func(i):
row_data = {}
easy_answers, hard_answers, results = sample_by_row_final(
row, proj_valid, proj_test, reverse_test,
meaningful_difference_setting=args.meaningful_difference_setting)
row_data['easy_answers'] = easy_answers
row_data['hard_answers'] = hard_answers
for k in results:
row_data[k] = results[k].dumps
return row_data
produced_size = 0
sample_size = args.num_samples
generated = set()
while produced_size < sample_size:
with Pool(args.ncpus) as p:
gets = p.map(sampler_func, list(range(sample_size - produced_size)))
for row_data in gets:
original = row_data['original']
if original in generated:
continue
else:
produced_size += 1
generated.add(original)
for k in row_data:
data[k].append(row_data[k])
else:
generated = set()
sampled_query = 0
while sampled_query < args.num_samples:
easy_answers, hard_answers, results = sample_by_row_final(
row, proj_valid, proj_test, reverse_test,
meaningful_difference_setting=args.meaningful_difference_setting)
if results['original'].dumps in generated:
continue
else:
generated.add(results['original'].dumps)
sampled_query += 1
data['easy_answers'].append(easy_answers)
data['hard_answers'].append(hard_answers)
for k in results:
data[k].append(results[k].dumps)
pd.DataFrame(data).to_csv(osp.join(out_folder, f"data-{fid}.csv"), index=False)