-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathdataset.py
269 lines (238 loc) · 11.7 KB
/
dataset.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
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
import numpy as np
import pandas as pd
import pickle
import random
import torch
import math
class Dataset:
def __init__(self, ds_name, args):
self.args = args
self.name = ds_name
self.batch_size = args.batch_size
self.dir = "datasets/" + self.name + "/"
self.ent2id = self.read_pickle('ent2id')
self.instype_all = self.read_pickle('ent2classes')
self.rel2id = self.read_pickle('rel2id')
self.class2id, self.class2id2ent2id, self.subclassof2id = self.read_pickle('class2id'), self.read_pickle('class2ents'), self.read_pickle('subclassof2id')
self.r2id2dom2id, self.r2id2range2id = self.read_pickle('rel2dom'), self.read_pickle('rel2range')
self.r2id2metadom2id, self.r2id2metarange2id = self.read_pickle('rel2metadom'), self.read_pickle('rel2metarange')
self.setting = args.setting
self.sem = args.sem
if self.args.model == 'ConvE' and (self.sem=='schema' or self.sem=='both'):
inv_r2id2dom2id = {}
for k,v in self.r2id2dom2id.items():
try:
inv_r2id2dom2id[k + max(self.rel2id.values()) + 1] = self.r2id2range2id[k]
except:
pass
inv_r2id2range2id = {}
for k,v in self.r2id2range2id.items():
try:
inv_r2id2range2id[k + max(self.rel2id.values()) + 1] = self.r2id2dom2id[k]
except:
pass
self.r2id2dom2id.update(inv_r2id2dom2id)
self.r2id2range2id.update(inv_r2id2range2id)
if 'FB15K' in self.name:
inv_r2id2metadom2id = {}
for k,v in self.r2id2metadom2id.items():
try:
inv_r2id2metadom2id[k + max(self.rel2id.values()) + 1] = self.r2id2metarange2id[k]
except:
pass
inv_r2id2metarange2id = {}
for k,v in self.r2id2metarange2id.items():
try:
inv_r2id2metarange2id[k + max(self.rel2id.values()) + 1] = self.r2id2metadom2id[k]
except:
pass
self.r2id2metadom2id.update(inv_r2id2metadom2id)
self.r2id2metarange2id.update(inv_r2id2metarange2id)
self.r2hs = self.read_pickle('heads2id')
if self.args.model == 'ConvE':
self.r2ts_origin = self.read_pickle('tails2id')
self.r2ts = self.r2ts_origin.copy()
for rel, tails in self.r2ts_origin.items():
self.r2ts[rel+len(self.rel2id)] = self.r2hs[rel]
else:
self.r2ts = self.read_pickle('tails2id')
self.data = {}
if self.args.model == 'ConvE':
self.data["pd_train"] = pd.read_csv(self.dir + "train2id_inv.txt", sep='\t', header=None, names=['h','r','t'])
self.data["pd_valid"] = pd.read_csv(self.dir + "valid2id_inv.txt", sep='\t', header=None, names=['h','r','t'])
self.data["pd_test"] = pd.read_csv(self.dir + "test2id_inv.txt", sep='\t', header=None, names=['h','r','t'])
self.data["train"] = self.data["pd_train"].to_numpy()
self.data["valid"] = self.data["pd_valid"].to_numpy()
self.data["test"] = self.data["pd_test"].to_numpy()
self.data["df"] = pd.concat([self.data["pd_train"],self.data["pd_valid"],self.data["pd_test"]]).to_numpy()
self.data["df_lst"] = self.data["df"].tolist()
self.data["dft"] = tuple(map(tuple, self.data["df"]))
else:
self.data["pd_train"] = pd.read_csv(self.dir + "train2id.txt", sep='\t', header=None, names=['h','r','t'])
self.data["pd_valid"] = pd.read_csv(self.dir + "valid2id.txt", sep='\t', header=None, names=['h','r','t'])
self.data["pd_test"] = pd.read_csv(self.dir + "test2id.txt", sep='\t', header=None, names=['h','r','t'])
self.data["train"] = self.data["pd_train"].to_numpy()
self.data["valid"] = self.data["pd_valid"].to_numpy()
self.data["test"] = self.data["pd_test"].to_numpy()
self.data["df"] = pd.concat([self.data["pd_train"],self.data["pd_valid"],self.data["pd_test"]]).to_numpy()
self.data["df_lst"] = self.data["df"].tolist()
self.data["dft"] = tuple(map(tuple, self.data["df"]))
self.neg_ratio = args.neg_ratio
self.neg_sampler = args.neg_sampler
self.all_rels = list(self.rel2id.values())
self.all_ents = list(self.ent2id.values())
self.batch_index = 0
self.id2ent = {v:k for k,v in self.ent2id.items()}
self.id2rel = {v:k for k,v in self.rel2id.items()}
if self.name not in ['Codex-S', 'Codex-M', 'WN18RR']:
self.id2class = {v:k for k,v in self.class2id.items()}
if 'YAGO' in self.name:
self.dist_classes = self.read_pickle('dist_classes')
self.dist_classes2id = self.read_pickle('dist_classes2id')
def inv_rel(self, dataset2id):
for sub, rel, obj in dataset2id:
rel_inv = rel + max(self.rel2id.values()) + 1
dataset2id = np.append(dataset2id, np.array([[obj, rel_inv, sub]]), axis=0)
return dataset2id
def read_pickle(self, file):
try:
with open(self.dir + file + ".pkl", 'rb') as f:
pckl = pickle.load(f)
return pckl
except:
pass
def num_ent(self):
return len(self.ent2id)
def num_rel(self):
if self.args.model == 'ConvE':
return len(self.rel2id)*2
return len(self.rel2id)
def num_batch(self):
return int(math.ceil(float(len(self.data["train"])) / self.batch_size))
def construct_adj(self):
edge_index, edge_type = [], []
for sub, rel, obj in self.data['train']:
edge_index.append((sub, obj))
edge_type.append(rel)
# Adding inverse edges
for sub, rel, obj in self.data['train']:
edge_index.append((obj, sub))
edge_type.append(rel + self.num_rel())
edge_index = torch.LongTensor(edge_index).to(self.device).t()
edge_type = torch.LongTensor(edge_type). to(self.device)
return edge_index, edge_type
def next_pos_batch(self, batch_size):
if self.batch_index + batch_size < len(self.data["train"]):
batch = self.data["train"][self.batch_index: self.batch_index+batch_size]
self.batch_index += batch_size
else:
batch = self.data["train"][self.batch_index:]
self.batch_index = 0
return batch
def random_negative_sampling(self, pos_batch, neg_ratio, side='all'):
if self.neg_ratio==1:
neg_ratio=2
neg_batch = np.repeat(np.copy(pos_batch), neg_ratio, axis=0)
M = neg_batch.shape[0]
corr = np.random.randint(self.num_ent() - 1, size=M)
if side=='all':
e_idxs = np.random.choice([0, 2], size=M)
neg_batch[np.arange(M), e_idxs] = corr
elif side=='tail':
neg_batch[np.arange(M), 2] = corr
elif side=='head':
neg_batch[np.arange(M), 0] = corr
return self.filtering(pos_batch, neg_batch)
def tc(self, r2class):
try:
return np.random.choice(self.class2id2ent2id[r2class], size=1)
except KeyError:
return np.random.choice(self.num_ent() - 1, size=1)
def anti_tc(self, r2class):
try:
other_cl = random.choice(list(self.class2id2ent2id.keys() - [r2class]))
return random.choice(self.class2id2ent2id[other_cl])
except KeyError:
return np.random.choice(self.num_ent() - 1, size=1)
def tcns(self, pos_batch, neg_ratio, side='all'):
neg_batch = np.repeat(np.copy(pos_batch), neg_ratio, axis=0)
M = neg_batch.shape[0]
if side=='all':
e_idxs = np.random.choice([0, 2], size=M)
for i in range(len(neg_batch)):
h,r,t = neg_batch[i, 0], neg_batch[i, 1], neg_batch[i, 2]
if e_idxs[i] == 0:
neg_batch[i, 0] = self.tc(self.r2id2dom2id[r])
else:
neg_batch[i, 2] = self.tc(self.r2id2range2id[r])
elif side=='tail':
for i in range(len(neg_batch)):
neg_batch[i, 2] = self.tc(self.r2id2range2id[r])
return self.filtering(pos_batch, neg_batch)
def meta_tcns(self, pos_batch, neg_ratio, side='all'):
neg_batch = np.repeat(np.copy(pos_batch), neg_ratio, axis=0)
M = neg_batch.shape[0]
if side=='all':
e_idxs = np.random.choice([0, 2], size=M)
for i in range(len(neg_batch)):
h,r,t = neg_batch[i, 0], neg_batch[i, 1], neg_batch[i, 2]
if e_idxs[i] == 0:
neg_batch[i, 0] = self.tc(self.r2id2metadom2id[r])
else:
neg_batch[i, 2] = self.tc(self.r2id2metarange2id[r])
elif side=='tail':
for i in range(len(neg_batch)):
neg_batch[i, 2] = self.tc(self.r2id2metarange2id[r])
return self.filtering(pos_batch, neg_batch)
def anti_tcns(self, pos_batch, neg_ratio, side='all'):
neg_batch = np.repeat(np.copy(pos_batch), neg_ratio, axis=0)
M = neg_batch.shape[0]
if side=='all':
e_idxs = np.random.choice([0, 2], size=M)
for i in range(len(neg_batch)):
h,r,t = neg_batch[i, 0], neg_batch[i, 1], neg_batch[i, 2]
if e_idxs[i] == 0:
neg_batch[i, 0] = self.anti_tc(self.r2id2dom2id[r])
else:
neg_batch[i, 2] = self.anti_tc(self.r2id2range2id[r])
elif side=='tail':
for i in range(len(neg_batch)):
neg_batch[i, 2] = self.anti_tc(self.r2id2range2id[r])
return self.filtering(pos_batch, neg_batch)
def socher(self, pos_batch, neg_ratio, side='all'):
neg_batch = np.repeat(np.copy(pos_batch), neg_ratio, axis=0)
M = neg_batch.shape[0]
if side=='all':
e_idxs = np.random.choice([0, 2], size=M)
for i in range(len(neg_batch)):
h,r,t = neg_batch[i, 0], neg_batch[i, 1], neg_batch[i, 2]
if e_idxs[i] == 0:
neg_batch[i, 0] = random.choice(list(set(self.H[r]) - set([h])))
else:
neg_batch[i, 2] = random.choice(list(set(self.T[r]) - set([t])))
elif side=='tail':
for i in range(len(neg_batch)):
r,t = neg_batch[i, 1], neg_batch[i, 2]
neg_batch[i, 2] = random.choice(list(self.set(T[r]) - set([t])))
return self.filtering(pos_batch, neg_batch)
def next_batch(self, batch_size, neg_ratio, neg_sampler, device):
pos_batch = self.next_pos_batch(batch_size)
if neg_sampler == 'rns':
neg_batch = self.random_negative_sampling(pos_batch, neg_ratio)
elif neg_sampler == 'tcns':
neg_batch = self.type_constrained_sampling(pos_batch, neg_ratio)
batch = np.append(pos_batch, neg_batch, axis=0)
batch = torch.tensor(batch)
return batch
def was_last_batch(self):
return (self.batch_index == 0)
def filtering(self, pos_batch, neg_batch):
nbt = tuple(map(tuple, neg_batch))
both = ((set(nbt).intersection((self.data['dft']))))
dupl = [nbt.index(x) for x in both]
nbt_rem = tuple(map(tuple, neg_batch[dupl]))
nbt_ok = tuple(set(nbt) - set(nbt_rem))
neg_batch = np.asarray(nbt_ok[:(len(nbt_ok)-(len(nbt_ok)%pos_batch.shape[0]))])
if self.neg_ratio == 1:
neg_batch = neg_batch[:pos_batch.shape[0]]
return neg_batch