-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathtrainer.py
421 lines (344 loc) · 14.1 KB
/
trainer.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
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
import json
import logging
import numpy as np
import torch
import torch.nn.functional as F
import random
from collections import defaultdict
from collections import deque
from torch import optim
from torch.autograd import Variable
from tqdm import tqdm
from args import read_options
from tensorboardX import SummaryWriter
from scipy.sparse import csc_matrix
import torch.nn as nn
from utils import *
from net import *
import time
import os
from networkx.algorithms.link_analysis import pagerank
import operator
import math
import sys
class Trainer(object):
def __init__(self, arg):
super(Trainer, self).__init__()
for k, v in vars(arg).items(): setattr(self, k, v)
#self.device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
self.device = torch.device("cpu")
if self.random_embed:
use_pretrain = False
else:
use_pretrain = True
self.use_pretrain = use_pretrain
self.load_embed()
self.T_GRUA = T_GRUA(self.kernel_num, self.embed_dim, self.hidden_dim, self.h_hrt_bg,self.ent2id, self.id2ent, self.id2rel, self.batch_size,self.edge_matrix,self.edge_nums,self.topk, self.rel_emb, self.ent_emb, self.device)
model_params = list(self.T_GRUA.parameters())
self.parameters = filter(lambda p: p.requires_grad, model_params)
'''
model_dict = self.T_GRUA.state_dict()
for k, v in model_dict.items():
print("model_dict:",k)
'''
self.optim = optim.Adam(self.parameters, lr=self.lr, weight_decay=self.weight_decay)
def loadtxt(self,path):
name2ids = {}
with open(path) as file:
for i, line in enumerate(file):
line = line.strip().split()
name,idx = line
name2ids[name] = int(idx)
return name2ids
def load_embedding(self,path):
def load_from_file(path,i):
embeds = []
with open(path) as file:
for line in file:
line = line.strip().split()
embeds.append(list(map(float, line)))
i = i+1
return embeds,i
i = 0
relation_embeds,i = load_from_file(path,i)
relation_embeds = torch.tensor(relation_embeds)
return relation_embeds
def load_embed(self):
rel_bg = json.load(open(self.dataset + '/relation2ids'))
ent_all = json.load(open(self.dataset+'/ent2ids'))
train_tasks = json.load(open(self.dataset+'/train_tasks.json'))
test_tasks = json.load(open(self.dataset+'/test_tasks.json'))
dev_tasks = json.load(open(self.dataset+'/dev_tasks.json'))
ent_embed = np.loadtxt(self.dataset + '/entity2vec.' + self.embed_model)
rel_embed = np.loadtxt(self.dataset + '/relation2vec.' + self.embed_model)
if self.embed_model=='ComplEx':
ent_mean = np.mean(ent_embed, axis=1, keepdims=True)
ent_std = np.std(ent_embed, axis=1, keepdims=True)
rel_mean = np.mean(rel_embed, axis=1, keepdims=True)
rel_std = np.std(rel_embed, axis=1, keepdims=True)
eps = 1e-3
ent_embed = (ent_embed - ent_mean) / (ent_std + eps)
rel_embed = (rel_embed - rel_mean) / (rel_std + eps)
self.rel2candidates = json.load(open(self.dataset + '/rel2candidates.json'))
self.e1rel_e2 = defaultdict(list)
self.e1rel_e2 = json.load(open(self.dataset + '/e1rel_e2.json'))
train_rel = train_tasks.keys()
test_rel = test_tasks.keys()
dev_rel = list(dev_tasks.keys())
bg_rel = rel_bg.keys()
rel2id = {}
ent2id = {}
rel_embedding = []
ent_embedding = []
i = 0
for key in rel_bg.keys():
rel2id[key] = i
i=i+1
rel_embedding.append(list(rel_embed[rel_bg[key],:]))
for rel in list(train_rel)+list(test_rel)+list(dev_rel):
rel2id[rel] = i
i=i+1
j = 0
for key in ent_all.keys():
ent2id[key] = j
j = j + 1
ent_embedding.append(list(ent_embed[ent_all[key],:]))
rel_embedding = torch.tensor(rel_embedding)
ent_embedding = torch.tensor(ent_embedding)
self.bg_rel_id_list = []
self.train_rel_id_list = []
self.test_rel_id_list = []
self.dev_rel_id_list = []
for i in range(len(list(train_rel))):
self.train_rel_id_list.append(rel2id[list(train_rel)[i]])
for i1 in range(len(list(test_rel))):
self.test_rel_id_list.append(rel2id[list(test_rel)[i1]])
for i2 in range(len(list(dev_rel))):
self.dev_rel_id_list.append(rel2id[list(dev_rel)[i2]])
for i3 in range(len(list(bg_rel))):
self.bg_rel_id_list.append(rel2id[list(bg_rel)[i3]])
facts_data = []
pg_facts_data = []
bg_data = []
with open(self.dataset+'/path_graph') as file:
for line in file:
fact = line.strip().split()
pg_facts_data.append([ent2id[fact[0]],rel2id[fact[1]],ent2id[fact[2]]])
pg_facts_data.append([ent2id[fact[2]],rel2id[fact[1]+'_inv'],ent2id[fact[0]]])
bg_data.append([ent2id[fact[0]],ent2id[fact[2]],rel2id[fact[1]]])
bg_data.append([ent2id[fact[2]],ent2id[fact[0]],rel2id[fact[1]+'_inv']])
file.close()
id2rel = {v: k for k, v in rel2id.items()}
id2ent = {v: k for k, v in ent2id.items()}
'''
# generate pagerank.txt
self.kg = KG(pg_facts_data, entity_num=len(ent2id), relation_num=len(rel_bg))
graph = networkx.DiGraph(self.kg.to_networkx())
print("Begin to compute pagerank")
self.pagerank = pagerank(graph)
self.pagerank = [self.pagerank[entity] for entity in range(len(self.pagerank))]
print("Begin to save pagerank")
with open(os.path.join(self.dataset, "pagerank.txt"), "w") as output:
for value in self.pagerank:
output.write("{}\n".format(value))
print("Complete save pagerank")
#'''
with open(os.path.join(self.dataset, 'pagerank.txt')) as file:
self.pagerank = list(map(lambda x: float(x.strip()), file.readlines()))
self.edge_data = [[] for _ in range(len(ent2id) + 1)]
for fact in bg_data:
e1,e2,rel = fact
self.edge_data[e1].append((e1, e2, rel))
for head in range(len(self.edge_data)):
self.edge_data[head].sort(key=lambda x: self.pagerank[x[1]], reverse=True)
self.edge_data[head] = self.edge_data[head][:self.neighbor_limit]
self.edge_nums = torch.tensor(list(map(len, self.edge_data)), dtype=torch.long)
edge_entities = [list(map(lambda x: x[1], edges)) for edges in self.edge_data]
edge_relations = [list(map(lambda x: x[2], edges)) for edges in self.edge_data]
edge_entities = list2tensor(edge_entities, padding_idx=len(ent2id), dtype=torch.int, device=self.device)
edge_relations = list2tensor(edge_relations, padding_idx=len(rel2id), dtype=torch.int,device=self.device)
self.edge_matrix = torch.stack((edge_entities, edge_relations), dim=2)
train_trip_id = reltri2tri(train_tasks, rel2id, ent2id)
test_trip_id = reltri2tri(test_tasks,rel2id,ent2id)
dev_trip_id = reltri2tri(dev_tasks,rel2id,ent2id)
self.rel_emb = nn.Embedding(len(rel2id.keys())+1, self.embed_dim)
self.rel_emb.weight.data[:len(rel_bg)] = rel_embedding
self.rel_emb.weight.data[-1] = torch.zeros(1,100)
self.rel_emb = self.rel_emb.to(self.device)
self.ent_emb = nn.Embedding(len(ent2id.keys())+1,self.embed_dim)
self.ent_emb.weight.data[:len(ent2id)] = ent_embedding
self.ent_emb.weight.data[-1] = torch.zeros(1,100)
self.ent_emb = self.ent_emb.to(self.device)
self.rel2id = rel2id
self.ent2id = ent2id
self.id2rel = id2rel
self.id2ent = id2ent
self.train_tasks = train_tasks
self.bg_data = bg_data
self.train_trip_id = train_trip_id
self.test_trip_id = test_trip_id
self.dev_trip_id = dev_trip_id
self.pg_facts_data = pg_facts_data
self.h_hrt_bg = h2hrt(self.pg_facts_data)
self.rel_test_trip = trip2rel2tripid(self.test_trip_id,self.test_rel_id_list)
self.rel_dev_trip = trip2rel2tripid(self.dev_trip_id,self.dev_rel_id_list)
self.rel2candidates = json.load(open(self.dataset + '/rel2candidates.json'))
self.rel_bg_trip = trip2rel2tripid(self.pg_facts_data, self.bg_rel_id_list)
rel_emb_n = self.rel_emb.weight.data[:-1]
rel_emb_bro = rel_emb_n.unsqueeze(0).repeat(rel_emb_n.size()[0],1,1)
cos_rel_all = torch.sigmoid(torch.cosine_similarity(rel_emb_bro,rel_emb_n.unsqueeze(1),dim=-1))
pad = nn.ZeroPad2d(padding=(0, 1, 0, 1)) # padding
self.cos_rel_all = pad(cos_rel_all)
path_dict_str = json.load(open(self.dataset+'/train_valid_test_pair2paths_name.json'))
self.trian_test_path, self.train_test_path_id = path_read(path_dict_str, self.rel2id, self.ent2id)
self.test2relkind_dict, self.test2relkind = test_relkind(self.rel_test_trip, self.train_test_path_id)
self.set_rel_sim = set_rel_sim_count(self.test2relkind)
def save(self, path = None):
if not path:
path = self.save_path
state_all = {'T_GRUA':self.T_GRUA.state_dict()}
torch.save(state_all, path)
def load(self):
checkpoint = torch.load(self.save_path+'_mrr_best')
self.T_GRUA.load_state_dict(checkpoint['T_GRUA'])
def train(self):
logging.info('START TRAINING...')
batch_num = 0
best_mrr = 0.0
for data in train_generate(self.few, self.dataset, self.batch_size, self.train_tasks,self.ent2id,self.rel2id, self.id2ent, self.id2rel, self.e1rel_e2, self.rel2candidates):
batch_num = batch_num+1
sys.stdout.write("\r{0}".format(str(batch_num)+'/'+str(self.max_batches)))
sys.stdout.flush()
self.optim.zero_grad()
support_pair, query_pair, one_tomany_train, candidates_id = data
support_pair_name = []
for i in range(len(support_pair)):
support_pair_name.append([self.id2ent[support_pair[i][0]], self.id2ent[support_pair[i][1]]])
query_pair_name = []
for i in range(len(query_pair)):
query_pair_name.append([self.id2ent[query_pair[i][0]], self.id2ent[query_pair[i][1]]])
support_rel = rel_submit(support_pair,self.train_test_path_id)
support_path = path_submit(support_pair, self.train_test_path_id)
query_path = path_submit(query_pair, self.train_test_path_id)
query_head = [pair[0] for pair in query_pair]
query_tail = [pair[1] for pair in query_pair]
query_head = torch.tensor(query_head)
query_tail = torch.tensor(query_tail)
loss = self.T_GRUA(support_pair, support_rel, support_path, query_head, query_tail, one_tomany_train, self.cos_rel_all, True, candidates_id)
if loss==0:
continue
loss.backward()
self.optim.step()
with torch.no_grad():
if batch_num % 50==0 :
hit1, hit5, hit10, mrr = self.eval(self.mode)
self.T_GRUA.train()
if mrr > best_mrr:
self.save(self.save_path + '_mrr_best')
best_mrr = mrr
if batch_num > self.max_batches:
self.test_(self.mode)
break
def test_(self, mode='test'):
logging.info('Checkpoint loaded')
self.load()
self.eval(mode, True)
'''
#support_pair_eval, query_paireval = self.find_sqs(values,lay_all,self.set_rel_sim[key])
def find_sqs(self,key):
pair_key = self.test2relkind_dict[key]
score = self.set_rel_sim[key]
index,score = zip(*sorted(enumerate(score), key=operator.itemgetter(1),reverse=True))
support_pair = []
query_paireval = []
for i in range(len(pair_key)):
if i < self.few:
support_pair.append([pair_key[index[i]][0],pair_key[index[i]][1]])
else:
query_paireval.append([pair_key[index[i]][0],pair_key[index[i]][1]])
return support_pair, query_paireval
'''
def find_sq(self,key):
trip_key = self.rel_test_trip[key]
support_pair = []
query_paireval = []
for i in range(len(trip_key)):
if i < self.few:
support_pair.append([trip_key[i][0],trip_key[i][2]])
else:
query_paireval.append([trip_key[i][0],trip_key[i][2]])
return support_pair, query_paireval
def eval(self, mode = 'test', get_result = False):
self.T_GRUA.eval()
hit1_sum = []
hit5_sum = []
hit10_sum = []
mrr_sum = []
if mode == 'test':
rel_trip = self.rel_test_trip
else:
rel_trip = self.rel_dev_trip
for key, values in rel_trip.items():
# logging.info('key:{}'.format(key))
# logging.info('values_len:{}'.format(len(values)))
if len(values)<2:
break
str_rel = self.id2rel[key]
candidate_ent = self.rel2candidates[str_rel]
candidate_ent_id = []
for i in range(len(candidate_ent)):
candidate_ent_id.append(self.ent2id[candidate_ent[i]])
support_pair_eval, query_paireval = self.find_sq(key)
support_name = []
for i in range(len(support_pair_eval)):
support_name.append([self.id2ent[support_pair_eval[i][0]],self.id2ent[support_pair_eval[i][1]]])
support_rel = rel_submit(support_pair_eval,self.train_test_path_id)
support_path = path_submit(support_pair_eval, self.train_test_path_id)
hit1, hit5, hit10, mrr = self.eval_score(key, candidate_ent_id, support_pair_eval, query_paireval, support_rel, support_path)
hit1_sum = hit1_sum + hit1
hit5_sum = hit5_sum + hit5
hit10_sum = hit10_sum + hit10
mrr_sum = mrr_sum + mrr
if get_result:
logging.critical('All------Hits1:{:.3f}, Hits5:{:.3f}, Hits10:{:.3f}, MRR:{:.3f}'.format(np.mean(hit1_sum), np.mean(hit5_sum), np.mean(hit10_sum), np.mean(mrr_sum)))
self.T_GRUA.train()
return np.mean(hit1_sum), np.mean(hit5_sum), np.mean(hit10_sum), np.mean(mrr_sum)
def eval_score(self, key, candidate_ent_id, support_pair, eval_pair, support_rel, support_path):
head = []
right_tail = []
support_pair_name = []
for i in range(len(support_pair)):
support_pair_name.append([self.id2ent[support_pair[i][0]], self.id2ent[support_pair[i][1]]])
i= 0
for i in range(len(eval_pair)):
head.append(eval_pair[i][0])
right_tail.append(eval_pair[i][1])
one2many_list_all = []
for i in range(len(head)):
one2many = self.e1rel_e2[self.id2ent[int(head[i])]+self.id2rel[int(key)]]
one2many2id = [self.ent2id[_] for _ in one2many]
one2many2id.remove(right_tail[i])
one2many_list_all.append(one2many2id)
head = torch.tensor(head)
right_tail = torch.tensor(right_tail)
num = head.size()[0]
num_count = math.ceil(num/float(self.batch_size))
hit1_all = []
hit5_all = []
hit10_all = []
mrr_all = []
for i in range(num_count):
if i == num_count-1:
head_batch = head[i*self.batch_size:]
right_tail_batch = right_tail[i*self.batch_size:]
one2many = one2many_list_all[i*self.batch_size:]
else:
head_batch = head[i*self.batch_size : (i+1)*self.batch_size]
right_tail_batch = right_tail[i*self.batch_size : (i+1)*self.batch_size]
one2many = one2many_list_all[i*self.batch_size : (i+1)*self.batch_size]
hit1,hit5,hit10,mrr = self.T_GRUA(support_pair, support_rel, support_path, head_batch, right_tail_batch, one2many, self.cos_rel_all, False, candidate_ent_id)
hit1_all = hit1_all+hit1
hit5_all = hit5_all+hit5
hit10_all = hit10_all+hit10
mrr_all = mrr_all + mrr
return hit1_all, hit5_all, hit10_all, mrr_all