-
Notifications
You must be signed in to change notification settings - Fork 3
/
run.py
394 lines (353 loc) · 17 KB
/
run.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
import argparse
import ast
import os
os.environ['CUDA_VISIBLE_DEVICES']='0'
from utils.args import ArgumentGroup,print_arguments
import logging
from reader.data_reader import read_input
from reader.data_loader import prepare_EC_info,prepare_adj_info,get_edge_labels
from model.DHGE import DHGE
import time
import math
import random
import torch
import torch.utils.data.dataset as Dataset
import torch.utils.data.dataloader as DataLoader
import copy
from itertools import cycle
from utils.evaluation import batch_evaluation,compute_metrics
torch.set_printoptions(precision=8)
logging.basicConfig(
format='%(asctime)s - %(levelname)s - %(name)s - %(message)s',
datefmt='%m/%d/%Y %H:%M:%S')
logger = logging.getLogger(__name__)
logger.setLevel(logging.INFO)
logger.info(logger.getEffectiveLevel())
parser = argparse.ArgumentParser(description='HyperKE4TI')
DHGE_g=ArgumentGroup(parser, "model", "model and checkpoint configuration.")
DHGE_g.add_arg('input', type=str, default='dataset/JW44K-6K/JW44K-6K-',help="") # db
DHGE_g.add_arg('output', type=str, default='/home/luohaoran/HyperKA-pytorch/output/results/',help="")
DHGE_g.add_arg('dim', type=int, default=256,help="")
DHGE_g.add_arg('onto_dim', type=int, default=256,help="")
DHGE_g.add_arg('ins_layer_num', type=int, default=3,help="")
DHGE_g.add_arg('onto_layer_num', type=int, default=3,help="")
DHGE_g.add_arg('neg_typing_margin', type=float, default=0.1,help="")
DHGE_g.add_arg('neg_triple_margin', type=float, default=0.2,help="")
DHGE_g.add_arg('nums_neg', type=int, default=30,help="")
DHGE_g.add_arg('mapping_neg_nums', type=int, default=30,help="")
DHGE_g.add_arg('learning_rate', type=float, default=5e-4,help="")
DHGE_g.add_arg('batch_size', type=int, default=1024,help="")
DHGE_g.add_arg('epochs', type=int, default=160,help="")
DHGE_g.add_arg('combine', type=ast.literal_eval, default=True,help="")
DHGE_g.add_arg('ent_top_k', type=list, default=[1, 3, 5, 10],help="")
DHGE_g.add_arg("use_cuda", bool, True, "If set, use GPU for training.")
DHGE_g.add_arg('ins_intermediate_size', type=int, default=512,help="")
DHGE_g.add_arg('onto_intermediate_size', type=int, default=512,help="")
DHGE_g.add_arg('num_hidden_layers', type=int, default=12,help="")
DHGE_g.add_arg('num_attention_heads', type=int, default=4,help="")
DHGE_g.add_arg('hidden_dropout_prob', type=float, default=0.1,help="")
DHGE_g.add_arg('attention_dropout_prob', type=float, default=0.1,help="")
DHGE_g.add_arg('num_edges', type=int, default=5,help="")
args = parser.parse_args(args=[])
class EDataset(Dataset.Dataset):
def __init__(self, triples1):
self.triples1=triples1
def __len__(self):
return len(self.triples1[0])
def __getitem__(self,index):
return self.triples1[0][index],self.triples1[1][index],self.triples1[2][index],self.triples1[3][index],self.triples1[4][index]
class CDataset(Dataset.Dataset):
def __init__(self, triples2):
self.triples2=triples2
def __len__(self):
return len(self.triples2[0])
def __getitem__(self,index):
return self.triples2[0][index],self.triples2[1][index],self.triples2[2][index],self.triples2[3][index],self.triples2[4][index]
class SDataset(Dataset.Dataset):
def __init__(self, cross_info,ins_info,onto_info,nums_neg,device):
self.device=device
self.nums_neg=nums_neg
self.seed_sup_ent1 = cross_info['train_cross'][0]
self.seed_sup_ent2 = cross_info['train_cross'][1]
self.ins_neg_sample= list(range(2+ins_info['rel_num'],ins_info['node_num']))
self.onto_neg_sample= list(range(2+onto_info['rel_num'],onto_info['node_num']))
self.seed_links = list()
self.seed_link_set=set()
for i in range(len(self.seed_sup_ent1)):
seed_link=(self.seed_sup_ent1[i], self.seed_sup_ent2[i])
self.seed_links.append(seed_link)
self.seed_link_set.add(seed_link)
self.links=list()
self.typing_negs=list()
for i in range(len(self.seed_links)):
typing_neg_links = list()
typing_sample=copy.copy(self.onto_neg_sample)
typing_sample.remove(self.seed_links[i][1])
typing_neg_ent2 = random.sample(typing_sample, self.nums_neg)
typing_neg_links.extend([(self.seed_links[i][0], typing_neg_ent2[k]) for k in range(self.nums_neg)])
typing_neg_links = torch.tensor(typing_neg_links).to(device)
seed_link=torch.tensor(list(self.seed_links[i])).to(device)
self.links.append(seed_link)
self.typing_negs.append(typing_neg_links)
def __len__(self):
return len(self.links)
def __getitem__(self,index):
return self.links[index],self.typing_negs[index]
def main(args):
config = vars(args)
if args.use_cuda:
device = torch.device("cuda")
config["device"]="cuda"
else:
device = torch.device("cpu")
config["device"]="cpu"
print("data folder:", args.input)
ins_info, onto_info, cross_info = read_input(args.input)
instance_info, ontology_info = prepare_EC_info (ins_info, onto_info, device)
ins_edge_labels = get_edge_labels(ins_info['max_n']).to(device)
onto_edge_labels = get_edge_labels(onto_info['max_n']).to(device)
ins_adj_info, onto_adj_info= prepare_adj_info(ins_info, onto_info, device)
model = DHGE(instance_info, ontology_info, ins_adj_info, onto_adj_info, config).to(device)
# E_train_dataloader
ins_train_facts=list()
for ins_train_fact in ins_info['train_facts']:
ins_train_fact=torch.tensor(ins_train_fact).to(device)
ins_train_facts.append(ins_train_fact)
train_data_E_reader=EDataset(ins_train_facts)
train_E_pyreader=DataLoader.DataLoader(train_data_E_reader,batch_size=args.batch_size,shuffle=True,drop_last=False)
# C_train_dataloader
onto_train_facts=list()
for onto_train_fact in onto_info['train_facts']:
onto_train_fact=torch.tensor(onto_train_fact).to(device)
onto_train_facts.append(onto_train_fact)
train_data_C_reader=CDataset(onto_train_facts)
train_C_pyreader=DataLoader.DataLoader(train_data_C_reader,batch_size=args.batch_size,shuffle=True,drop_last=False)
# S_train_dataloader
train_data_S_reader=SDataset(cross_info,ins_info,onto_info,config['nums_neg'],device)
train_S_batch_size=max(math.ceil(args.batch_size/ len(ins_info['train_facts'][0]) *train_data_S_reader.__len__())-1,2)
train_S_pyreader=DataLoader.DataLoader(train_data_S_reader,batch_size=train_S_batch_size,shuffle=True,drop_last=False)
# train_information
logging.info("train_ins_batch_size: "+str(args.batch_size))
logging.info("train_onto_batch_size: "+str(args.batch_size))
logging.info("train_cross_batch_size: "+str(train_S_batch_size))
steps = math.ceil(len(ins_info['train_facts']) / args.batch_size)
logging.info("train_steps_per_epoch: "+str(steps))
# E_valid_dataloader
ins_valid_facts=list()
for ins_valid_fact in ins_info['valid_facts']:
ins_valid_fact=torch.tensor(ins_valid_fact).to(device)
ins_valid_facts.append(ins_valid_fact)
valid_data_E_reader=EDataset(ins_valid_facts)
valid_E_pyreader=DataLoader.DataLoader(
valid_data_E_reader,
batch_size=args.batch_size,
shuffle=True,
drop_last=False)
# C_valid_dataloader
onto_valid_facts=list()
for onto_valid_fact in onto_info['valid_facts']:
onto_valid_fact=torch.tensor(onto_valid_fact).to(device)
onto_valid_facts.append(onto_valid_fact)
valid_data_C_reader=CDataset(onto_valid_facts)
valid_C_pyreader=DataLoader.DataLoader(
valid_data_C_reader,
batch_size=args.batch_size,
shuffle=True,
drop_last=False)
# S_valid_dataloader
ref_ent1 = cross_info['valid_cross'][0]
all_ref = [cross_info['all_cross'][str(k)] for k in ref_ent1]
ref_ent1=torch.tensor(ref_ent1).to('cuda')
# E_valid_dataloader
ins_test_facts=list()
for ins_test_fact in ins_info['test_facts']:
ins_test_fact=torch.tensor(ins_test_fact).to(device)
ins_test_facts.append(ins_test_fact)
test_data_E_reader=EDataset(ins_test_facts)
test_E_pyreader=DataLoader.DataLoader(
test_data_E_reader,
batch_size=args.batch_size,
shuffle=True,
drop_last=False)
# C_valid_dataloader
onto_test_facts=list()
for onto_test_fact in onto_info['test_facts']:
onto_test_fact=torch.tensor(onto_test_fact).to(device)
onto_test_facts.append(onto_test_fact)
test_data_C_reader=CDataset(onto_test_facts)
test_C_pyreader=DataLoader.DataLoader(
test_data_C_reader,
batch_size=args.batch_size,
shuffle=True,
drop_last=False)
# S_valid_dataloader
ref_ent1_test = cross_info['test_cross'][0]
all_ref_test = [cross_info['all_cross'][str(k)] for k in ref_ent1_test]
ref_ent1_test=torch.tensor(ref_ent1_test).to('cuda')
# ECS_optimizers
ins_optimizer=torch.optim.Adam(model.parameters(),lr=config['learning_rate'])
onto_optimizer=torch.optim.Adam(model.parameters(),lr=config['learning_rate'])
cross_optimizer=torch.optim.Adam(model.parameters(),lr=config['learning_rate']/5.0)
# Start Training
iterations = 1
for iteration in range(1, args.epochs // iterations + 1):
logger.info("iteration "+str(iteration))
t1 = time.time()
model.train()
for i in range(iterations):
ins_epoch_loss = 0
onto_epoch_loss =0
mapping_loss = 0
start = time.time()
for j,data in enumerate(zip(train_E_pyreader,cycle(train_C_pyreader),cycle(train_S_pyreader))):
ins_pos,onto_pos,pos_list=data
# E_training
ins_optimizer.zero_grad()
ins_loss,_= model.forward_E(ins_pos,ins_edge_labels)
ins_loss.backward()
ins_optimizer.step()
ins_epoch_loss += ins_loss
# C_training
onto_optimizer.zero_grad()
onto_loss,_= model.forward_C(onto_pos,onto_edge_labels)
onto_loss.backward()
onto_optimizer.step()
onto_epoch_loss += onto_loss
# S_training
cross_optimizer.zero_grad()
loss2 = model.forward_S(pos_list)
loss2.backward()
cross_optimizer.step()
mapping_loss += loss2
# print_ECS_loss_per_step
if j%100==0:
logger.info(str(j)+' , ins_loss: '+str(ins_loss.item())+' , onto_loss: '+str(onto_loss.item())+' , loss2: '+str(loss2.item()))
# print_ECS_loss_per_epoch
ins_epoch_loss /= steps
onto_epoch_loss /= steps
mapping_loss /= steps
end = time.time()
t2=round(end - start, 2)
logger.info("ins_epoch_loss = {:.3f}, onto_epoch_loss = {:.3f}, typing_loss = {:.3f}, time = {:.3f} s".format(ins_epoch_loss, onto_epoch_loss, mapping_loss, t2))
# Start validation and testing
model.eval()
with torch.no_grad():
# EC_validation
h1EC = predict(
model=model,
ins_test_pyreader=valid_E_pyreader,
onto_test_pyreader=valid_C_pyreader,
ins_all_facts=ins_info['all_facts'],
onto_all_facts=onto_info['all_facts'],
ins_edge_labels=ins_edge_labels,
onto_edge_labels=onto_edge_labels,
device=device)
# S_validation
h1 = model.test(ref_ent1,all_ref)
# EC_testing
h2EC = predict(
model=model,
ins_test_pyreader=test_E_pyreader,
onto_test_pyreader=test_C_pyreader,
ins_all_facts=ins_info['all_facts'],
onto_all_facts=onto_info['all_facts'],
ins_edge_labels=ins_edge_labels,
onto_edge_labels=onto_edge_labels,
device=device)
# S_testing
h2 = model.test(ref_ent1_test,all_ref_test)
logger.info("stop")
def predict(model, ins_test_pyreader, onto_test_pyreader,
ins_all_facts,onto_all_facts,
ins_edge_labels,onto_edge_labels,device):
start=time.time()
step = 0
ins_ret_ranks=dict()
ins_ret_ranks['entity']=torch.empty(0).to(device)
ins_ret_ranks['relation']=torch.empty(0).to(device)
ins_ret_ranks['2-r']=torch.empty(0).to(device)
ins_ret_ranks['2-ht']=torch.empty(0).to(device)
ins_ret_ranks['n-r']=torch.empty(0).to(device)
ins_ret_ranks['n-ht']=torch.empty(0).to(device)
ins_ret_ranks['n-a']=torch.empty(0).to(device)
ins_ret_ranks['n-v']=torch.empty(0).to(device)
onto_ret_ranks=dict()
onto_ret_ranks['entity']=torch.empty(0).to(device)
onto_ret_ranks['relation']=torch.empty(0).to(device)
onto_ret_ranks['2-r']=torch.empty(0).to(device)
onto_ret_ranks['2-ht']=torch.empty(0).to(device)
onto_ret_ranks['n-r']=torch.empty(0).to(device)
onto_ret_ranks['n-ht']=torch.empty(0).to(device)
onto_ret_ranks['n-a']=torch.empty(0).to(device)
onto_ret_ranks['n-v']=torch.empty(0).to(device)
#while steps < max_train_steps:
for i, data in enumerate(zip(ins_test_pyreader,cycle(onto_test_pyreader))):
ins_pos,onto_pos=data
_,ins_np_fc_out= model.forward_E(ins_pos,ins_edge_labels)
_,onto_np_fc_out= model.forward_C(onto_pos,onto_edge_labels)
ins_ret_ranks=batch_evaluation(ins_np_fc_out, ins_pos, ins_all_facts,ins_ret_ranks,device)
onto_ret_ranks=batch_evaluation(onto_np_fc_out, onto_pos, onto_all_facts,onto_ret_ranks,device)
#print("Processing prediction steps: %d | ins_examples: %d | onto_examples: %d " % (step, ins_global_idx,onto_global_idx))
step += 1
ins_eval_performance = compute_metrics(ins_ret_ranks)
ins_all_entity = "ENTITY\t\t%.4f\t%.4f\t%.4f\t%.4f\t%.4f" % (
ins_eval_performance['entity']['mrr'],
ins_eval_performance['entity']['hits1'],
ins_eval_performance['entity']['hits3'],
ins_eval_performance['entity']['hits5'],
ins_eval_performance['entity']['hits10'])
ins_all_relation = "RELATION\t%.4f\t%.4f\t%.4f\t%.4f\t%.4f" % (
ins_eval_performance['relation']['mrr'],
ins_eval_performance['relation']['hits1'],
ins_eval_performance['relation']['hits3'],
ins_eval_performance['relation']['hits5'],
ins_eval_performance['relation']['hits10'])
ins_all_ht = "HEAD/TAIL\t%.4f\t%.4f\t%.4f\t%.4f\t%.4f" % (
ins_eval_performance['ht']['mrr'],
ins_eval_performance['ht']['hits1'],
ins_eval_performance['ht']['hits3'],
ins_eval_performance['ht']['hits5'],
ins_eval_performance['ht']['hits10'])
ins_all_r = "PRIMARY_R\t%.4f\t%.4f\t%.4f\t%.4f\t%.4f" % (
ins_eval_performance['r']['mrr'],
ins_eval_performance['r']['hits1'],
ins_eval_performance['r']['hits3'],
ins_eval_performance['r']['hits5'],
ins_eval_performance['r']['hits10'])
logger.info("\n-------- E Evaluation Performance --------\n%s\n%s\n%s\n%s\n%s" % (
"\t".join(["TASK\t", "MRR", "Hits@1", "Hits@3", "Hits@5", "Hits@10"]),
ins_all_ht, ins_all_r, ins_all_entity, ins_all_relation))
onto_eval_performance = compute_metrics(onto_ret_ranks)
onto_all_entity = "ENTITY\t\t%.4f\t%.4f\t%.4f\t%.4f\t%.4f" % (
onto_eval_performance['entity']['mrr'],
onto_eval_performance['entity']['hits1'],
onto_eval_performance['entity']['hits3'],
onto_eval_performance['entity']['hits5'],
onto_eval_performance['entity']['hits10'])
onto_all_relation = "RELATION\t%.4f\t%.4f\t%.4f\t%.4f\t%.4f" % (
onto_eval_performance['relation']['mrr'],
onto_eval_performance['relation']['hits1'],
onto_eval_performance['relation']['hits3'],
onto_eval_performance['relation']['hits5'],
onto_eval_performance['relation']['hits10'])
onto_all_ht = "HEAD/TAIL\t%.4f\t%.4f\t%.4f\t%.4f\t%.4f" % (
onto_eval_performance['ht']['mrr'],
onto_eval_performance['ht']['hits1'],
onto_eval_performance['ht']['hits3'],
onto_eval_performance['ht']['hits5'],
onto_eval_performance['ht']['hits10'])
onto_all_r = "PRIMARY_R\t%.4f\t%.4f\t%.4f\t%.4f\t%.4f" % (
onto_eval_performance['r']['mrr'],
onto_eval_performance['r']['hits1'],
onto_eval_performance['r']['hits3'],
onto_eval_performance['r']['hits5'],
onto_eval_performance['r']['hits10'])
logger.info("\n-------- C Evaluation Performance --------\n%s\n%s\n%s\n%s\n%s" % (
"\t".join(["TASK\t", "MRR", "Hits@1", "Hits@3", "Hits@5", "Hits@10"]),
onto_all_ht, onto_all_r, onto_all_entity, onto_all_relation))
end=time.time()
logger.info("INS and ONTO time: "+str(round(end - start, 3))+'s')
return (ins_eval_performance['entity']['hits1']+onto_eval_performance['entity']['hits1'])/2.0
if __name__ == '__main__':
print_arguments(args)
main(args)