-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathmain.py
210 lines (179 loc) · 9 KB
/
main.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
import json
import torch
import pickle
import numpy as np
import argparse
import sys
import os
import math
import time
from os.path import join
import torch.backends.cudnn as cudnn
from evaluation import ranking_and_hits
from model import ConvE, DistMult, Complex, DistMultDecompress, ComplexDecompress, EnDistMultDecompress, RelDistMultDecompress, LDecomDistMult, LEnDecomDistMult, LRelDecomDistMult, LComplexDeComplex
from model import EnComplexDecompress
from spodernet.preprocessing.pipeline import Pipeline, DatasetStreamer
from spodernet.preprocessing.processors import JsonLoaderProcessors, Tokenizer, AddToVocab, SaveLengthsToState, StreamToHDF5, SaveMaxLengthsToState, CustomTokenizer
from spodernet.preprocessing.processors import ConvertTokenToIdx, ApplyFunction, ToLower, DictKey2ListMapper, ApplyFunction, StreamToBatch
from spodernet.utils.global_config import Config, Backends
from spodernet.utils.logger import Logger, LogLevel
from spodernet.preprocessing.batching import StreamBatcher
from spodernet.preprocessing.pipeline import Pipeline
from spodernet.preprocessing.processors import TargetIdx2MultiTarget
from spodernet.hooks import LossHook, ETAHook
from spodernet.utils.util import Timer
from spodernet.preprocessing.processors import TargetIdx2MultiTarget
np.set_printoptions(precision=3)
cudnn.benchmark = True
# parse console parameters and set global variables
Config.backend = Backends.TORCH
Config.parse_argv(sys.argv)
Config.cuda = True
#Config.embedding_dim = 400
#Logger.GLOBAL_LOG_LEVEL = LogLevel.DEBUG
print(Config.epochs)
#model_name = 'DistMult_{0}_{1}'.format(Config.input_dropout, Config.dropout)
model_name = '{2}_{0}_{1}_{3}'.format(Config.input_dropout, Config.dropout, Config.model_name, Config.learning_rate)
log_name = '{2}_{0}_{1}_{3}.txt'.format(Config.input_dropout, Config.dropout, Config.model_name, Config.learning_rate)
print(log_name)
if os.path.exists(log_name):
os.remove(log_name)
epochs = Config.epochs
load = False
if Config.dataset is None:
Config.dataset = 'FB15k-237'
model_path = 'saved_models/{0}_{1}.model'.format(Config.dataset, model_name)
''' Preprocess knowledge graph using spodernet. '''
def preprocess(dataset_name, delete_data=False):
full_path = 'data/{0}/e1rel_to_e2_full.json'.format(dataset_name)
train_path = 'data/{0}/e1rel_to_e2_train.json'.format(dataset_name)
dev_ranking_path = 'data/{0}/e1rel_to_e2_ranking_dev.json'.format(dataset_name)
test_ranking_path = 'data/{0}/e1rel_to_e2_ranking_test.json'.format(dataset_name)
keys2keys = {}
keys2keys['e1'] = 'e1' # entities
keys2keys['rel'] = 'rel' # relations
keys2keys['rel_eval'] = 'rel' # relations
keys2keys['e2'] = 'e1' # entities
keys2keys['e2_multi1'] = 'e1' # entity
keys2keys['e2_multi2'] = 'e1' # entity
input_keys = ['e1', 'rel', 'rel_eval', 'e2', 'e2_multi1', 'e2_multi2']
d = DatasetStreamer(input_keys)
d.add_stream_processor(JsonLoaderProcessors())
d.add_stream_processor(DictKey2ListMapper(input_keys))
# process full vocabulary and save it to disk
d.set_path(full_path)
p = Pipeline(Config.dataset, delete_data, keys=input_keys, skip_transformation=True)
p.add_sent_processor(ToLower())
p.add_sent_processor(CustomTokenizer(lambda x: x.split(' ')),keys=['e2_multi1', 'e2_multi2'])
p.add_token_processor(AddToVocab())
p.execute(d)
p.save_vocabs()
# process train, dev and test sets and save them to hdf5
p.skip_transformation = False
for path, name in zip([train_path, dev_ranking_path, test_ranking_path], ['train', 'dev_ranking', 'test_ranking']):
d.set_path(path)
p.clear_processors()
p.add_sent_processor(ToLower())
p.add_sent_processor(CustomTokenizer(lambda x: x.split(' ')),keys=['e2_multi1', 'e2_multi2'])
p.add_post_processor(ConvertTokenToIdx(keys2keys=keys2keys), keys=['e1', 'rel', 'rel_eval', 'e2', 'e2_multi1', 'e2_multi2'])
p.add_post_processor(StreamToHDF5(name, samples_per_file=1000, keys=input_keys))
p.execute(d)
def main():
if Config.process: preprocess(Config.dataset, delete_data=True)
input_keys = ['e1', 'rel', 'rel_eval', 'e2', 'e2_multi1', 'e2_multi2']
p = Pipeline(Config.dataset, keys=input_keys)
p.load_vocabs()
vocab = p.state['vocab']
num_entities = vocab['e1'].num_token
train_batcher = StreamBatcher(Config.dataset, 'train', Config.batch_size, randomize=True, keys=input_keys)
dev_rank_batcher = StreamBatcher(Config.dataset, 'dev_ranking', Config.batch_size, randomize=False, loader_threads=4, keys=input_keys)
test_rank_batcher = StreamBatcher(Config.dataset, 'test_ranking', Config.batch_size, randomize=False, loader_threads=4, keys=input_keys)
if Config.model_name is None:
model = ConvE(vocab['e1'].num_token, vocab['rel'].num_token)
elif Config.model_name == 'ConvE':
model = ConvE(vocab['e1'].num_token, vocab['rel'].num_token)
elif Config.model_name == 'DistMult':
model = DistMult(vocab['e1'].num_token, vocab['rel'].num_token)
elif Config.model_name == 'ComplEx':
model = Complex(vocab['e1'].num_token, vocab['rel'].num_token)
elif Config.model_name == 'DistMultDecompress':
model = DistMultDecompress(vocab['e1'].num_token, vocab['rel'].num_token)
elif Config.model_name == 'EnDistMultDecompress':
model = EnDistMultDecompress(vocab['e1'].num_token, vocab['rel'].num_token)
elif Config.model_name == 'RelDistMultDecompress':
model = RelDistMultDecompress(vocab['e1'].num_token, vocab['rel'].num_token)
elif Config.model_name == 'ComplexDecompress':
model = ComplexDecompress(vocab['e1'].num_token, vocab['rel'].num_token)
elif Config.model_name == 'LDecomDistMult':
model = LDecomDistMult(vocab['e1'].num_token, vocab['rel'].num_token)
elif Config.model_name == 'LEnDecomDistMult':
model = LEnDecomDistMult(vocab['e1'].num_token, vocab['rel'].num_token)
elif Config.model_name == 'LRelDecomDistMult':
model = LRelDecomDistMult(vocab['e1'].num_token, vocab['rel'].num_token)
elif Config.model_name == 'LDeComplex':
model = LDeComplex(vocab['e1'].num_token, vocab['rel'].num_token)
elif Config.model_name == 'EnComplexDecompress':
model = EnComplexDecompress(vocab['e1'].num_token, vocab['rel'].num_token)
else:
print('Unknown model: {0}', Config.model_name)
raise Exception("Unknown model!")
print('num of paras is %d' %sum(p.numel() for p in model.parameters()))
train_batcher.at_batch_prepared_observers.insert(1,TargetIdx2MultiTarget(num_entities, 'e2_multi1', 'e2_multi1_binary'))
eta = ETAHook('train', print_every_x_batches=100)
train_batcher.subscribe_to_events(eta)
train_batcher.subscribe_to_start_of_epoch_event(eta)
train_batcher.subscribe_to_events(LossHook('train', print_every_x_batches=100))
if Config.cuda:
model.cuda()
if load:
model_params = torch.load(model_path)
total_param_size = []
params = [(key, value.size(), value.numel()) for key, value in model_params.items()]
for key, size, count in params:
total_param_size.append(count)
print(key, size, count)
print(np.array(total_param_size).sum())
model.load_state_dict(model_params)
model.eval()
ranking_and_hits(model, test_rank_batcher, vocab, 'test_evaluation')
ranking_and_hits(model, dev_rank_batcher, vocab, 'dev_evaluation')
else:
model.init()
total_param_size = []
params = [value.numel() for value in model.parameters()]
# print(np.sum(params))
opt = torch.optim.Adam(model.parameters(), lr=Config.learning_rate, weight_decay=Config.L2)
best_mrr = 0.
for epoch in range(epochs):
# model.eval()
# with torch.no_grad():
# start_time = time.time()
# dev_mrr = ranking_and_hits(model, dev_rank_batcher, vocab, 'dev_evaluation', log_name, epoch)
# print(time.time() - start_time)
model.train()
for i, str2var in enumerate(train_batcher):
opt.zero_grad()
e1 = str2var['e1']
rel = str2var['rel']
e2_multi = str2var['e2_multi1_binary'].float()
# label smoothing
e2_multi = ((1.0-Config.label_smoothing_epsilon)*e2_multi) + (1.0/e2_multi.size(1))
pred = model.forward(e1, rel)
loss = model.loss(pred, e2_multi)
# print(pred)
# print(e2_multi)
# bp = input()
loss.backward()
opt.step()
train_batcher.state.loss = loss.cpu()
print('saving to {0}'.format(model_path))
torch.save(model.state_dict(), model_path)
model.eval()
with torch.no_grad():
dev_mrr = ranking_and_hits(model, dev_rank_batcher, vocab, 'dev_evaluation', log_name, epoch)
if dev_mrr > best_mrr:
start_time = time.time()
best_mrr = dev_mrr
ranking_and_hits(model, test_rank_batcher, vocab, 'test_evaluation', log_name, epoch)
if __name__ == '__main__':
main()