forked from Nicoleqwerty/RDF-to-Text
-
Notifications
You must be signed in to change notification settings - Fork 0
/
LSTMModel.py
265 lines (213 loc) · 12.8 KB
/
LSTMModel.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
from beam_search_PG import Beam, sort_beams
from torch_geometric.data import DataLoader
from Encoder_Decoder import Encoder, Decoder, ReduceState, GCNEncoder
from utils import config
import torch
use_cuda = config.use_cuda and torch.cuda.is_available()
class LSTMModel(object):
def __init__(self, model_file_path=None, is_eval=False):
encoder = Encoder()
decoder = Decoder()
reduce_state = ReduceState()
# shared the embedding between encoder and decoder
decoder.embedding.weight = encoder.embedding.weight
if is_eval:
encoder = encoder.eval()
decoder = decoder.eval()
reduce_state = reduce_state.eval()
if use_cuda:
encoder = encoder.cuda()
decoder = decoder.cuda()
reduce_state = reduce_state.cuda()
self.encoder = encoder
self.decoder = decoder
self.reduce_state = reduce_state
if model_file_path is not None:
state = torch.load(model_file_path, map_location=lambda storage, location: storage)
self.encoder.load_state_dict(state['encoder_state_dict'])
self.decoder.load_state_dict(state['decoder_state_dict'], strict=False)
self.reduce_state.load_state_dict(state['reduce_state_dict'])
def forward(self, iterator, device, optimizer, epoch):
epoch_loss = 0
for i, (src_seqs_nl, trg_seqs_nl, src_seqs, src_lengths, trg_seqs, trg_lengths) in enumerate(iterator):
src_seqs = src_seqs.to(device) # (batch_size, seq_len)
trg_seqs = trg_seqs.to(device)
max_dec_len = max(trg_lengths)-1
enc_padding_mask = torch.gt(src_seqs,0).float() # (batch_size, seq_len)
dec_padding_mask = torch.gt(trg_seqs,0).float()
c_t_1 = torch.zeros((src_seqs.size()[0], 2 * config.hidden_dim)).to(device) # (batch_size, 2*hidden_dim)
coverage = None
if config.is_coverage:
coverage = torch.zeros(src_seqs.size()).to(device)
optimizer.zero_grad()
# encoder_outputs: (batch_size, seq_len, 2*hidden_dim)
# encoder_feature: (batch_size*seq_len, 2*hidden_dim)
# encoder_hidden: (2, batch_size, hidden_dim)
# (2, batch_size, hidden_dim)
encoder_outputs, encoder_feature, encoder_hidden = self.encoder(src_seqs, src_lengths)
s_t_1 = self.reduce_state(encoder_hidden) # ((1, batch_size, hidden_dim), (1, batch_size, hidden_dim))
step_losses = []
# preds = []
for di in range(min(max_dec_len, config.max_dec_steps)):
y_t_1 = trg_seqs[:, di] # Teacher forcing
final_dist, s_t_1, c_t_1, attn_dist, p_gen, next_coverage = self.decoder(y_t_1, s_t_1,
encoder_outputs,
encoder_feature,
enc_padding_mask, c_t_1,
extra_zeros=None,
enc_batch_extend_vocab=src_seqs,
coverage=coverage,
step = di)
target = trg_seqs[:, di+1]
gold_probs = torch.gather(final_dist, 1, target.unsqueeze(1)).squeeze()
step_loss = -torch.log(gold_probs + config.eps)
if config.is_coverage:
step_coverage_loss = torch.sum(torch.min(attn_dist, coverage), 1)
step_loss = step_loss + config.cov_loss_wt * step_coverage_loss
coverage = next_coverage
step_mask = dec_padding_mask[:, di+1]
step_loss = step_loss * step_mask
step_losses.append(step_loss)
sum_losses = torch.sum(torch.stack(step_losses, 1), 1)
batch_avg_loss = sum_losses / torch.tensor(trg_lengths).to(device)
loss = torch.mean(batch_avg_loss)
loss.backward()
torch.nn.utils.clip_grad_norm_(self.encoder.parameters(), config.max_grad_norm)
torch.nn.utils.clip_grad_norm_(self.decoder.parameters(), config.max_grad_norm)
torch.nn.utils.clip_grad_norm_(self.reduce_state.parameters(), config.max_grad_norm)
optimizer.step()
epoch_loss += float(loss.item()) #
avg_los = epoch_loss / (i + 1)
print('epoch : %d %10d batch loss: %.4f, avg_loss: %.4f' % (epoch, i + 1, loss, avg_los))
return epoch_loss/len(iterator)
def eval(self, iterator, device):
epoch_loss = 0
for i, (src_seqs_nl, trg_seqs_nl, src_seqs, src_lengths, trg_seqs, trg_lengths) in enumerate(iterator):
src_seqs = src_seqs.to(device) # (batch_size, seq_len)
trg_seqs = trg_seqs.to(device)
max_dec_len = max(trg_lengths) - 1
enc_padding_mask = torch.gt(src_seqs, 0).float()
dec_padding_mask = torch.gt(trg_seqs, 0).float()
c_t_1 = torch.zeros((src_seqs.size()[0], 2 * config.hidden_dim)).to(device)
coverage = None
if config.is_coverage:
coverage = torch.zeros(src_seqs.size()).to(device)
encoder_outputs, encoder_feature, encoder_hidden = self.encoder(src_seqs, src_lengths)
s_t_1 = self.reduce_state(encoder_hidden)
step_losses = []
for di in range(min(max_dec_len, config.max_dec_steps)):
y_t_1 = trg_seqs[:, di] # Teacher forcing
final_dist, s_t_1, c_t_1, attn_dist, p_gen, next_coverage = self.decoder(y_t_1, s_t_1,
encoder_outputs,
encoder_feature,
enc_padding_mask, c_t_1,
extra_zeros=None,
enc_batch_extend_vocab=src_seqs,
coverage=coverage,
step=di)
target = trg_seqs[:, di + 1]
gold_probs = torch.gather(final_dist, 1, target.unsqueeze(1)).squeeze()
step_loss = -torch.log(gold_probs + config.eps)
if config.is_coverage:
step_coverage_loss = torch.sum(torch.min(attn_dist, coverage), 1)
step_loss = step_loss + config.cov_loss_wt * step_coverage_loss
coverage = next_coverage
step_mask = dec_padding_mask[:, di + 1]
step_loss = step_loss * step_mask
step_losses.append(step_loss)
sum_losses = torch.sum(torch.stack(step_losses, 1), 1)
batch_avg_loss = sum_losses / torch.tensor(trg_lengths).to(device)
loss = torch.mean(batch_avg_loss)
epoch_loss += float(loss.item())
return epoch_loss
def beam_search(self, f, iterator, device, word2id, id2word):
for i, (src_seqs_nl, trg_seqs_nl, src_seqs, src_lengths, trg_seqs, trg_lengths) in enumerate(iterator):
src_seqs = torch.cat([src_seqs for n in range(config.beam_size)], dim=0).to(device) # (batch_size, seq_len)
# trg_seqs = torch.cat([trg_seqs for n in range(config.beam_size)], dim=0).to(device)
src_lengths = [src_lengths[0] for n in range(config.beam_size)]
max_dec_len = max(trg_lengths) - 1
enc_padding_mask = torch.gt(src_seqs, 0)
# dec_padding_mask = torch.gt(trg_seqs, 0)
c_t_0 = torch.zeros((src_seqs.size()[0], 2 * config.hidden_dim)).to(device)
encoder_outputs, encoder_feature, encoder_hidden = self.encoder(src_seqs, src_lengths)
s_t_0 = self.reduce_state(encoder_hidden)
dec_h, dec_c = s_t_0 # 1 x 2*hidden_size
dec_h = dec_h.squeeze()
dec_c = dec_c.squeeze()
coverage_t_0 = torch.zeros(src_seqs.size()).to(device)
# decoder batch preparation, it has beam_size example initially everything is repeated
beams = [Beam(tokens=[word2id['<start>']],
log_probs=[0.0],
state=(dec_h[0], dec_c[0]),
context=c_t_0[0],
coverage=(coverage_t_0[0] if config.is_coverage else None))
for _ in range(config.beam_size)]
results = []
steps = 0
while steps < config.max_dec_steps and len(results) < config.beam_size:
latest_tokens = [h.latest_token for h in beams]
latest_tokens = [t if t < len(word2id) else word2id['<unk>'] \
for t in latest_tokens]
y_t_1 = torch.LongTensor(latest_tokens)
if config.use_cuda:
y_t_1 = y_t_1.to(device)
all_state_h = []
all_state_c = []
all_context = []
for h in beams:
state_h, state_c = h.state
all_state_h.append(state_h)
all_state_c.append(state_c)
all_context.append(h.context)
s_t_1 = (torch.stack(all_state_h, 0).unsqueeze(0), torch.stack(all_state_c, 0).unsqueeze(0))
c_t_1 = torch.stack(all_context, 0)
coverage_t_1 = None
if config.is_coverage:
all_coverage = []
for h in beams:
all_coverage.append(h.coverage)
coverage_t_1 = torch.stack(all_coverage, 0)
final_dist, s_t, c_t, attn_dist, p_gen, coverage_t = self.decoder(y_t_1, s_t_1,
encoder_outputs, encoder_feature,
enc_padding_mask, c_t_1,
extra_zeros=None,
enc_batch_extend_vocab=src_seqs,
coverage=coverage_t_1,
step=steps)
log_probs = torch.log(final_dist)
topk_log_probs, topk_ids = torch.topk(log_probs, config.beam_size * 2)
dec_h, dec_c = s_t
dec_h = dec_h.squeeze()
dec_c = dec_c.squeeze()
all_beams = []
num_orig_beams = 1 if steps == 0 else len(beams)
for i in range(num_orig_beams):
h = beams[i]
state_i = (dec_h[i], dec_c[i])
context_i = c_t[i]
coverage_i = (coverage_t[i] if config.is_coverage else None)
for j in range(config.beam_size * 2): # for each of the top 2*beam_size hyps:
new_beam = h.extend(token=topk_ids[i, j].item(),
log_prob=topk_log_probs[i, j].item(),
state=state_i,
context=context_i,
coverage=coverage_i)
all_beams.append(new_beam)
beams = []
for h in sort_beams(all_beams):
if h.latest_token == word2id['<end>']:
if steps >= config.min_dec_steps:
results.append(h)
else:
beams.append(h)
if len(beams) == config.beam_size or len(results) == config.beam_size:
break
steps += 1
if len(results) == 0:
results = beams
beams_sorted = sort_beams(results)
generated_tokens = [id2word[int(i)] for i in beams_sorted[0].tokens]
print(trg_seqs_nl[0].strip())
print(' '.join(generated_tokens))
print()
f.write(' '.join(generated_tokens).replace('<start>', '').replace('<end>', '').strip() + '\n')