-
Notifications
You must be signed in to change notification settings - Fork 0
/
model.py
208 lines (168 loc) · 7.99 KB
/
model.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
import torch
from torch import nn
import torch.nn.functional as F
class LSTMClassifier(nn.Module):
def __init__(self, vocab_size=50000, emb_dim=100, emb_vectors=None,
emb_dropout=0.3,
lstm_dim=256, lstm_n_layer=2, lstm_dropout=0.3,
bidirectional=True, lstm_combine='add',
n_linear=2, linear_dropout=0.5, n_classes=1,
crit=nn.CrossEntropyLoss()):
super().__init__()
vocab_size, emb_dim = emb_vectors.shape
n_dirs = bidirectional + 1
lstm_dir_dim = lstm_dim // n_dirs if lstm_combine == 'concat' else lstm_dim
self.lstm_n_layer = lstm_n_layer
self.n_dirs = n_dirs
self.lstm_dir_dim = lstm_dir_dim
self.lstm_combine = lstm_combine
self.embedding_layer = nn.Embedding(*emb_vectors.shape)
self.embedding_layer.from_pretrained(emb_vectors, padding_idx=1)
# pad=1 in torchtext; embedding weights trainable
self.embedding_dropout = nn.Dropout(p=emb_dropout)
self.lstm = nn.LSTM(emb_dim, lstm_dir_dim,
num_layers=lstm_n_layer,
bidirectional=bidirectional,
batch_first=True)
if lstm_n_layer > 1: self.lstm.dropout = lstm_dropout
self.lstm_dropout = nn.Dropout(p=lstm_dropout)
self.att_w = nn.Parameter(torch.randn(1, lstm_dim, 1))
self.linear_layers = [nn.Linear(lstm_dim, lstm_dim) for _ in
range(n_linear - 1)]
self.linear_layers = nn.ModuleList(self.linear_layers)
self.linear_dropout = nn.Dropout(p=linear_dropout)
self.label = nn.Linear(lstm_dim, n_classes)
self.crit = crit
self.opts = {
'vocab_size': vocab_size,
'emb_dim': emb_dim,
'emb_dropout': emb_dropout,
'emb_vectors': emb_vectors,
'lstm_dim': lstm_dim,
'lstm_n_layer': lstm_n_layer,
'lstm_dropout': lstm_dropout,
'lstm_combine': lstm_combine,
'n_linear': n_linear,
'linear_dropout': linear_dropout,
'n_classes': n_classes,
'crit': crit,
}
def attention_net(self, lstm_output, final_state):
"""
Now we will incorporate Attention mechanism in our LSTM model. In this new model, we will use attention to compute soft alignment score corresponding
between each of the hidden_state and the last hidden_state of the LSTM. We will be using torch.bmm for the batch matrix multiplication.
Arguments
---------
lstm_output : Final output of the LSTM which contains hidden layer outputs for each sequence.
final_state : Final time-step hidden state (h_n) of the LSTM
---------
Returns : It performs attention mechanism by first computing weights for each of the sequence present in lstm_output and and then finally computing the
new hidden state.
Tensor Size :
hidden.size() = (batch_size, hidden_size)
attn_weights.size() = (batch_size, num_seq)
soft_attn_weights.size() = (batch_size, num_seq)
new_hidden_state.size() = (batch_size, hidden_size)
"""
attn_weights = torch.bmm(lstm_output, final_state.unsqueeze(2)).squeeze(
2)
soft_attn_weights = F.softmax(attn_weights, 1).unsqueeze(
2) # shape = (batch_size, seq_len, 1)
new_hidden_state = torch.bmm(lstm_output.transpose(1, 2),
soft_attn_weights).squeeze(2)
return new_hidden_state
def re_attention(self, lstm_output, final_h, input):
batch_size, seq_len = input.shape
final_h = final_h.view(self.lstm_n_layer, self.n_dirs, batch_size,
self.lstm_dir_dim)[-1]
final_h = final_h.permute(1, 0, 2)
final_h = final_h.sum(dim=1) # (batch_size, 1, self.half_dim)
# final_h.size() = (batch_size, hidden_size)
# output.size() = (batch_size, num_seq, hidden_size)
if self.lstm_combine == 'add':
lstm_output = lstm_output.view(batch_size, seq_len, 2,
self.lstm_dir_dim)
lstm_output = lstm_output.sum(dim=2)
# lstm_output(batch_size, seq_len, lstm_dir_dim)
att = torch.bmm(torch.tanh(lstm_output),
self.att_w.repeat(batch_size, 1, 1))
att = F.softmax(att, dim=1) # att(batch_size, seq_len, 1)
att = torch.bmm(lstm_output.transpose(1, 2), att).squeeze(2)
attn_output = torch.tanh(att) # attn_output(batch_size, lstm_dir_dim)
return attn_output
def forward(self, input):
batch_size, seq_len, *_ = input.shape
inp = self.embedding_layer(input)
inp = self.embedding_dropout(inp)
lstm_output, (final_h, final_c) = self.lstm(inp)
# outputs = []
# for i in range(seq_len):
# cur_emb = inp[i:i + 1] # .view(1, inp.size(1), inp.size(2))
#
# o, hidden = self.lstm(cur_emb) if i == 0 else self.lstm(cur_emb, hidden)
# import pdb;pdb.set_trace()
# outputs += [o.unsqueeze(0)]
#
# outputs = torch.cat(outputs, dim=0)
lstm_output = self.lstm_dropout(lstm_output)
attn_output = self.re_attention(lstm_output, final_h, input)
output = self.linear_dropout(attn_output)
for layer in self.linear_layers:
output = layer(output)
output = self.linear_dropout(output)
output = F.relu(output)
logits = self.label(output)
return logits
def forward_normal_attention(self):
batch_size = len(input)
inp = self.embedding_layer(input)
inp = self.embedding_dropout(inp)
lstm_output, (final_h, final_c) = self.lstm(inp)
final_h = final_h.view(self.lstm_n_layer, self.n_dirs, batch_size,
self.lstm_dim // self.n_dirs)[-1]
final_h = final_h.permute(1, 0,
2) # (batch_size, 2, self.lstm_dim // self.n_dirs)
final_h = final_h.contiguous().view(batch_size, self.lstm_dim)
# final_h.size() = (batch_size, hidden_size)
# output.size() = (batch_size, num_seq, hidden_size)
attn_output = self.attention_net(lstm_output, final_h)
output = self.linear_dropout(attn_output)
for layer in self.linear_layers:
output = layer(output)
output = self.linear_dropout(output)
output = F.relu(output)
logits = self.label(output)
return logits
def forward_normal_lstm(self):
inp = self.embedding_layer(input)
inp = self.embedding_dropout(inp)
lstm_output, (final_h, final_c) = self.lstm(inp)
# output.size() = (batch_size, num_seq, hidden_size)
output = lstm_output[:, -1]
output = self.linear_dropout(output)
for layer in self.linear_layers:
output = layer(output)
output = self.linear_dropout(output)
output = F.relu(output)
logits = self.label(output)
return logits
def loss(self, input, target):
logits = self.forward(input)
logits_flat = logits.view(-1, logits.size(-1))
target_flat = target.view(-1)
loss = self.crit(logits_flat, target_flat) # mean_score per batch
return loss
def predict(self, input):
logits = self.forward(input)
logits[:, :2] = float('-inf')
preds = logits.max(dim=-1)[1]
preds = preds.detach().cpu().numpy().tolist()
return preds
def loss_n_acc(self, input, target):
logits = self.forward(input)
logits_flat = logits.view(-1, logits.size(-1))
target_flat = target.view(-1)
loss = self.crit(logits_flat, target_flat) # mean_score per batch
pred_flat = logits_flat.max(dim=-1)[1]
acc = (pred_flat == target_flat).sum()
return loss, acc.item()