-
Notifications
You must be signed in to change notification settings - Fork 1
/
cnn.py
246 lines (195 loc) · 7.91 KB
/
cnn.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
import time
import os
import argparse
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from torch.optim import lr_scheduler
from torch.utils.data import Dataset, DataLoader
from sklearn.metrics import accuracy_score, f1_score
import numpy as np
from preprocess import load_cnn_trainset, load_cnn_valset, \
load_cnn_testset, load_embeddings_as_tensor
from utils import save_result
class CNNMessageClassifier(nn.Module):
def __init__(self, n_classes, embeddings, dropout_prob=0.5,
kernel_num=100, kernel_sizes=[2, 3, 4, 5]):
super(CNNMessageClassifier, self).__init__()
self.embedding = nn.Embedding.from_pretrained(embeddings, freeze=True)
self.convs = nn.ModuleList(
[nn.Conv2d(1, kernel_num, (k, embeddings.size(1))) for k in kernel_sizes])
self.dropout = nn.Dropout(p=0.5)
self.fc = nn.Sequential(
nn.Linear(kernel_num * len(kernel_sizes), 200),
nn.ReLU(),
nn.Linear(200, n_classes)
)
def forward(self, x):
# [N, W]
x = self.embedding(x)
# [N, W, embedding_dim]
x = x.unsqueeze(1)
# [N, 1, W, embedding_dim]
x_list = [F.relu(conv(x)).squeeze(3) for conv in self.convs]
# [N, kernel_num, W] * len(kernel_sizes)
x_list = [F.max_pool1d(x, x.size(2)).squeeze(2) for x in x_list]
# [N, kernel_num*len(kernel_sizes)]
x = torch.cat(x_list, 1)
# [N, kernel_num*len(kernel_sizes)]
x = self.dropout(x)
x = self.fc(x)
# [N, n_classes]
return x
class Text2EmojiTrainset(Dataset):
def __init__(self):
X, y = load_cnn_trainset()
self.X, self.y = torch.LongTensor(X), torch.LongTensor(y)
def __len__(self):
return self.X.size(0)
def __getitem__(self, index):
return self.X[index], self.y[index]
class Text2EmojiValset(Dataset):
def __init__(self):
X, y = load_cnn_valset()
self.X, self.y = torch.LongTensor(X), torch.LongTensor(y)
def __len__(self):
return self.X.size(0)
def __getitem__(self, index):
return self.X[index], self.y[index]
class Text2EmojiTestset(Dataset):
def __init__(self):
X = load_cnn_testset()
self.X = torch.LongTensor(X)
def __len__(self):
return self.X.size(0)
def __getitem__(self, index):
return self.X[index]
MODELS_DIR = 'models'
def save_model(model, name):
save_path = os.path.join(MODELS_DIR, name + '.pt')
torch.save(model.state_dict(), save_path)
def load_model(model, name):
load_path = os.path.join(MODELS_DIR, name + '.pt')
model.load_state_dict(torch.load(load_path, map_location='cpu'))
def train(model, name, max_epoch=10, batch_size=100,
lr=1e-4, weight_decay=0.0):
trainset = Text2EmojiTrainset()
valset = Text2EmojiValset()
trainloader = DataLoader(trainset, batch_size=batch_size,
shuffle=True, num_workers=4)
validateloader = DataLoader(valset, batch_size=10000,
shuffle=False, num_workers=4)
criterion = nn.CrossEntropyLoss()
optimizer = optim.Adam(model.parameters(), lr=lr,
weight_decay=weight_decay)
scheduler = lr_scheduler.ExponentialLR(optimizer, 0.8)
print('Begin training. batch size: %d, epoch: %d, learning rate: %f' %
(batch_size, max_epoch, lr))
best_score = 0.0
model.train()
for epoch in range(max_epoch):
scheduler.step()
n_sample = 0
iter_ctr = 0
train_loss = 0.0
train_score = 0
time_start = time.time()
for X_train, y_train in trainloader:
optimizer.zero_grad()
outputs = model(X_train)
loss = criterion(outputs, y_train)
loss.backward()
optimizer.step()
iter_ctr += 1
n_sample += batch_size
train_loss += loss.item()
train_score += eval_score(outputs, y_train)
if iter_ctr == 10:
# Calculate validation loss and score
val_loss, val_score = validate(
model, criterion, validateloader)
# Average training loss.
train_loss /= iter_ctr
train_score /= iter_ctr
# Show results.
print('[%d, %5d] training loss: %.3f, validation loss: %.3f, '
'train f1: %.3f, validation f1: %.3f, time %.2f sec' %
(epoch, n_sample, train_loss, val_loss,
train_score, val_score, time.time() - time_start))
# Save the model if it is better than before
if val_score >= best_score:
save_model(model, name)
best_score = val_score + 0.001
print('Model is saved with score: %.3f' % val_score)
train_loss = 0.0
train_score = 0
iter_ctr = 0
time_start = time.time()
print('Training complete')
def eval_score(outputs, labels):
pred = predict(outputs)
pred, labels = pred.cpu(), labels.cpu()
return f1_score(labels, pred, average='micro')
def predict(outputs):
return torch.argmax(outputs, dim=1)
def validate(model, criterion, validateloader):
model.eval()
with torch.no_grad():
outputs_list, labels_list = [], []
for X_val, y_val in validateloader:
outputs_list.append(model(X_val))
labels_list.append(y_val)
outputs = torch.cat(outputs_list, 0)
labels = torch.cat(labels_list)
loss = criterion(outputs, labels)
score = eval_score(outputs, labels)
model.train()
return loss.item(), score
def test(model, result_file):
print('Begin testing')
model.eval()
dataset = Text2EmojiTestset()
testloader = DataLoader(dataset, batch_size=10000,
shuffle=False, num_workers=4)
pred = []
with torch.no_grad():
for data in testloader:
pred += list(predict(model(data)))
save_result(result_file, pred)
model.train()
print('Testing complete')
def parse_args():
parser = argparse.ArgumentParser(description='Train and test by cnn')
parser.add_argument('cmd', choices=['train', 'test'],
help='sub-commands')
parser.add_argument('-n', '--name', default='model',
help='the name of model')
parser.add_argument('-d', '--dropout', type=float, default=0.5,
help='the dropout probability of the model')
parser.add_argument('-k', '--kernel-num', dest='kernel_num',
type=float, default=200,
help='the kernel num of the model')
parser.add_argument('-s', '--kernel-sizes', dest='kernel_sizes',
type=int, nargs='+', default=[2, 3, 4, 5],
help='the kernel num of the model')
parser.add_argument('-e', '--epoch', type=int, default=5,
help='the max epoch for training')
parser.add_argument('-b', '--batch', type=int, default=100,
help='the batch size for training')
parser.add_argument('-l', '--lr', type=float, default=1e-4,
help='the learning rate for training')
parser.add_argument('-o', '--output', default='cnn_result.csv',
help='the output result for testing')
return parser.parse_args()
if __name__ == "__main__":
args = parse_args()
net = CNNMessageClassifier(
72, load_embeddings_as_tensor(), dropout_prob=args.dropout,
kernel_num=args.kernel_num, kernel_sizes=args.kernel_sizes)
if args.cmd == 'train':
train(net, args.name, max_epoch=args.epoch,
batch_size=args.batch, lr=args.lr)
elif args.cmd == 'test':
load_model(net, args.name)
test(net, args.output)