-
Notifications
You must be signed in to change notification settings - Fork 50
/
run.py
169 lines (135 loc) · 6.33 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
from model import CNN
import utils
from torch.autograd import Variable
import torch
import torch.optim as optim
import torch.nn as nn
from sklearn.utils import shuffle
from gensim.models.keyedvectors import KeyedVectors
import numpy as np
import argparse
import copy
def train(data, params):
if params["MODEL"] != "rand":
# load word2vec
print("loading word2vec...")
word_vectors = KeyedVectors.load_word2vec_format("GoogleNews-vectors-negative300.bin", binary=True)
wv_matrix = []
for i in range(len(data["vocab"])):
word = data["idx_to_word"][i]
if word in word_vectors.vocab:
wv_matrix.append(word_vectors.word_vec(word))
else:
wv_matrix.append(np.random.uniform(-0.01, 0.01, 300).astype("float32"))
# one for UNK and one for zero padding
wv_matrix.append(np.random.uniform(-0.01, 0.01, 300).astype("float32"))
wv_matrix.append(np.zeros(300).astype("float32"))
wv_matrix = np.array(wv_matrix)
params["WV_MATRIX"] = wv_matrix
model = CNN(**params).cuda(params["GPU"])
parameters = filter(lambda p: p.requires_grad, model.parameters())
optimizer = optim.Adadelta(parameters, params["LEARNING_RATE"])
criterion = nn.CrossEntropyLoss()
pre_dev_acc = 0
max_dev_acc = 0
max_test_acc = 0
for e in range(params["EPOCH"]):
data["train_x"], data["train_y"] = shuffle(data["train_x"], data["train_y"])
for i in range(0, len(data["train_x"]), params["BATCH_SIZE"]):
batch_range = min(params["BATCH_SIZE"], len(data["train_x"]) - i)
batch_x = [[data["word_to_idx"][w] for w in sent] +
[params["VOCAB_SIZE"] + 1] * (params["MAX_SENT_LEN"] - len(sent))
for sent in data["train_x"][i:i + batch_range]]
batch_y = [data["classes"].index(c) for c in data["train_y"][i:i + batch_range]]
batch_x = Variable(torch.LongTensor(batch_x)).cuda(params["GPU"])
batch_y = Variable(torch.LongTensor(batch_y)).cuda(params["GPU"])
optimizer.zero_grad()
model.train()
pred = model(batch_x)
loss = criterion(pred, batch_y)
loss.backward()
nn.utils.clip_grad_norm(parameters, max_norm=params["NORM_LIMIT"])
optimizer.step()
dev_acc = test(data, model, params, mode="dev")
test_acc = test(data, model, params)
print("epoch:", e + 1, "/ dev_acc:", dev_acc, "/ test_acc:", test_acc)
if params["EARLY_STOPPING"] and dev_acc <= pre_dev_acc:
print("early stopping by dev_acc!")
break
else:
pre_dev_acc = dev_acc
if dev_acc > max_dev_acc:
max_dev_acc = dev_acc
max_test_acc = test_acc
best_model = copy.deepcopy(model)
print("max dev acc:", max_dev_acc, "test acc:", max_test_acc)
return best_model
def test(data, model, params, mode="test"):
model.eval()
if mode == "dev":
x, y = data["dev_x"], data["dev_y"]
elif mode == "test":
x, y = data["test_x"], data["test_y"]
x = [[data["word_to_idx"][w] if w in data["vocab"] else params["VOCAB_SIZE"] for w in sent] +
[params["VOCAB_SIZE"] + 1] * (params["MAX_SENT_LEN"] - len(sent))
for sent in x]
x = Variable(torch.LongTensor(x)).cuda(params["GPU"])
y = [data["classes"].index(c) for c in y]
pred = np.argmax(model(x).cpu().data.numpy(), axis=1)
acc = sum([1 if p == y else 0 for p, y in zip(pred, y)]) / len(pred)
return acc
def main():
parser = argparse.ArgumentParser(description="-----[CNN-classifier]-----")
parser.add_argument("--mode", default="train", help="train: train (with test) a model / test: test saved models")
parser.add_argument("--model", default="rand", help="available models: rand, static, non-static, multichannel")
parser.add_argument("--dataset", default="TREC", help="available datasets: MR, TREC")
parser.add_argument("--save_model", default=False, action='store_true', help="whether saving model or not")
parser.add_argument("--early_stopping", default=False, action='store_true', help="whether to apply early stopping")
parser.add_argument("--epoch", default=100, type=int, help="number of max epoch")
parser.add_argument("--learning_rate", default=1.0, type=float, help="learning rate")
parser.add_argument("--gpu", default=-1, type=int, help="the number of gpu to be used")
options = parser.parse_args()
data = getattr(utils, f"read_{options.dataset}")()
data["vocab"] = sorted(list(set([w for sent in data["train_x"] + data["dev_x"] + data["test_x"] for w in sent])))
data["classes"] = sorted(list(set(data["train_y"])))
data["word_to_idx"] = {w: i for i, w in enumerate(data["vocab"])}
data["idx_to_word"] = {i: w for i, w in enumerate(data["vocab"])}
params = {
"MODEL": options.model,
"DATASET": options.dataset,
"SAVE_MODEL": options.save_model,
"EARLY_STOPPING": options.early_stopping,
"EPOCH": options.epoch,
"LEARNING_RATE": options.learning_rate,
"MAX_SENT_LEN": max([len(sent) for sent in data["train_x"] + data["dev_x"] + data["test_x"]]),
"BATCH_SIZE": 50,
"WORD_DIM": 300,
"VOCAB_SIZE": len(data["vocab"]),
"CLASS_SIZE": len(data["classes"]),
"FILTERS": [3, 4, 5],
"FILTER_NUM": [100, 100, 100],
"DROPOUT_PROB": 0.5,
"NORM_LIMIT": 3,
"GPU": options.gpu
}
print("=" * 20 + "INFORMATION" + "=" * 20)
print("MODEL:", params["MODEL"])
print("DATASET:", params["DATASET"])
print("VOCAB_SIZE:", params["VOCAB_SIZE"])
print("EPOCH:", params["EPOCH"])
print("LEARNING_RATE:", params["LEARNING_RATE"])
print("EARLY_STOPPING:", params["EARLY_STOPPING"])
print("SAVE_MODEL:", params["SAVE_MODEL"])
print("=" * 20 + "INFORMATION" + "=" * 20)
if options.mode == "train":
print("=" * 20 + "TRAINING STARTED" + "=" * 20)
model = train(data, params)
if params["SAVE_MODEL"]:
utils.save_model(model, params)
print("=" * 20 + "TRAINING FINISHED" + "=" * 20)
else:
model = utils.load_model(params).cuda(params["GPU"])
test_acc = test(data, model, params)
print("test acc:", test_acc)
if __name__ == "__main__":
main()