-
Notifications
You must be signed in to change notification settings - Fork 0
/
main.py
203 lines (164 loc) · 8.21 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
import argparse
import os
from utils import load_data, prepare_data, save_model, save_result, make_batch, acc_func, loss_func
from neuralMT import NMT_Model
import numpy as np
from sklearn.model_selection import train_test_split
import time
rnn_arch = ['gru', 'lstm', 'bidirectional']
embed_dim = ['50','100','200','300']
parser = argparse.ArgumentParser()
parser.add_argument('--rnn_arch', '-a', metavar='RNN', default='bidirectional',
choices=rnn_arch,
help='RNN architecture: ' + ' | '.join(rnn_arch) +
' (default: rbidirectional)')
parser.add_argument('--epochs', default=20, type=int, metavar='N',
help='number of total epochs to run')
parser.add_argument('--hidden', default=1024, type=int, metavar='N',
help='number of hidden units of Recurrent Layer')
parser.add_argument('--embedding_dim', default=200, type=int, metavar='N',
help='dimension of embedding layer' + ' | '.join(embed_dim) +
' (default: 200)')
parser.add_argument('--batch_size', default=64, type=int,
metavar='N', help='mini-batch size (default: 128)')
parser.add_argument('--lr', '--learning-rate', default=0.01, type=float,
metavar='LR', help='initial learning rate')
parser.add_argument('--momentum', default=0.9, type=float, metavar='M',
help='momentum')
parser.add_argument('--weight_decay', '--wd', default=1e-4, type=float,
metavar='W', help='weight decay (default: 1e-4)')
parser.add_argument('--checkpoint', dest='checkpoint',
help='The directory used to save the trained models',
default='checkpoint', type=str)
parser.add_argument('--best_model_dir', dest='best_model_dir',
help='The directory used to save the best trained models',
default='best_model', type=str)
parser.add_argument('--dataset', dest='dataset',
help='The directory used to save the dataset',
default='fra.txt', type=str)
parser.add_argument('--data_dir', dest='data_dir',
help='The directory containing dataset',
default='data', type=str)
parser.add_argument('--embedding_dir', dest='embedding_dir',
help='The directory containing pretrained embeddings',
default='embedding', type=str)
parser.add_argument('--result_dir', dest='result',
help='The directory used to save the results',
default='result', type=str)
parser.add_argument('-glove', dest='glove', action='store_true',
help='using glove pretrained embedding')
args = parser.parse_args()
working_dir = '.'
data_dir = os.path.join(working_dir, args.data_dir )
checkpoint_dir = os.path.join(working_dir, args.checkpoint +'_'+args.rnn_arch )
if not os.path.exists(checkpoint_dir):
os.makedirs(checkpoint_dir)
best_model_dir = os.path.join(working_dir, args.best_model_dir+'_'+args.rnn_arch )
if not os.path.exists(best_model_dir):
os.makedirs(best_model_dir)
result_dir = os.path.join(working_dir, args.result+'_'+args.rnn_arch )
if not os.path.exists(result_dir):
os.makedirs(result_dir)
embedding_dir = os.path.join(working_dir,args.embedding_dir)
embedding = 'glove.6B.'+str(args.embedding_dim)+'d.txt'
def train_one_epoch(epoch, model,X_train, Y_train):
print("epoch:", epoch + 1)
loss, acc, data_count = 0.0, 0.0, 0
dataset = make_batch(X_train, Y_train, batch_size=args.batch_size)
for batch, inp, targ in dataset:
data_count += len(inp)
decoder_inp = np.zeros((len(targ), Ty))
decoder_inp[:, 1:] = targ[:, :-1]
decoder_inp[:, 0] = targ_vocab.word_idx['<sos>']
targ_one_hot = np.zeros((len(targ), Ty, targ_vocab_size), dtype='float32')
for idx, tokVec in enumerate(targ):
for tok_idx, tok in enumerate(tokVec):
if (tok > 0):
targ_one_hot[idx, tok_idx, tok] = 1
history = model.fit(inp, decoder_inp, targ_one_hot, batch_size=args.batch_size, verbose=0)
loss_b, acc_b = history.history['loss'][0], history.history['acc_func'][0]
loss += (loss_b * len(inp))
acc += (acc_b * len(inp))
if batch % 100 == 0:
print('Epoch {} Batch {} Loss {:.4f} Accuracy {:.4f}'.format(epoch + 1,
batch,
loss / data_count, acc / data_count))
def evaluate(model, dataset, batch_size=64, verbose=0):
loss, acc, data_count = 0.0, 0.0, 0
for batch, inp, targ in dataset:
data_count += len(inp)
decoder_inp = np.zeros((len(targ), Ty))
decoder_inp[:, 1:] = targ[:, :-1]
decoder_inp[:, 0] = targ_vocab.word_idx['<sos>']
targ_one_hot = np.zeros((len(targ), Ty, targ_vocab_size), dtype='float32')
for idx, tokVec in enumerate(targ):
for tok_idx, tok in enumerate(tokVec):
if (tok > 0):
targ_one_hot[idx, tok_idx, tok] = 1
loss_b, acc_b = model.evaluate(inp, decoder_inp, targ_one_hot, batch_size=batch_size, verbose=verbose)
loss += loss_b * len(inp)
acc += acc_b * len(inp)
return loss / data_count, acc / data_count
if __name__ == '__main__':
sentence_pairs = load_data(os.path.join(data_dir, args.dataset))
X, Y, inp_vocab, targ_vocab, Tx, Ty = prepare_data(sentence_pairs , num_examples= 100, Tx= 100, Ty=100)
X_train, X_test, Y_train, Y_test = train_test_split(X, Y, test_size=0.2)
del X
del Y
inp_vocab_size = len(inp_vocab.word_idx)
targ_vocab_size = len(targ_vocab.word_idx)
# setting
HIDDEN_UNITS = args.hidden
EMBEDDING_DIM = args.embedding_dim
encoder_units = HIDDEN_UNITS
decoder_units = HIDDEN_UNITS
wordVec = {}
print('Loading wordVec')
# load in word vectors in a dict
word_embedding = np.zeros((inp_vocab_size, EMBEDDING_DIM))
# if args.glove:
# with open(os.path.join(embedding_dir, embedding)) as f:
# for line in f:
# data = line.split()
# word = data[0]
# vec = np.asarray(data[1:], dtype='float32')
# wordVec[word] = vec
#
# print('Finished loading wordVec.')
#
#
# # create word embedding by fetching each word vector
# for tok, idx in inp_vocab.word_idx.items():
# if idx < inp_vocab_size:
# word_vector = wordVec.get(tok)
# if word_vector is not None:
# word_embedding[idx] = word_vector
model = NMT_Model(args.rnn_arch, Tx, Ty, encoder_units, decoder_units, EMBEDDING_DIM, inp_vocab_size,
targ_vocab_size, word_embedding)
model.compile(opt='adam', loss=loss_func, metrics=[acc_func])
# final final debug
### debug
EPOCHS = args.epochs
dataset = make_batch(X_train, Y_train, batch_size=args.batch_size)
test_dataset = make_batch(X_test, Y_test, shuffle=False, batch_size=args.batch_size)
loss, acc = [], []
loss_test, acc_test = [],[]
best_acc =0.0
for epoch in range(EPOCHS):
start = time.time()
train_one_epoch(epoch, model, X_train, Y_train)
print('Time taken for 1 epoch training {} sec\n'.format(time.time() - start))
start = time.time()
loss_train_e, acc_train_e = evaluate(model, dataset, batch_size=args.batch_size)
loss_test_e, acc_test_e = evaluate(model, test_dataset, batch_size=args.batch_size)
print('Epoch {} Loss {:.4f} Accuracy {:.4f}'.format(epoch + 1, loss_train_e, acc_train_e))
print('Epoch {} Loss on test {:.4f} Accuracy on test {:.4f}'.format(epoch + 1, loss_test_e, acc_test_e))
print('Time taken for 1 epoch evaluating {} sec\n'.format(time.time() - start))
if loss_test_e > best_acc:
save_model(model, epoch, best_model_dir)
loss.append(loss_train_e)
acc.append(acc_train_e)
loss_test.append(loss_test_e)
acc_test.append(acc_test_e)
save_model(model, epoch, checkpoint_dir)
save_result(loss, acc, loss_test, acc_test, best_acc, result_dir)