-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
180 lines (118 loc) · 4.56 KB
/
train.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
# for data
import pandas as pd
import numpy as np
# for deep learning
from tensorflow.keras.layers import Embedding, Bidirectional, LSTM, Dense, Permute, multiply, Activation, Input, concatenate
from tensorflow.keras.models import Model
from tensorflow.keras.callbacks import EarlyStopping, ModelCheckpoint
from tensorflow.keras.preprocessing.text import Tokenizer
from tensorflow.keras.preprocessing.sequence import pad_sequences
# constants
DATA_PATH = 'data/'
MODEL_PATH = 'models/'
MAX_LEN = 30
'''
Read prepared train dataset
'''
# read dataset
df_train = pd.read_csv(DATA_PATH + 'train.csv')
print(df_train.head())
# get X
X_train = df_train['text'].values
# get other numerical and categorical features
feat_train = df_train.drop(['text', 'y'], axis=1).values
# get target
y_train = df_train["y"].values
'''
Load prepared word embeddings
'''
emb_matrix = np.genfromtxt(DATA_PATH + 'word_emb.csv', delimiter=',')
embedding_layer = Embedding(input_dim=emb_matrix.shape[0], output_dim=emb_matrix.shape[1], weights=[
emb_matrix], input_length=MAX_LEN, trainable=False)
'''
Define Bi-directional LSTM model
'''
# code attention layer
def attention_layer(inputs, neurons):
X = Permute((2, 1))(inputs)
X = Dense(neurons, activation="softmax")(X)
X = Permute((2, 1), name="attention")(X)
X = multiply([inputs, X])
return X
# model
def bid_lstm(input_shape, additional_input_shape=None):
# input
X_indices = Input(input_shape, name="text_input")
inputs = X_indices
# additional input layer of other features besides text
if additional_input_shape:
feat_input = Input(additional_input_shape, name='feat_input')
inputs = [X_indices, feat_input]
feat_input = Activation('tanh')(feat_input)
# embedding
embeddings = embedding_layer(X_indices)
# apply attention
X = attention_layer(embeddings, neurons=MAX_LEN)
# 2 layers of bidirectional lstm
X = Bidirectional(LSTM(
units=MAX_LEN, dropout=0.2, return_sequences=True))(X)
X = Bidirectional(LSTM(units=MAX_LEN, dropout=0.2))(X)
# concatenating additional features before prediction layer
if 'feat_input' in locals():
X = concatenate([X, feat_input])
# final dense layers
X = Dense(64, activation='relu')(X)
y_out = Dense(1, activation='sigmoid')(X)
# compile
model = Model(inputs=inputs, outputs=y_out)
model.compile(loss='binary_crossentropy',
optimizer='adam', metrics=['accuracy'])
return model
# tokenizing sequences from training data
tokenizer = Tokenizer()
tokenizer.fit_on_texts(X_train)
X_train_indices = tokenizer.texts_to_sequences(X_train)
# padding sequences to uniform length
X_train_indices = pad_sequences(X_train_indices, maxlen=MAX_LEN)
'''
1. Text-only model
'''
# model for text features only
model = bid_lstm((MAX_LEN,))
print('Text-only model')
model.summary()
# early stopping and model checkpoints
early_stopping = EarlyStopping(monitor='val_loss', patience=5)
bst_model_path = MODEL_PATH + 'bid_lstm.h5'
model_checkpoint = ModelCheckpoint(
bst_model_path, save_best_only=True, save_weights_only=False)
# training
hist = model.fit(X_train_indices, y_train, validation_split=0.2,
batch_size=2048, epochs=50, callbacks=[early_stopping, model_checkpoint])
bst_val_score = min(hist.history['val_loss'])
print('\n-----------------------------------\n')
print('best val score: {}'.format(bst_val_score))
print('model saved to {}'.format(bst_model_path))
print('\n-----------------------------------\n')
'''
2. Text + additional features model
'''
# model for text + additional features
model = bid_lstm((MAX_LEN,), (feat_train.shape[1],))
print('Text + additional features model')
model.summary()
# early stopping and model checkpoints
early_stopping = EarlyStopping(monitor='val_loss', patience=5)
bst_model_path = MODEL_PATH + 'bid_lstm_feats.h5'
model_checkpoint = ModelCheckpoint(
bst_model_path, save_best_only=True, save_weights_only=False)
# training
# text + additional features model
hist_text_feat = model.fit([X_train_indices, feat_train], y_train, validation_split=0.2,
batch_size=2048, epochs=50, callbacks=[early_stopping, model_checkpoint])
bst_val_score = min(hist.history['val_loss'])
print('\n-----------------------------------\n')
print('best val score: {}'.format(bst_val_score))
print('model saved to {}'.format(bst_model_path))
print('\n-----------------------------------\n')
# Note: The models will be saved to disk after training. Run test.py to evaluate the saved models.