-
Notifications
You must be signed in to change notification settings - Fork 12
/
Copy pathtrain_random.py
185 lines (143 loc) · 6.26 KB
/
train_random.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
"""
Created on Sat Mar 17 11:01:27 2018
@author: leo
"""
import warnings
warnings.filterwarnings("ignore")
import pickle
import numpy as np
#import matplotlib as mpl
#mpl.use('Agg')
import matplotlib.pyplot as plt
from keras.models import Sequential
from keras.layers.core import Dense, Dropout, Activation, Flatten
from keras.layers.embeddings import Embedding
from keras.layers import MaxPooling2D
from keras.layers.convolutional import Conv2D
from keras import regularizers
from keras.optimizers import Adadelta
from preprocessing.word_embedding import rand_embed_sentences
from sklearn.utils import shuffle
# Build model
def build_model(input_shape, conv_window_size, num_filters, reg, dropout, word2vec = True, max_token = None, sequence_len= 190):
"""
If random init
max_token is the vocabulary size
sequence_len is the number of words in the largenst sentence
"""
model = Sequential()
#model.add(Embedding(max_features,300))
# we add a Convolution 1D, which will learn num_filters
# word group filters of size conv_window_size:
if (not(word2vec)) :
model.add(Embedding(max_token,300, input_length=sequence_len))
input_shape = (1, sequence_len, 300)
model.add(Conv2D(input_shape=input_shape,
filters=num_filters,
kernel_size=(1, conv_window_size),
padding="valid",
activation="relu",
strides=1,
data_format='channels_first'))
model.add(MaxPooling2D(pool_size=(num_filters, 1)))
#Fully Connected + Dropout + sigmoid
model.add(Flatten())
model.add(Dropout(dropout))
model.add(Dense(1, activation='sigmoid', kernel_regularizer=regularizers.l2(reg)))
#In addition, an l2−norm constraint of the weights w_r is imposed during training as well
model.compile(loss='binary_crossentropy',
optimizer=Adadelta(),
metrics=['mae'])
return model
def train(model, x_train, y_train, val_train_ratio=0.2, epochs=1000, batch_size=128):
history = model.fit(x_train, y_train,
batch_size=batch_size,
epochs=epochs,
validation_split=val_train_ratio,
shuffle=False,
verbose=1)
return history
def load_data():
print("loading pickle files...")
data1 = pickle.load(open("preprocessing/wordEmbeddingsToSaliency1.pickle", "rb"))
data2 = pickle.load(open("preprocessing/wordEmbeddingsToSaliency2.pickle", "rb"))
data3 = pickle.load(open("preprocessing/wordEmbeddingsToSaliency3.pickle", "rb"))
data4 = pickle.load(open("preprocessing/wordEmbeddingsToSaliency4.pickle", "rb"))
data5 = pickle.load(open("preprocessing/wordEmbeddingsToSaliency5.pickle", "rb"))
data6 = pickle.load(open("preprocessing/wordEmbeddingsToSaliency6.pickle", "rb"))
data7 = pickle.load(open("preprocessing/wordEmbeddingsToSaliency7.pickle", "rb"))
data8 = pickle.load(open("preprocessing/wordEmbeddingsToSaliency8.pickle", "rb"))
print("concatenating data...")
data = np.concatenate((data1, data2, data3, data4, data5, data6, data7, data8), axis=0)
print("extracting x and y...")
x = data[::2]
y = data[1::2]
del data
print("converting x to np tensor...")
x = np.dstack(x)
x = np.rollaxis(x, -1)
x = np.expand_dims(x, axis=1)
mask = y==-1
print("removing -1s...")
x = x[~mask, :]
y = y[~mask]
print("data loaded.")
x, y = shuffle(x, y)
return x, y
def dummy_load_data(data):
x,y = rand_embed_sentences(data)
return (x,y)
def main():
# Model Hyperparameters
conv_window_size = 300
num_filters = 400
reg = 0.01
dropout = 0.5
# Training parameters
epochs = 25
batch_size = 128
test_train_ratio = 0.2
val_train_ratio = 0.2
"""
#Training with word2vec
x_train, y_train = load_data()
print("training data:", x_train.shape, y_train.shape)
model = build_model((1, x_train.shape[2], x_train.shape[3]), conv_window_size, num_filters, reg, dropout)
history = train(model, x_train, y_train, val_train_ratio, epochs, batch_size)
print("Saving model...")
model.model.save('model.h5')
print("Plotting...")
f, (ax1, ax2) = plt.subplots(2, 1)
ax1.plot(range(1, epochs+1), history.history['val_mean_absolute_error'], 'tab:blue', label="validation MAE")
ax1.plot(range(1, epochs+1), history.history['mean_absolute_error'], 'tab:red', label="training MAE")
ax2.plot(range(1, epochs+1), history.history['loss'], 'tab:orange', label="loss")
ax2.plot(range(1, epochs+1), history.history['val_loss'], 'tab:green', label="validation loss")
ax1.legend()
ax2.legend()
f.savefig('training.png', dpi=300)
plt.show()
print("Done.")
"""
#Training with random init
data = np.array([[0,"This is a sentence for doc0.",0.9], [1,"That is a sentence for doc1!" , 0.8], [2," And this is a sentence for doc2.", 0.6]])
x_train, y_train = dummy_load_data(data)
print("training data:", x_train.shape, y_train.shape)
vocab_size = np.max(x_train)
seq_len = len(x_train[0])
model = build_model((1,1,1), conv_window_size, num_filters, reg, dropout, word2vec = False, max_token = vocab_size, sequence_len = seq_len)
history = train(model, x_train, y_train, val_train_ratio, epochs, batch_size)
print("Saving model...")
model.model.save('model.h5')
print("Plotting...")
f, (ax1, ax2) = plt.subplots(2, 1)
ax1.plot(range(1, epochs+1), history.history['val_mean_absolute_error'], 'tab:blue', label="validation MAE")
ax1.plot(range(1, epochs+1), history.history['mean_absolute_error'], 'tab:red', label="training MAE")
ax2.plot(range(1, epochs+1), history.history['loss'], 'tab:orange', label="loss")
ax2.plot(range(1, epochs+1), history.history['val_loss'], 'tab:green', label="validation loss")
ax1.legend()
ax2.legend()
f.savefig('training.png', dpi=300)
plt.show()
print("Done.")
if __name__ == "__main__":
main()