/
Model.py
245 lines (178 loc) · 9.31 KB
/
Model.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
import numpy as np
import pandas as pd
import os
from sklearn.cross_validation import train_test_split
from keras.models import Sequential
from keras.models import Model
from keras.layers import Input, Dense, Conv2D, MaxPooling2D, Flatten
from keras.layers.core import Dropout, Activation
from keras.layers import BatchNormalization
import tensorflow as tf
from keras import regularizers
import keras.backend
from keras.optimizers import Adam
from keras.utils import plot_model
from keras.callbacks import TensorBoard,EarlyStopping
from Datasets import Dataset
class Model(object):
def __init__(self, config):
self.config = config
self.data_init()
self.model_init()
def data_init(self):
pass
def model_init(self):
pass
def train_predict(self):
pass
class Model_Multi(Model):
def __init__(self, config):
Model.__init__(self, config)
def data_init(self):
self.dataset = Dataset_Multi(self.config)
self.data = self.dataset.frame_data
self.X = self.data.iloc[:,1]
self.y = self.data.iloc[:,2:]
self.ids_train, self.ids_val, self.y_train, selfy_val = train_test_split(X, y, test_size=0.25, random_state=1)
self.y_train_vect = [self.y_train["cadran_1"], self.y_train["cadran_2"], self.y_train["cadran_3"], self.y_train["cadran_4"]]
self.y_val_vect = [self.y_val["cadran_1"], self.y_val["cadran_2"], self.y_val["cadran_3"], self.y_val["cadran_4"]]
self.X_train = self.dataset.convert_to_arrays(ids_train)
self.X_val = self.dataset.convert_to_arrays(ids_val)
def model_init(self):
model_input = Input((100,246,1))
x = Conv2D(32, (3, 3), padding='same', name='conv2d_hidden_1', kernel_regularizer=regularizers.l2(0.01))(model_input)
x = BatchNormalization()(x)
x = Activation('relu')(x)
x = MaxPooling2D(pool_size=(2, 2), strides=(3, 3),name='maxpool_2d_hidden_1')(x)
x = Dropout(0.30)(x)
x = Conv2D(64, (3, 3), padding='same', name='conv2d_hidden_2', kernel_regularizer=regularizers.l2(0.01))(x)
x = BatchNormalization()(x)
x = Activation('relu')(x)
x = MaxPooling2D(pool_size=(2, 2), strides=(3, 3),name='maxpool_2d_hidden_2')(x)
x = Dropout(0.30)(x)
x = Conv2D(128, (3, 3), padding='same', name='conv2d_hidden_3', kernel_regularizer=regularizers.l2(0.01))(x)
x = BatchNormalization()(x)
x = Activation('relu')(x)
x = MaxPooling2D(pool_size=(2, 2), strides=(3, 3),name='maxpool_2d_hidden_3')(x)
x = Dropout(0.30)(x)
x = Flatten()(x)
x = Dense(256, activation ='relu', kernel_regularizer=regularizers.l2(0.01))(x)
digit1 = (Dense(output_dim =11,activation = 'softmax', name='digit_1'))(x)
digit2 = (Dense(output_dim =11,activation = 'softmax', name='digit_2'))(x)
digit3 = (Dense(output_dim =11,activation = 'softmax', name='digit_3'))(x)
digit4 = (Dense(output_dim =11,activation = 'softmax', name='digit_4'))(x)
outputs = [digit1, digit2, digit3, digit4]
self.model = Model(input = model_input , output = outputs)
self.model._make_predict_function()
def train(self, lr = 1e-3, epochs=100):
optimizer = Adam(lr=lr, decay=lr/10)
self.model.compile(loss="sparse_categorical_crossentropy", optimizer= optimizer, metrics = ['accuracy'])
keras.backend.get_session().run(tf.initialize_all_variables())
history = self.model.fit(self.X_train, self.y_train_vect, batch_size= 50, nb_epoch=100, verbose=1, validation_data=(self.X_val, self.y_val_vect))
def plot_loss(self):
for i in range(1,5):
plt.figure(figsize=[8,6])
plt.plot(history.history['digit_%i_acc' %i],'r',linewidth=0.5)
plt.plot(history.history['val_digit_%i_acc' %i],'b',linewidth=0.5)
plt.legend(['Training Accuracy', 'Validation Accuracy'],fontsize=18)
plt.xlabel('Epochs ',fontsize=16)
plt.ylabel('Accuracy',fontsize=16)
plt.title('Accuracy Curves Digit %i' %i,fontsize=16)
plt.show()
def plot_acc(self):
for i in range(1,5):
plt.figure(figsize=[8,6])
plt.plot(history.history['digit_%i_loss' %i],'r',linewidth=0.5)
plt.plot(history.history['val_digit_%i_loss' %i],'b',linewidth=0.5)
plt.legend(['Training loss', 'Validation Loss'],fontsize=18)
plt.xlabel('Epochs ',fontsize=16)
plt.ylabel('Loss',fontsize=16)
plt.title('Loss Curves Digit %i' %i,fontsize=16)
plt.show()
def predict(self):
preds = self.model.predict(self.X_val)
correct_preds = 0
for i in range(X_val.shape[0]):
pred_list_i = [np.argmax(pred[i]) for pred in y_pred]
val_list_i = y_val.values[i].astype('int')
if np.array_equal(val_list_i, pred_list_i):
correct_preds = correct_preds + 1
print('exact accuracy', correct_preds / X_val.shape[0])
mse = 0
diff = []
for i in range(X_val.shape[0]):
pred_list_i = [np.argmax(pred[i]) for pred in y_pred]
pred_number = 1000* pred_list_i[0] + 100* pred_list_i[1] + 10 * pred_list_i[2] + 1* pred_list_i[3]
val_list_i = y_val.values[i].astype('int')
val_number = 1000* val_list_i[0] + 100* val_list_i[1] + 10 * val_list_i[2] + 1* val_list_i[3]
diff.append(val_number - pred_number)
print('difference label vs. prediction', diff)
def train_predict(self):
self.train(self.config.lr, self.config.epochs)
self.plot_loss()
self.plot_acc()
self.predict()
class Model_Digit(Model):
def __init__(self, config):
Model.__init__(self, config)
def data_init(self):
self.dataset = Dataset_Digit(self.config)
self.data = self.digits_data
self.X = self.data.iloc[:,0]
self.y = self.data.iloc[:,1]
self.ids_train, self.ids_val, self.y_train, self.y_val = train_test_split(self.X, self.y, test_size=0.25, random_state=1)
self.X_train = self.dataset.convert_to_arrays(ids_train)
self.X_val = self.dataset.convert_to_arrays(ids_val)
def model_init(self):
model_input = Input((30,50,1))
x = Conv2D(32, (3, 3), padding='same', name='conv2d_hidden_1', kernel_regularizer=regularizers.l2(0.01))(model_input)
x = BatchNormalization()(x)
x = Activation('relu')(x)
x = MaxPooling2D(pool_size=(2, 2), strides=(3, 3),name='maxpool_2d_hidden_1')(x)
x = Dropout(0.30)(x)
x = Conv2D(63, (3, 3), padding='same', name='conv2d_hidden_2', kernel_regularizer=regularizers.l2(0.01))(x)
x = BatchNormalization()(x)
x = Activation('relu')(x)
x = MaxPooling2D(pool_size=(2, 2), strides=(3, 3),name='maxpool_2d_hidden_2')(x)
x = Dropout(0.30)(x)
x = Conv2D(128, (3, 3), padding='same', name='conv2d_hidden_3', kernel_regularizer=regularizers.l2(0.01))(x)
x = BatchNormalization()(x)
x = Activation('relu')(x)
x = MaxPooling2D(pool_size=(2, 2), strides=(3, 3),name='maxpool_2d_hidden_3')(x)
x = Dropout(0.30)(x)
x = Flatten()(x)
x = Dense(1024, activation ='relu', kernel_regularizer=regularizers.l2(0.01))(x)
output = Dense(output_dim =11,activation = 'softmax', name='output')(x)
self.model = Model(input = model_input , output = output)
self.model._make_predict_function()
def train(self, lr = 1e-3, epochs=100):
optimizer = Adam(lr=lr, decay=lr/10)
self.model.compile(loss="sparse_categorical_crossentropy", optimizer= optimizer, metrics = ['accuracy'])
keras.backend.get_session().run(tf.initialize_all_variables())
history = self.model.fit(self.X_train, self.y_train, batch_size= 32, nb_epoch=30, verbose=1, validation_data=(self.X_val, self.y_val))
def plot_acc(self):
plt.figure(figsize=[8,6])
plt.plot(history.history['acc'],'r',linewidth=0.5)
plt.plot(history.history['val_acc'],'b',linewidth=0.5)
plt.legend(['Training Accuracy', 'Validation Accuracy'],fontsize=18)
plt.xlabel('Epochs ',fontsize=16)
plt.ylabel('Accuracy',fontsize=16)
plt.title('Accuracy Curves Digit',fontsize=16)
plt.show()
def plot_loss(self):
plt.figure(figsize=[8,6])
plt.plot(history.history['loss'],'r',linewidth=0.5)
plt.plot(history.history['val_loss'],'b',linewidth=0.5)
plt.legend(['Training loss', 'Validation Loss'],fontsize=18)
plt.xlabel('Epochs ',fontsize=16)
plt.ylabel('Loss',fontsize=16)
plt.title('Loss Curves Digit',fontsize=16)
plt.show()
def predict(self):
preds = self.model.predict(self.X_val)
correct_preds = 0
def train_predict(self):
self.train(self.config.lr, self.config.epochs)
self.plot_loss()
self.plot_acc()
self.predict()