-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain.py
executable file
·278 lines (236 loc) · 10.4 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
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
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
# python train.py --dataset mnist --model vgg16 --reshape '(32,32)' --batch_size 128 --epoch 10 --learning_rate 0.01 --dropout_rate 0.2 --activation_ch softmax --optimizer_ch sgd
# python train.py --dataset cifar --model vgg16 --reshape '(32,32)' --batch_size 128 --epoch 10 --learning_rate 0.01 --dropout_rate 0.2 --activation_ch softmax --optimizer_ch sgd
# python train.py --dataset mnist --model googlenet --reshape '(224,224)' --batch_size 128 --epoch 10 --learning_rate 0.001 --dropout_rate 0.2 --activation_ch softmax --optimizer_ch sgd
# python train.py --dataset cifar --model googlenet --reshape '(224,224)' --batch_size 128 --epoch 10 --learning_rate 0.001 --dropout_rate 0.2 --activation_ch softmax --optimizer_ch sgd
import os,time
from ast import literal_eval
import argparse
import matplotlib.pyplot as plt
import pandas as pd
# os.environ["CUDA_VISIBLE_DEVICES"]="1"
import numpy as np
from sklearn.metrics import classification_report, confusion_matrix
import scikitplot as skplt
import keras
from tensorflow.keras.utils import plot_model
from processData import dataset_selection
from networks import googlenet_layers
from networks import evaluate_test
from networks import select_optimiser
if not os.path.exists("model"):
os.makedirs("model")
ap = argparse.ArgumentParser()
ap.add_argument("-d", "--dataset", required=True,
help="dataset name, mnist or cifar10")
ap.add_argument("-m", "--model", required=True,
help="model to train, vgg16 or googlenet")
ap.add_argument("-r", "--reshape", required=True, type=str, default="(224,224)",
help="minimum reshape size for input image")
ap.add_argument("-b", "--batch_size", required=True, type=int,
help="batch_size")
ap.add_argument("-e", "--epochs", required=True, type=int,
help="number of epochs")
ap.add_argument("-l", "--learning_rate", required=True, type=float,
help="learning rate")
ap.add_argument("-dr", "--dropout_rate", required=True, type=float,
help="dropout_rate")
ap.add_argument("-a", "--activation_ch", required=True,
help="activation choice")
ap.add_argument("-o", "--optimizer_ch", required=True,
help="optimizer choice")
args = vars(ap.parse_args())
model_output_csv = pd.DataFrame()
dataset_name = args['dataset']
model_name = args['model']
batch_size = args['batch_size']
buffer_size = 10000
epochs = args['epochs']
learning_rate = args['learning_rate']
dropout_rate = args['dropout_rate']
activation_ch = args['activation_ch']
opt = args['optimizer_ch']
num_classes = 10
input_dims = literal_eval(args['reshape'])
data = dataset_selection(dataset_name,final_shape=input_dims)
print("Train Dataset",data.x_train.shape,data.y_train.shape)
print("Validation Dataset",data.x_val.shape, data.y_val.shape)
print("Test Dataset",data.x_test.shape,data.y_test.shape)
for i in range(9):
plt.subplot(330 + 1 + i)
plt.imshow(data.x_train[i], cmap="gray")
plt.show()
plt.savefig('data.png')
input_shape = data.x_train.shape[1:]
print("Creating network with Paramters")
print()
print("batch_size = ", batch_size)
print("buffer_size = ", buffer_size)
print("epochs = ", epochs)
print("learning_rate = ", learning_rate)
print("dropout_rate = ", dropout_rate)
print("activation_ch = ", activation_ch)
print("opt = ", opt)
print("num_classes = ", num_classes)
print("input_dims = ", str(input_shape))
if model_name == "vgg16":
inputs = keras.layers.Input(shape=input_shape)
model_output = vgg16_layers(inputs,num_classes,dropout_rate,activation_ch)
opt_type = select_optimiser(opt,learning_rate)
model = keras.models.Model(inputs=inputs, outputs=model_output)
model.compile(loss=keras.losses.categorical_crossentropy,
optimizer=opt_type,
metrics=['accuracy'])
plot_model(model, to_file='vgg16.png')
print(model.summary())
start = time.time()
history = model.fit(data.x_train, data.y_train,
batch_size=batch_size,
epochs=epochs,
verbose=1,
validation_data=(data.x_val, data.y_val))
end = time.time()
total_exec_time = end - start
print("Training Completed in ", str(total_exec_time/60), "mins")
print("Saving Network and weights")
model.save_weights('model/vgg16_wt.h5')
print("Evaluating network performance on Test Data")
score = model.evaluate(data.x_test, data.y_test, verbose=1,batch_size=batch_size)
print('Test loss:', score[0])
print('Test accuracy:', score[1])
predictions = model.predict(data.x_test, verbose=0)
# print(confusion_matrix(np.argmax(data.y_test,axis=1),np.argmax(predictions,axis=1), normalize='true'))
skplt.metrics.plot_confusion_matrix(np.argmax(data.y_test,axis=1),np.argmax(predictions,axis=1), normalize=True,figsize = (7,7))
print(classification_report(np.argmax(data.y_test,axis=1),np.argmax(predictions,axis=1)))
fig = plt.figure(figsize=(12,5))
subfig = fig.add_subplot(121)
subfig.plot(history.history['accuracy'])
subfig.plot(history.history['val_accuracy'])
subfig.set_title('model accuracy')
subfig.set_ylabel('accuracy')
subfig.set_xlabel('epoch')
subfig.legend(['train', 'val'], loc='upper left')
subfig = fig.add_subplot(122)
subfig.plot(history.history['loss'])
subfig.plot(history.history['val_loss'])
subfig.set_title('model loss')
subfig.set_ylabel('loss')
subfig.set_xlabel('epoch')
subfig.legend(['train', 'val'], loc='upper left')
plt.show()
plt.savefig('vgg_acc_loss.png')
p = evaluate_test(data.y_test,predictions)
print("Error Analysis")
print("True: {}".format(np.argmax(data.y_test[p[0][0:5]], axis =1)))
print("classified as: {}".format(np.argmax(predictions[p[0][0:5]], axis=1)))
print("Error Cases")
print("True: {}".format(np.argmax(data.y_test[p[1][6:11]], axis =1)))
print("classified as: {}".format(np.argmax(predictions[p[1][6:11]], axis=1)))
print('Saving Network Results')
hist = {}
hist['model_name'] = model_name
hist['dataset'] = dataset_name
hist['dropout_rate'] = dropout_rate
hist['activation'] = activation_ch
hist['lr_rate'] = learning_rate
hist['epochs'] = epochs
hist['opt'] = opt
hist['input_dims'] = data.x_train.shape[1:]
hist = dict(hist, **history.history)
hist['test_loss'] = score[0]
hist['test_accuracy'] = score[1]
np.save('model.txt',hist)
model_output_csv = model_output_csv.append(hist, ignore_index=True)
model_output_csv.to_csv("vgg_results.csv", index=False)
print("Releasing GPU Memory")
keras.backend.clear_session()
if model_name == 'googlenet':
inputs = keras.layers.Input(shape=input_shape)
(main, aux1, aux2) = googlenet_layers(inputs,num_classes,dropout_rate,activation_ch)
opt_type = select_optimiser(opt,learning_rate)
model = keras.models.Model(inputs=inputs, outputs=[main, aux1, aux2])
model.compile(loss=keras.losses.categorical_crossentropy,
optimizer=opt_type,
metrics=['accuracy'])
plot_model(model, to_file='inception.png')
print(model.summary())
start = time.time()
history = model.fit(data.x_train, [data.y_train,data.y_train,data.y_train],
batch_size=batch_size,
epochs=epochs,
verbose=1,
validation_data=(data.x_val, [data.y_val,data.y_val,data.y_val]))
end = time.time()
total_exec_time = end - start
print("Training Completed in ", str(total_exec_time/60), "mins")
print("Saving Network and weights")
weight_path = 'model/inception_wt.h5'
model.save_weights(weight_path)
print("Evaluating network performance on Test Data")
score = model.evaluate(data.x_test, [data.y_test,data.y_test,data.y_test], verbose=1,batch_size=batch_size)
print('Test loss:', score[0])
print('Test accuracy:', score[4])
predictions = model.predict(data.x_test, verbose=0)
# print(confusion_matrix(np.argmax(data.y_test,axis=1),np.argmax(predictions[0],axis=1), normalize='true'))
skplt.metrics.plot_confusion_matrix(np.argmax(data.y_test,axis=1),np.argmax(predictions[0],axis=1), normalize=True,figsize = (7,7))
print(classification_report(np.argmax(data.y_test,axis=1),np.argmax(predictions[0],axis=1)))
fig = plt.figure(figsize=(12,5))
subfig = fig.add_subplot(121)
subfig.plot(history.history['main_accuracy'])
subfig.plot(history.history['val_main_accuracy'])
subfig.plot(history.history['aux1_accuracy'])
subfig.plot(history.history['val_aux1_accuracy'])
subfig.plot(history.history['aux2_accuracy'])
subfig.plot(history.history['val_aux2_accuracy'])
subfig.set_title('model accuracy')
subfig.set_ylabel('accuracy')
subfig.set_xlabel('epoch')
subfig.legend(['main_accuracy', 'val_main_accuracy','aux1_accuracy','val_aux1_accuracy','aux2_accuracy','val_aux2_accuracy'], loc='upper left')
subfig = fig.add_subplot(122)
# summarize history for loss
subfig.plot(history.history['main_loss'])
subfig.plot(history.history['val_main_loss'])
subfig.plot(history.history['aux1_loss'])
subfig.plot(history.history['val_aux1_loss'])
subfig.plot(history.history['aux2_loss'])
subfig.plot(history.history['val_aux2_loss'])
subfig.set_title('model loss')
subfig.set_ylabel('loss')
subfig.set_xlabel('epoch')
subfig.legend(['main_loss', 'val_main_loss','aux1_loss','val_aux1_loss','aux2_loss','val_aux2_loss'], loc='upper left')
plt.show()
plt.savefig('googlenet_acc_loss.png')
# loss
fig = plt.figure(figsize=(12,5))
subfig = fig.add_subplot(121)
subfig.plot(history.history['loss'])
subfig.plot(history.history['val_loss'])
subfig.set_title('model loss')
subfig.set_ylabel('loss')
subfig.set_xlabel('epoch')
subfig.legend(['train', 'val'], loc='upper left')
plt.show()
plt.savefig('googlenet_loss.png')
p = evaluate_test(data.y_test,predictions[0])
print("True: {}".format(np.argmax(data.y_test[p[0][0:5]], axis =1)))
print("classified as: {}".format(np.argmax(predictions[0][p[0][0:5]], axis=1)))
print("Error Cases")
print("True: {}".format(np.argmax(data.y_test[p[1][6:11]], axis =1)))
print("classified as: {}".format(np.argmax(predictions[0][p[1][6:11]], axis=1)))
print('Saving Network Results')
hist = {}
hist['model_name'] = 'inception'
hist['dataset'] = dataset_name
hist['dropout_rate'] = dropout_rate
hist['activation'] = activation_ch
hist['lr_rate'] = learning_rate
hist['epochs'] = epochs
hist['opt'] = opt
hist['input_dims'] = data.x_train.shape[1:]
hist = dict(hist, **history.history)
hist['test_loss'] = score[0]
hist['test_accuracy'] = score[4]
np.save('model.txt',hist)
model_output_csv = model_output_csv.append(hist, ignore_index=True)
model_output_csv.to_csv("inception_results.csv", index=False)
print("Releasing GPU Memory")
keras.backend.clear_session()