-
Notifications
You must be signed in to change notification settings - Fork 0
/
cnn_keras.py
282 lines (228 loc) · 10.7 KB
/
cnn_keras.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
279
280
281
282
#!/usr/bin/env python2
# -*- coding: utf-8 -*-
"""
Created on Thu Mar 9 10:38:34 2017
@author: bychkov
"""
import os
import theano
import shutil
import pickle
import platform
import pandas as pd
import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as plt
if platform.node() == 'lm4-fim-mg3qd.pc.helsinki.fi':
plt.ioff()
plt.style.use('ggplot')
#import skimage
import pylab as pl
import numpy as np
#import pandas as pd
import sklearn
from sklearn.metrics import roc_auc_score
#from matplotlib import offsetbox
#from sklearn.feature_extraction import image
import keras
from keras import backend as K
os.environ['KERAS_BACKEND'] = 'theano'
if platform.node() == 'lm4-fim-mg3qd.pc.helsinki.fi':
os.environ['THEANO_FLAGS']='mode=FAST_RUN,device=cpu,floatX=float32'
np.random.seed(1337) # for reproducibility
#np.set_printoptions(precision=5, suppress=True)
#==============================================================================
# Custom model initializer Class:
#==============================================================================
class Builder(object):
def __init__(self, opts, train_data, test_data, train_ids, test_ids):
self.opts = opts
self.test_data = test_data
self.train_data = train_data
self.train_ids = train_ids
self.test_ids = test_ids
# Initialize model architecture:
self.model = opts['zoo_model']
# Compile model:
self.model.name = opts['exp_name']
self.model.compile(loss = opts['loss'], optimizer = opts['optimizer'],metrics=['accuracy'])
# Prepare working directory:
self.wrk_dir = os.path.join(opts['exp_path'], opts['exp_name'])
self.state_dir = os.path.join(self.wrk_dir, 'state')
if self.opts['continue'] and os.path.exists(self.state_dir):
# Continue training existing model:
self.routines = self.EpochRoutine(builder=self, history=True)
curr_epoch = self.routines.history['epoch'][-1]
curr_state = os.path.join(self.state_dir, 'epoch_%s.h5' % str(curr_epoch))
self.model.load_weights( curr_state )
print('\nRestoring the model from: %s' % self.wrk_dir)
print('continue from epoch: %s ' % str(curr_epoch))
else:
# Initialize a new model:
print('\ninitializing a new model...')
if os.path.exists(self.wrk_dir):
shutil.rmtree(self.wrk_dir)
# Create a working directory:
os.makedirs(self.wrk_dir)
os.makedirs(self.state_dir)
# Save model structure:
with open(os.path.join(self.wrk_dir, 'model.yml'), "w") as text_file:
text_file.write( self.model.to_yaml() )
self.routines = self.EpochRoutine(builder=self, history=False)
print('Model initialized.')
# ------------------------------------------------------------------------------
def train(self, epoches, batch, verbose=1):
history = self.model.fit( x = self.train_data[0],
y = self.train_data[1],
batch_size = batch,
nb_epoch = epoches,
verbose = verbose,
validation_split = self.opts['val_split'],
#validation_data = (self.test_data[0], self.test_data[1]),
callbacks = [ self.routines ] )
return history
# ------------------------------------------------------------------------------
# EpochRoutine Class:
# ------------------------------------------------------------------------------
class EpochRoutine(keras.callbacks.Callback):
def __init__(self, builder, history=False, *args, **kwargs):
super(builder.EpochRoutine, self).__init__(*args, **kwargs)
self.opts = builder.opts
self.builder = builder
if not history:
self.history = dict()
self.history['epoch'] = []
self.history['train_loss'] = []
self.history['valid_loss'] = []
self.history['test_loss'] = []
self.history['train_acc'] = []
self.history['valid_acc'] = []
self.history['test_acc'] = []
self.epoch_count = 0
else:
history_path = os.path.join(builder.wrk_dir, 'history.pkl')
self.history = pickle.load(open(history_path, 'rb'))
self.epoch_count = self.history['epoch'][-1]
# ------------------------------------------------------------------------------
def on_epoch_end(self, epoch, logs={}):
epoch += 1
self.epoch_count += 1
# Available entities:
# epoch, logs, self.model, self.params, self.opts, self.history
# Get statistics on test set:
test_eval = self.model.model.evaluate(
self.builder.test_data[0],
self.builder.test_data[1],
batch_size=20
)
# Save training performance (loss):
self.history['epoch'].append( self.epoch_count )
self.history['train_loss'].append( logs.get('loss') )
self.history['valid_loss'].append( logs.get('val_loss') )
self.history['test_loss'].append( test_eval[0] )
self.history['train_acc'].append( logs.get('acc') )
self.history['valid_acc'].append( logs.get('val_acc') )
self.history['test_acc'].append( test_eval[1] )
# Visualize loss at each apoch:
self.plot_loss(y1='train_loss', y2='valid_loss', y3='test_loss')
self.plot_accuracy(y1='train_acc', y2='valid_acc', y3='test_acc')
# Dump model weights & update training history:
if ( self.epoch_count % self.opts['dump_freq'] == 0 or self.epoch_count == 1):
self.dump_state()
self.dump_history()
# ------------------------------------------------------------------------------
def dump_state(self):
self.model.save_weights(
os.path.join(self.builder.state_dir,'epoch_%d.h5' % self.epoch_count) )
def dump_history(self):
path = os.path.join(self.builder.wrk_dir, 'history.pkl')
pickle.dump( self.history, open(path, 'wb') )
def plot_accuracy(self, y1, y2, y3):
file_name = 'Accuracy.png'
# Make a plot:
pl.figure()
pl.title( 'Accuracy (1-error)' )
pl.plot(self.history['epoch'], self.history[y1], label=y1, color='#348ABD')
pl.plot(self.history['epoch'], self.history[y2], label=y2, color='#8EBA42')
pl.plot(self.history['epoch'], self.history[y3], label=y3, color='#BA4252')
pl.ylim(0.0, 1.0)
pl.legend(loc='upper center', ncol=3)
pl.savefig(os.path.join(self.builder.wrk_dir, file_name))
plt.close()
def plot_loss(self, y1, y2, y3):
file_name = 'Loss.png'
# Make a plot:
fig, ax1 = plt.subplots()
ax2 = ax1.twinx()
lns1 = ax1.plot(self.history['epoch'], self.history[y1], label=y1, color='#348ABD')
lns2 = ax2.plot(self.history['epoch'], self.history[y2], label=y2, color='#8EBA42')
lns3 = ax2.plot(self.history['epoch'], self.history[y3], label=y3, color='#BA4252')
ax1.set_ylabel(y1); ax2.set_ylabel('valid/test losses')
# legend:
lns = lns1+lns2+lns3
labs = [l.get_label() for l in lns]
ax1.legend(lns, labs, loc='upper center', ncol=3)
plt.savefig(os.path.join(self.builder.wrk_dir, file_name))
plt.close()
def get_activations(self, x, model, layer, learn_phase = 0):
func = K.function([model.layers[0].input, K.learning_phase()], [ model.layers[layer].output ])
return func([x, learn_phase])
def total_epoch_count(self):
return self.epoch_count
#==============================================================================
# Model
#==============================================================================
import keras.models as models
from keras.layers import InputLayer
from keras.layers.core import Dense, Dropout, Flatten, Reshape, Activation
from keras.layers.convolutional import Convolution2D, MaxPooling2D, UpSampling2D, ZeroPadding2D
from keras.layers.recurrent import SimpleRNN, LSTM, GRU
from keras.layers.normalization import BatchNormalization
from keras.regularizers import l1l2
model = models.Sequential()
model.add( InputLayer(input_shape=(3, 90, 90,), name='InputLayer') )
model.add( Convolution2D(32, 3, 3, activation='relu') )
model.add( MaxPooling2D(pool_size=(2, 2)) )
model.add( Convolution2D(32, 3, 3, activation='relu') )
model.add( MaxPooling2D(pool_size=(2, 2)) )
model.add( Convolution2D(64, 3, 3, activation='relu') )
model.add( MaxPooling2D(pool_size=(2, 2)) )
model.add( Convolution2D(64, 3, 3, activation='relu') )
model.add( MaxPooling2D(pool_size=(2, 2)) )
model.add( Flatten() )
model.add( Dense(256, activation='sigmoid') )
model.add( Dense(1, activation='sigmoid') )
#==============================================================================
# Main code:
#==============================================================================
# Dataset npz-file:
data = np.load('synthetic_imgs/first.npz')
class_names = np.array(['benign','cancer'])
train_x = data['train_x']
train_y = data['train_y']
test_x = data['test_x']
test_y = data['test_y']
#----------------------------------------------------------------------
# Set model & training parameters:
#----------------------------------------------------------------------
opts = dict()
opts['dump_freq'] = 1
opts['zoo_model'] = model
opts['val_split'] = 0.3
opts['loss'] = 'binary_crossentropy'
opts['optimizer'] = 'adadelta'
opts['exp_path'] = '/Users/bychkov/GDD/projects/simu'
opts['exp_name'] = 'first'
opts['continue'] = True
#----------------------------------------------------------------------
# Define the model:
#----------------------------------------------------------------------
network = Builder( opts = opts,
train_data = ( train_x, train_y ),
test_data = ( test_x, test_y ),
train_ids = data['train_id'],
test_ids = data['test_id'] )
history = network.train(epoches=15, batch=25, verbose=1)
#==============================================================================
# Done.
#==============================================================================