-
Notifications
You must be signed in to change notification settings - Fork 47
/
model.py
404 lines (330 loc) · 20.8 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
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
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
import tensorflow as tf
import os
class BigModel:
def __init__(self, args, model_type):
self.learning_rate = 0.001
self.num_steps = args.num_steps
self.batch_size = args.batch_size
self.display_step = args.display_step
self.num_input = 784 # MNIST data input (img shape: 28*28)
self.num_classes = 10
self.dropoutprob = args.dropoutprob
self.checkpoint_dir = args.checkpoint_dir
self.checkpoint_file = "bigmodel"
self.temperature = args.temperature
self.checkpoint_path = os.path.join(self.checkpoint_dir, self.checkpoint_file + ".ckpt")
self.log_dir = os.path.join(args.log_dir, self.checkpoint_file)
self.model_type = model_type
# Store layers weight & bias
self.weights = {
# 5x5 conv, 1 input, 32 outputs
'wc1': tf.Variable(tf.random_normal([5, 5, 1, 32]), name="%s_%s" % (self.model_type, "wc1")),
# 5x5 conv, 32 inputs, 64 outputs
'wc2': tf.Variable(tf.random_normal([5, 5, 32, 64]), name="%s_%s" % (self.model_type, "wc2")),
# fully connected, 7*7*64 inputs, 1024 outputs
'wd1': tf.Variable(tf.random_normal([7 * 7 * 64, 1024]), name="%s_%s" % (self.model_type, "wd1")),
# 1024 inputs, 10 outputs (class prediction)
'out': tf.Variable(tf.random_normal([1024, self.num_classes]), name="%s_%s" % (self.model_type, "out"))
}
self.biases = {
'bc1': tf.Variable(tf.random_normal([32]), name="%s_%s" % (self.model_type, "bc1")),
'bc2': tf.Variable(tf.random_normal([64]), name="%s_%s" % (self.model_type, "bc2")),
'bd1': tf.Variable(tf.random_normal([1024]), name="%s_%s" % (self.model_type, "bd1")),
'out': tf.Variable(tf.random_normal([self.num_classes]), name="%s_%s" % (self.model_type, "out"))
}
self.build_model()
self.saver = tf.train.Saver()
def conv2d(self, x, W, b, strides=1):
# Conv2D wrapper, with bias and relu activation
with tf.name_scope("%sconv2d" % (self.model_type)), tf.variable_scope("%sconv2d" % (self.model_type)):
x = tf.nn.conv2d(x, W, strides=[1, strides, strides, 1], padding='SAME')
x = tf.nn.bias_add(x, b)
return tf.nn.relu(x)
def maxpool2d(self, x, k=2):
# MaxPool2D wrapper
with tf.name_scope("%smaxpool2d" % (self.model_type)), tf.variable_scope("%smaxpool2d" % (self.model_type)):
return tf.nn.max_pool(x, ksize=[1, k, k, 1], strides=[1, k, k, 1],
padding='SAME')
# Create model
def build_model(self):
self.X = tf.placeholder(tf.float32, [None, self.num_input], name="%s_%s" % (self.model_type, "xinput"))
self.Y = tf.placeholder(tf.float32, [None, self.num_classes], name="%s_%s" % (self.model_type, "yinput"))
self.keep_prob = tf.placeholder(tf.float32,
name="%s_%s" % (self.model_type, "dropoutprob")) # dropout (keep probability)
self.softmax_temperature = tf.placeholder(tf.float32, name="%s_%s" % (self.model_type, "softmaxtemp"))
# MNIST data input is a 1-D vector of 784 features (28*28 pixels)
# Reshape to match picture format [Height x Width x Channel]
# Tensor input become 4-D: [Batch Size, Height, Width, Channel]
with tf.name_scope("%sinputreshape" % (self.model_type)), tf.variable_scope(
"%sinputreshape" % (self.model_type)):
x = tf.reshape(self.X, shape=[-1, 28, 28, 1])
# Convolution Layer
with tf.name_scope("%sconvmaxpool" % (self.model_type)), tf.variable_scope("%sconvmaxpool" % (self.model_type)):
conv1 = self.conv2d(x, self.weights['wc1'], self.biases['bc1'])
# Max Pooling (down-sampling)
conv1 = self.maxpool2d(conv1, k=2)
# Convolution Layer
conv2 = self.conv2d(conv1, self.weights['wc2'], self.biases['bc2'])
# Max Pooling (down-sampling)
conv2 = self.maxpool2d(conv2, k=2)
# Fully connected layer
# Reshape conv2 output to fit fully connected layer input
with tf.name_scope("%sfclayer" % (self.model_type)), tf.variable_scope("%sfclayer" % (self.model_type)):
fc1 = tf.reshape(conv2, [-1, self.weights['wd1'].get_shape().as_list()[0]])
fc1 = tf.add(tf.matmul(fc1, self.weights['wd1']), self.biases['bd1'])
fc1 = tf.nn.relu(fc1)
# Apply Dropout
fc1 = tf.nn.dropout(fc1, self.keep_prob)
# Output, class prediction
logits = tf.add(tf.matmul(fc1, self.weights['out']), self.biases['out']) / self.softmax_temperature
with tf.name_scope("%sprediction" % (self.model_type)), tf.variable_scope("%sprediction" % (self.model_type)):
self.prediction = tf.nn.softmax(logits)
# Evaluate model
correct_pred = tf.equal(tf.argmax(self.prediction, 1), tf.argmax(self.Y, 1))
self.accuracy = tf.reduce_mean(tf.cast(correct_pred, tf.float32))
with tf.name_scope("%soptimization" % (self.model_type)), tf.variable_scope(
"%soptimization" % (self.model_type)):
# Define loss and optimizer
self.loss_op = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(
logits=logits, labels=self.Y))
optimizer = tf.train.AdamOptimizer(learning_rate=self.learning_rate)
self.train_op = optimizer.minimize(self.loss_op)
with tf.name_scope("%ssummarization" % (self.model_type)), tf.variable_scope(
"%ssummarization" % (self.model_type)):
tf.summary.scalar("loss", self.loss_op)
# Create a summary to monitor accuracy tensor
tf.summary.scalar("accuracy", self.accuracy)
for var in tf.trainable_variables():
tf.summary.histogram(var.name, var)
# Merge all summaries into a single op
# If using TF 1.6 or above, simply use the following merge_all function
# which supports scoping
# self.merged_summary_op = tf.summary.merge_all(scope=self.model_type)
# Explicitly using scoping for TF versions below 1.6
def mymergingfunction(scope_str):
with tf.name_scope("%s_%s" % (self.model_type, "summarymerger")), tf.variable_scope(
"%s_%s" % (self.model_type, "summarymerger")):
from tensorflow.python.framework import ops as _ops
key = _ops.GraphKeys.SUMMARIES
summary_ops = _ops.get_collection(key, scope=scope_str)
if not summary_ops:
return None
else:
return tf.summary.merge(summary_ops)
self.merged_summary_op = mymergingfunction(self.model_type)
def start_session(self):
self.sess = tf.Session()
def close_session(self):
self.sess.close()
def train(self, dataset):
# Initialize the variables (i.e. assign their default value)
self.sess.run(tf.global_variables_initializer())
print("Starting Training")
train_data = dataset.get_train_data()
train_summary_writer = tf.summary.FileWriter(self.log_dir, graph=self.sess.graph)
max_accuracy = 0
for step in range(1, self.num_steps + 1):
batch_x, batch_y = train_data.next_batch(self.batch_size)
_, summary = self.sess.run([self.train_op, self.merged_summary_op],
feed_dict={self.X: batch_x, self.Y: batch_y, self.keep_prob: self.dropoutprob,
self.softmax_temperature: self.temperature})
if (step % self.display_step) == 0 or step == 1:
# Calculate Validation loss and accuracy
validation_x, validation_y = dataset.get_validation_data()
loss, acc = self.sess.run([self.loss_op, self.accuracy], feed_dict={self.X: validation_x,
self.Y: validation_y,
self.keep_prob: 1.0,
self.softmax_temperature: 1.0})
if acc > max_accuracy:
save_path = self.saver.save(self.sess, self.checkpoint_path)
print("Model Checkpointed to %s " % (save_path))
print("Step " + str(step) + ", Validation Loss= " + "{:.4f}".format(
loss) + ", Validation Accuracy= " + "{:.3f}".format(acc))
else:
# Final Evaluation and checkpointing before training ends
validation_x, validation_y = dataset.get_validation_data()
loss, acc = self.sess.run([self.loss_op, self.accuracy], feed_dict={self.X: validation_x,
self.Y: validation_y,
self.keep_prob: 1.0,
self.softmax_temperature: 1.0})
if acc > max_accuracy:
save_path = self.saver.save(self.sess, self.checkpoint_path)
print("Model Checkpointed to %s " % (save_path))
train_summary_writer.close()
print("Optimization Finished!")
def predict(self, data_X, temperature=1.0):
return self.sess.run(self.prediction,
feed_dict={self.X: data_X, self.keep_prob: 1.0, self.softmax_temperature: temperature})
def run_inference(self, dataset):
test_images, test_labels = dataset.get_test_data()
print("Testing Accuracy:", self.sess.run(self.accuracy, feed_dict={self.X: test_images,
self.Y: test_labels,
self.keep_prob: 1.0,
self.softmax_temperature: 1.0
}))
def load_model_from_file(self, load_path):
ckpt = tf.train.get_checkpoint_state(load_path)
if ckpt and tf.train.checkpoint_exists(ckpt.model_checkpoint_path):
print("Reading model parameters from %s" % ckpt.model_checkpoint_path)
self.saver.restore(self.sess, ckpt.model_checkpoint_path)
else:
print("Created model with fresh parameters.")
self.sess.run(tf.global_variables_initializer())
class SmallModel:
def __init__(self, args, model_type):
self.learning_rate = 0.001
self.num_steps = args.num_steps
self.batch_size = args.batch_size
self.display_step = args.display_step
self.n_hidden_1 = 256 # 1st layer number of neurons
self.n_hidden_2 = 256 # 2nd layer number of neurons
self.num_input = 784 # MNIST data input (img shape: 28*28)
self.num_classes = 10
self.temperature = args.temperature
self.checkpoint_dir = args.checkpoint_dir
self.checkpoint_file = "smallmodel"
self.checkpoint_path = os.path.join(self.checkpoint_dir, self.checkpoint_file)
self.max_checkpoint_path = os.path.join(self.checkpoint_dir, self.checkpoint_file + "max")
self.log_dir = os.path.join(args.log_dir, self.checkpoint_file)
self.model_type = model_type
self.weights = {
'h1': tf.Variable(tf.random_normal([self.num_input, self.n_hidden_1]),
name="%s_%s" % (self.model_type, "h1")),
'h2': tf.Variable(tf.random_normal([self.n_hidden_1, self.n_hidden_2]),
name="%s_%s" % (self.model_type, "h2")),
'out': tf.Variable(tf.random_normal([self.n_hidden_2, self.num_classes]),
name="%s_%s" % (self.model_type, "out")),
'linear': tf.Variable(tf.random_normal([self.num_input, self.num_classes]),
name="%s_%s" % (self.model_type, "linear"))
}
self.biases = {
'b1': tf.Variable(tf.random_normal([self.n_hidden_1]), name="%s_%s" % (self.model_type, "b1")),
'b2': tf.Variable(tf.random_normal([self.n_hidden_2]), name="%s_%s" % (self.model_type, "b2")),
'out': tf.Variable(tf.random_normal([self.num_classes]), name="%s_%s" % (self.model_type, "out")),
'linear': tf.Variable(tf.random_normal([self.num_classes]), name="%s_%s" % (self.model_type, "linear"))
}
self.build_model()
self.saver = tf.train.Saver()
# Create model
def build_model(self):
self.X = tf.placeholder(tf.float32, [None, self.num_input], name="%s_%s" % (self.model_type, "xinput"))
self.Y = tf.placeholder(tf.float32, [None, self.num_classes], name="%s_%s" % (self.model_type, "yinput"))
self.flag = tf.placeholder(tf.bool, None, name="%s_%s" % (self.model_type, "flag"))
self.soft_Y = tf.placeholder(tf.float32, [None, self.num_classes], name="%s_%s" % (self.model_type, "softy"))
self.softmax_temperature = tf.placeholder(tf.float32, name="%s_%s" % (self.model_type, "softmaxtemperature"))
with tf.name_scope("%sfclayer" % (self.model_type)), tf.variable_scope("%sfclayer" % (self.model_type)):
# Hidden fully connected layer with 256 neurons
# layer_1 = tf.add(tf.matmul(self.X, self.weights['h1']), self.biases['b1'])
# # Hidden fully connected layer with 256 neurons
# layer_2 = tf.add(tf.matmul(layer_1, self.weights['h2']), self.biases['b2'])
# # Output fully connected layer with a neuron for each class
# logits = (tf.matmul(layer_2, self.weights['out']) + self.biases['out'])
logits = tf.add(tf.matmul(self.X, self.weights['linear']), self.biases['linear'])
with tf.name_scope("%sprediction" % (self.model_type)), tf.variable_scope("%sprediction" % (self.model_type)):
self.prediction = tf.nn.softmax(logits)
self.correct_pred = tf.equal(tf.argmax(logits, 1), tf.argmax(self.Y, 1))
self.accuracy = tf.reduce_mean(tf.cast(self.correct_pred, tf.float32))
with tf.name_scope("%soptimization" % (self.model_type)), tf.variable_scope(
"%soptimization" % (self.model_type)):
# Define loss and optimizer
self.loss_op_standard = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(
logits=logits, labels=self.Y))
self.total_loss = self.loss_op_standard
self.loss_op_soft = tf.cond(self.flag,
true_fn=lambda: tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(
logits=logits / self.softmax_temperature, labels=self.soft_Y)),
false_fn=lambda: 0.0)
self.total_loss += tf.square(self.softmax_temperature) * self.loss_op_soft
# optimizer = tf.train.AdamOptimizer(learning_rate=self.learning_rate)
# optimizer = tf.train.GradientDescentOptimizer(0.05)
optimizer = tf.train.AdamOptimizer(learning_rate=self.learning_rate)
self.train_op = optimizer.minimize(self.total_loss)
with tf.name_scope("%ssummarization" % (self.model_type)), tf.variable_scope(
"%ssummarization" % (self.model_type)):
tf.summary.scalar("loss_op_standard", self.loss_op_standard)
tf.summary.scalar("total_loss", self.total_loss)
# Create a summary to monitor accuracy tensor
tf.summary.scalar("accuracy", self.accuracy)
for var in tf.trainable_variables():
tf.summary.histogram(var.name, var)
# Merge all summaries into a single op
# If using TF 1.6 or above, simply use the following merge_all function
# which supports scoping
# self.merged_summary_op = tf.summary.merge_all(scope=self.model_type)
# Explicitly using scoping for TF versions below 1.6
def mymergingfunction(scope_str):
with tf.name_scope("%s_%s" % (self.model_type, "summarymerger")), tf.variable_scope(
"%s_%s" % (self.model_type, "summarymerger")):
from tensorflow.python.framework import ops as _ops
key = _ops.GraphKeys.SUMMARIES
summary_ops = _ops.get_collection(key, scope=scope_str)
if not summary_ops:
return None
else:
return tf.summary.merge(summary_ops)
self.merged_summary_op = mymergingfunction(self.model_type)
def start_session(self):
self.sess = tf.Session()
def close_session(self):
self.sess.close()
def train(self, dataset, teacher_model=None):
teacher_flag = False
if teacher_model is not None:
teacher_flag = True
# Initialize the variables (i.e. assign their default value)
self.sess.run(tf.global_variables_initializer())
train_data = dataset.get_train_data()
train_summary_writer = tf.summary.FileWriter(self.log_dir, graph=self.sess.graph)
max_accuracy = 0
print("Starting Training")
def dev_step():
validation_x, validation_y = dataset.get_validation_data()
loss, acc = self.sess.run([self.loss_op_standard, self.accuracy], feed_dict={self.X: validation_x,
self.Y: validation_y,
# self.soft_Y: validation_y,
self.flag: False,
self.softmax_temperature: 1.0})
if acc > max_accuracy:
save_path = self.saver.save(self.sess, self.checkpoint_path)
print("Model Checkpointed to %s " % (save_path))
print("Step " + str(step) + ", Validation Loss= " + "{:.4f}".format(
loss) + ", Validation Accuracy= " + "{:.3f}".format(acc))
for step in range(1, self.num_steps + 1):
batch_x, batch_y = train_data.next_batch(self.batch_size)
soft_targets = batch_y
if teacher_flag:
soft_targets = teacher_model.predict(batch_x, self.temperature)
# self.sess.run(self.train_op,
_, summary = self.sess.run([self.train_op, self.merged_summary_op],
feed_dict={self.X: batch_x,
self.Y: batch_y,
self.soft_Y: soft_targets,
self.flag: teacher_flag,
self.softmax_temperature: self.temperature}
)
if (step % self.display_step) == 0 or step == 1:
dev_step()
else:
# Final Evaluation and checkpointing before training ends
dev_step()
train_summary_writer.close()
print("Optimization Finished!")
def predict(self, data_X, temperature=1.0):
return self.sess.run(self.prediction,
feed_dict={self.X: data_X, self.flag: False, self.softmax_temperature: temperature})
def run_inference(self, dataset):
test_images, test_labels = dataset.get_test_data()
print("Testing Accuracy:", self.sess.run(self.accuracy, feed_dict={self.X: test_images,
self.Y: test_labels,
# self.soft_Y: test_labels,
self.flag: False,
self.softmax_temperature: 1.0
}))
def load_model_from_file(self, load_path):
ckpt = tf.train.get_checkpoint_state(load_path)
if ckpt and tf.train.checkpoint_exists(ckpt.model_checkpoint_path):
print("Reading model parameters from %s" % ckpt.model_checkpoint_path)
self.saver.restore(self.sess, ckpt.model_checkpoint_path)
else:
print("Created model with fresh parameters.")
self.sess.run(tf.global_variables_initializer())