/
model.py
167 lines (117 loc) · 7.41 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
import json
import argparse
import numpy as np
import tensorflow as tf
import matplotlib.pyplot as plt
from tensorflow.examples.tutorials.mnist import input_data
from distill_optimizer import DistillOptimizer
import utils as U
class Model(object):
def __init__(self, network_type):
self.graph = tf.Graph()
self.session = U.get_session(self.graph)
self.config = json.load(open("config.json"))
self.network_type = network_type
self.get_data()
self.create_placeholders()
self.build_model()
def get_data(self):
self.mnist = input_data.read_data_sets('MNIST_data', one_hot=False, seed=self.config["seed"])
self.num_itr = self.mnist.train.num_examples // self.config["batch_size"]
def create_placeholders(self):
with self.graph.as_default():
self.x = tf.placeholder(tf.float32, shape=(None, self.config["input_dim"]), name="x")
self.y = tf.placeholder(tf.int32, shape=(None,), name="y")
self.probs_placeholder = tf.placeholder(tf.float32, shape=(None, self.config["output_dim"]), name="probs")
self.keep_prob_visible_unit = tf.placeholder_with_default(1.0, shape=None, name="keep_prob_visible_unit")
self.keep_prob_hidden_unit = tf.placeholder_with_default(1.0, shape=None, name="keep_prob_hidden_unit")
self.learning_rate = tf.placeholder(tf.float32, shape=None, name="learning_rate")
self.momentum = tf.placeholder(tf.float32, shape=None, name="momentum")
self.max_norm = tf.placeholder(tf.float32, shape=None, name="max_norm")
self.temperature_ph = tf.placeholder(tf.float32, shape=None, name="temperature")
def model_architecture(self, num_hidden_units, use_probs):
with self.graph.as_default():
U.set_random_seed()
w1 = tf.Variable(tf.random_normal(shape=(self.config["input_dim"], num_hidden_units), mean=0.0, stddev=0.01), name="w1")
b1 = tf.Variable(tf.zeros(shape=(num_hidden_units,), dtype=tf.float32), name="b1")
w2 = tf.Variable(tf.random_normal(shape=(num_hidden_units, num_hidden_units), mean=0.0, stddev=0.01), name="w2")
b2 = tf.Variable(tf.zeros(shape=(num_hidden_units,), dtype=tf.float32), name="b2")
w3 = tf.Variable(tf.random_normal(shape=(num_hidden_units, self.config["output_dim"]), mean=0.0, stddev=0.01), name="w3")
b3 = tf.Variable(tf.zeros(shape=(self.config["output_dim"],), dtype=tf.float32), name="b3")
x_dropout = tf.nn.dropout(self.x, self.keep_prob_visible_unit)
h1 = tf.nn.relu(tf.matmul(x_dropout, w1) + b1)
h1_dropout = tf.nn.dropout(h1, self.keep_prob_hidden_unit)
h2 = tf.nn.relu(tf.matmul(h1_dropout, w2) + b2)
h2_dropout = tf.nn.dropout(h2, self.keep_prob_hidden_unit)
self.logits = tf.matmul(h2_dropout, w3) + b3
self.probs = tf.nn.softmax(self.logits / self.temperature_ph)
if use_probs:
self.loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits_v2(labels=self.probs_placeholder,
logits=self.logits / self.temperature_ph))
else:
self.loss = tf.reduce_mean(
tf.nn.sparse_softmax_cross_entropy_with_logits(labels=self.y, logits=self.logits))
self.pred_ = tf.argmax(self.logits, axis=1, output_type=tf.int32)
self.missclassification_error = tf.reduce_sum(tf.cast(tf.not_equal(self.pred_, self.y), tf.float32))
self.train_op = self.get_optimizer().minimize(self.loss)
self.initialize = tf.global_variables_initializer()
self.saver = tf.train.Saver()
def get_optimizer(self):
optimizer_name = self.config[self.network_type]["optimizer"]
if optimizer_name == "DistillOptimizer":
return U.get_optimizer(optimizer_name)(self.learning_rate, self.momentum, self.max_norm)
elif optimizer_name == "Momentum":
return U.get_optimizer(optimizer_name)(self.learning_rate, self.momentum)
def build_model(self):
self.model_architecture(self.config[self.network_type]["num_hidden_units"],
self.config[self.network_type]["use_probs"])
def train(self):
self.training_miss_classifications = []
self.testing_miss_classifications = []
self.session.run(self.initialize)
lr = self.config[self.network_type]["initial_learning_rate"]
x_test, y_test = self.mnist.test.images, self.mnist.test.labels
for epoch in range(self.config["num_epoch"]):
miss_classes = []
mu = U.get_momentum(epoch)
for itr in range(self.num_itr):
x_train, y_train = self.mnist.train.next_batch(self.config["batch_size"])
if self.config[self.network_type]["jitter_images"]:
x_train = U.jitter_images(x_train)
_, miss_class, logits_val = self.session.run([self.train_op, self.missclassification_error, self.logits],
feed_dict={self.x: x_train,
self.y: y_train,
self.keep_prob_hidden_unit: self.config[self.network_type]["keep_prob_hidden_unit"],
self.keep_prob_visible_unit: self.config[self.network_type]["keep_prob_visible_unit"],
self.learning_rate: lr,
self.momentum: mu,
self.max_norm: self.config["max_norm_val"]})
miss_classes.append(miss_class)
lr *= self.config["learning_rate_decay"]
if (epoch + 1) % self.config["show_every"] == 0:
test_miss_class = self.session.run(self.missclassification_error,
feed_dict={self.x: x_test,
self.y: y_test})
print("epoc: {0}, train_miss_class: {1:0.0f}, test_miss_class: {2:0.0f}"
.format(epoch, np.sum(miss_classes), test_miss_class))
self.training_miss_classifications.append(np.sum(miss_classes))
self.testing_miss_classifications.append(test_miss_class)
U.save_data(self.training_miss_classifications, self.testing_miss_classifications, self.network_type)
U.plot_results(self.training_miss_classifications, self.testing_miss_classifications, self.network_type)
def save(self):
names = ["probs", "logits", "inputs", "temperature_ph"]
vals = [self.probs, self.logits, self.x, self.temperature_ph]
with self.graph.as_default():
U.add_to_collection(names, vals)
self.saver.save(self.session, "checkpoint/" + self.network_type)
if __name__ == "__main__":
parser = argparse.ArgumentParser(description="Train a model")
parser.add_argument("-n", default="ensemble", choices=["ensemble", "small", "distill"], dest="model_name")
args = parser.parse_args()
print("--" * 30)
print("Training a --- {} --- model".format(args.model_name))
print("--" * 30)
model = Model(args.model_name)
model.train()
model.save()
#####