forked from HyperGAN/hyperchamber
-
Notifications
You must be signed in to change notification settings - Fork 0
/
mnist.py
140 lines (106 loc) · 4.05 KB
/
mnist.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
import hyperchamber as hc
from shared.ops import *
from shared.util import *
import os
import time
import numpy as np
import tensorflow as tf
from tensorflow.python.framework import ops
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("MNIST_data/", one_hot=True)
learning_rates = list(np.linspace(0.0001, 1, num=30))
hc.set("learning_rate", learning_rates)
hidden_layers = [ [], [26], [128], [16, 32], [32,16,8], [16,8,8,4], [64,64]]
hc.set("hidden_layer", hidden_layers)
hc.set("activation", [tf.nn.elu, tf.nn.relu, tf.nn.relu6, tf.tanh, tf.sigmoid, lrelu]);
hc.set("batch_size", 128)
X_DIMS=[28,28]
Y_DIMS=10
def hidden_layers(config, x):
output = tf.reshape(x, [config["batch_size"], X_DIMS[0]*X_DIMS[1]])
for i, layer in enumerate(config['hidden_layer']):
output = linear(output, layer, scope="l"+str(i))
output = config['activation'](output)
return output
def output_layer(config, x):
return linear(x, Y_DIMS)
def create(config):
batch_size = config["batch_size"]
x = tf.placeholder(tf.float32, [batch_size, X_DIMS[0], X_DIMS[1], 1], name="x")
y = tf.placeholder(tf.float32, [batch_size, Y_DIMS], name="y")
hidden = hidden_layers(config, x)
output = output_layer(config, hidden)
loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(output, y), name="loss")
output = tf.nn.softmax(output)
correct_prediction = tf.equal(tf.argmax(output,1), tf.argmax(y,1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
variables = tf.trainable_variables()
optimizer = tf.train.GradientDescentOptimizer(config['learning_rate']).minimize(loss)
set_tensor("x", x)
set_tensor("y", y)
set_tensor("loss", loss)
set_tensor("optimizer", optimizer)
set_tensor("accuracy", accuracy)
def train(sess, config, x_input, y_labels):
x = get_tensor("x")
y = get_tensor("y")
cost = get_tensor("loss")
optimizer = get_tensor("optimizer")
accuracy = get_tensor("accuracy")
_, accuracy, cost = sess.run([optimizer, accuracy, cost], feed_dict={x:x_input, y:y_labels})
#hc.cost(config, cost)
#print("Accuracy %.2f Cost %.2f" % (accuracy, cost))
def test(sess, config, x_input, y_labels):
x = get_tensor("x")
y = get_tensor("y")
cost = get_tensor("loss")
accuracy = get_tensor("accuracy")
accuracy, cost = sess.run([accuracy, cost], feed_dict={x:x_input, y:y_labels})
print("Accuracy %.2f Cost %.2f" % (accuracy, cost))
return accuracy, cost
def epoch(sess, config):
batch_size = config["batch_size"]
n_samples = mnist.train.num_examples
total_batch = int(n_samples / batch_size)
for i in range(total_batch):
x, y = mnist.train.next_batch(batch_size)
x=np.reshape(x, [batch_size, X_DIMS[0], X_DIMS[1], 1])
train(sess, config, x, y)
def test_config(sess, config):
batch_size = config["batch_size"]
n_samples = mnist.test.num_examples
total_batch = int(n_samples / batch_size)
accuracies = []
costs = []
for i in range(total_batch):
x, y = mnist.test.next_batch(batch_size)
x=np.reshape(x, [batch_size, X_DIMS[0], X_DIMS[1], 1])
accuracy, cost = test(sess, config, x, y)
accuracies.append(accuracy)
costs.append(cost)
return accuracies, costs
print("Searching randomly with %d possible configurations." % hc.count_configs())
for i in range(100):
config = hc.random_config()
print("Testing configuration", config)
sess = tf.Session()
graph = create(config)
init = tf.initialize_all_variables()
sess.run(init)
for i in range(10):
epoch(sess, config)
accuracies, costs = test_config(sess, config)
accuracy, cost = np.mean(accuracies), np.mean(costs)
results = {
'accuracy':accuracy,
'cost':cost
}
hc.record(config, results)
ops.reset_default_graph()
sess.close()
def by_accuracy(x):
config,result = x
return 1-result['accuracy']
for config, result in hc.top(by_accuracy):
print("RESULTS")
print(config, result)