-
Notifications
You must be signed in to change notification settings - Fork 0
/
cnn.py
257 lines (221 loc) · 10.5 KB
/
cnn.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
from __future__ import print_function
from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("/tmp/data/", one_hot=True)
import tensorflow as tf
import numpy as np
posdigit = 8
negdigit = 3
labelled_unlabelled_split_barrier = 4000
theta_p = 0.5
theta_n = 1-theta_p
train_data_x = mnist.train.images
train_labels = mnist.train.labels
train_data_x,unlabelled_data_x = train_data_x[:labelled_unlabelled_split_barrier],train_data_x[labelled_unlabelled_split_barrier:]
train_labels,unlabelled_labels = train_labels[:labelled_unlabelled_split_barrier],train_labels[labelled_unlabelled_split_barrier:]
pos_x = train_data_x[np.logical_or(np.logical_or(np.logical_or(np.logical_or(train_labels[:,0]==1,train_labels[:,2]==1),train_labels[:,4]==1),train_labels[:,6]==1),train_labels[:,8]==1)]
neg_x = train_data_x[np.logical_or(np.logical_or(np.logical_or(np.logical_or(train_labels[:,1]==1,train_labels[:,3]==1),train_labels[:,5]==1),train_labels[:,7]==1),train_labels[:,9]==1)]
# pos_x,neg_x = pos_x[:min(pos_x.shape[0],neg_x.shape[0])],neg_x[:min(pos_x.shape[0],neg_x.shape[0])]
train_data_x = np.concatenate((pos_x,neg_x))
train_labels = np.concatenate((np.array([1.0]*pos_x.shape[0]),np.array([0.0]*neg_x.shape[0]))).reshape(-1,1)
train_labels = np.array(list(zip(1.0-train_labels[:,0],train_labels[:,0])))
# del train_labels,train_data_x
def gen_pos_labels(len):
return np.array([[0.0,1.0]]*len)
def gen_neg_labels(len):
return np.array([[1.0,0.0]]*len)
test_data_x = mnist.test.images
test_labels = mnist.test.labels
test_labels = np.array(list(zip(list(test_labels[:,0]+test_labels[:,2]+test_labels[:,4]+test_labels[:,6]+test_labels[:,8]),list(test_labels[:,1]+test_labels[:,3]+test_labels[:,5]+test_labels[:,7]+test_labels[:,9]))))
# # Equalise data:
# test_pos = test_data_x[test_labels[:,posdigit]==1]
# test_neg = test_data_x[test_labels[:,negdigit]==1]
test_pos = test_data_x[np.logical_or(np.logical_or(np.logical_or(np.logical_or(test_labels[:,0]==1,test_labels[:,2]==1),test_labels[:,4]==1),test_labels[:,6]==1),test_labels[:,8]==1)]
test_neg = test_data_x[np.logical_or(np.logical_or(np.logical_or(np.logical_or(test_labels[:,1]==1,test_labels[:,3]==1),test_labels[:,5]==1),test_labels[:,7]==1),test_labels[:,9]==1)]
test_pos,test_neg = test_pos[:min(test_pos.shape[0],test_neg.shape[0])],test_neg[:min(test_pos.shape[0],test_neg.shape[0])]
test_data_x = np.concatenate((test_pos,test_neg))
test_labels = np.concatenate((np.array([1.0]*test_pos.shape[0]),np.array([0.0]*test_neg.shape[0]))).reshape(-1,1)
test_labels = np.array(list(zip(1.0-test_labels[:,0],test_labels[:,0])))
test_pos_labels = np.array([0.0]*test_pos.shape[0])
test_pos_labels = np.array(list(zip(1.0-test_pos_labels,test_pos_labels)))
test_neg_labels = np.array([1.0]*test_neg.shape[0])
test_neg_labels = np.array(list(zip(1.0-test_neg_labels,test_neg_labels)))
# print(test_pos.shape,test_neg.shape)
unlabelled_pos = unlabelled_data_x[unlabelled_labels[:,posdigit]==1]
unlabelled_neg = unlabelled_data_x[unlabelled_labels[:,negdigit]==1]
unlabelled_pos,unlabelled_neg = unlabelled_pos[:min(unlabelled_pos.shape[0],unlabelled_neg.shape[0])],unlabelled_neg[:min(unlabelled_pos.shape[0],unlabelled_neg.shape[0])]
# print(unlabelled_pos.shape,unlabelled_neg.shape)
unlabelled_data_x = np.concatenate((unlabelled_pos,unlabelled_neg))
del unlabelled_neg,unlabelled_pos,unlabelled_labels
print(unlabelled_data_x.shape)
print(pos_x.shape,neg_x.shape)
print(test_pos.shape,test_neg.shape)
# Parameters
learning_rate = 0.001
num_steps = 10
batch_size = 100
display_step = 1
num_input = 784
num_classes = 2
# tf Graph input
X = tf.placeholder("float", [None, num_input])
X_pos = tf.placeholder("float", [None, num_input])
X_neg = tf.placeholder("float", [None, num_input])
X_unlabelled = tf.placeholder("float", [None, num_input])
pos_pos_labels = tf.placeholder("float", [None, num_classes])
pos_neg_labels = tf.placeholder("float", [None, num_classes])
neg_pos_labels = tf.placeholder("float", [None, num_classes])
neg_neg_labels = tf.placeholder("float", [None, num_classes])
unlabelled_pos_labels = tf.placeholder("float", [None,num_classes])
unlabelled_neg_labels = tf.placeholder("float", [None,num_classes])
Y = tf.placeholder("float", [None, num_classes])
# Store layers weight & bias
def weight(out1,out2,out3,name):
return {
# 1x20 conv, 1 input, 32 outputs
'wc1': tf.Variable(tf.random_normal([5, 5, 1, out1]),name=name),
'wc2': tf.Variable(tf.random_normal([5, 5, out1, out2]),name=name),
# fully connected, inputs, 1024 outputs
'wd1': tf.Variable(tf.random_normal([7*7*out2, out3]),name=name),
# 1024 inputs, 10 outputs (class prediction)
'out': tf.Variable(tf.random_normal([out3, num_classes]),name=name)
}
def bias(out1,out2,out3,name):
return {
'bc1': tf.Variable(tf.random_normal([out1]),name=name),
'bc2': tf.Variable(tf.random_normal([out2]),name=name),
'bd1': tf.Variable(tf.random_normal([out3]),name=name),
'out': tf.Variable(tf.random_normal([num_classes]),name=name)
}
weights = weight(3,6,5,'0')
biases = bias(3,6,5,'0')
# Create some wrappers for simplicity
def conv2d(x, W, b, strides=1):
# Conv2D wrapper, with bias and relu activation
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(x, k=2):
# MaxPool2D wrapper
return tf.nn.max_pool(x, ksize=[1, k, k, 1], strides=[1, k, k, 1],
padding='SAME')
# Create model
def conv_net_2c2d(x,dropout):
x = tf.reshape(x, shape=[-1, 28, 28, 1])
# Convolution Layer
conv1 = conv2d(x, weights['wc1'], biases['bc1'])
# Max Pooling (down-sampling)
conv1 = maxpool2d(conv1, k=2)
conv2 = conv2d(conv1, weights['wc2'], biases['bc2'])
# Max Pooling (down-sampling)
conv2 = maxpool2d(conv2, k=2)
# Fully connected layer
# Reshape conv2 output to fit fully connected layer input
fc1 = tf.reshape(conv2, [-1, weights['wd1'].get_shape().as_list()[0]])
fc1 = tf.add(tf.matmul(fc1, weights['wd1']), biases['bd1'])
fc1 = tf.nn.relu(fc1)
# Apply Dropout
fc1 = tf.nn.dropout(fc1, dropout)
# Output, class prediction
out = tf.add(tf.matmul(fc1, weights['out']), biases['out'])
return out
# Construct model
logits = conv_net_2c2d(X,0.8)
pos_logits = conv_net_2c2d(X_pos,0.8)
neg_logits = conv_net_2c2d(X_neg,0.8)
unlabelled_logits = conv_net_2c2d(X_unlabelled,0.8)
prediction = tf.round(tf.nn.softmax(logits))
correct_prediction = tf.equal(tf.argmax(prediction, 1), tf.argmax(Y, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))
pn_loss = theta_p*tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=pos_logits, labels=pos_pos_labels)) \
+ theta_n*tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=neg_logits, labels=neg_neg_labels))
pu_loss = theta_p*tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=pos_logits, labels=pos_pos_labels)) \
+ tf.maximum( tf.zeros(1), \
- theta_p*tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=pos_logits, labels=pos_neg_labels)) \
+ tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=unlabelled_logits, labels=unlabelled_neg_labels)) \
)
nu_loss = theta_n*tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=neg_logits, labels=neg_neg_labels)) \
+ tf.maximum( tf.zeros(1), \
- theta_n*tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=neg_logits, labels=neg_pos_labels)) \
+ tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=unlabelled_logits, labels=unlabelled_pos_labels)) \
)
def run(gamma,run_type):
if run_type == 'pn':
loss_op = pn_loss
elif run_type == 'punu':
loss_op = nu_loss*gamma + pu_loss*(1-gamma) # PUNU
else:
if gamma >=0 : # PNU
loss_op = pu_loss*gamma + pn_loss*(1-gamma)
else:
gamma = -1*gamma
loss_op = nu_loss*gamma + pn_loss*(1-gamma)
gamma = -1*gamma
train_op = tf.train.AdamOptimizer(learning_rate=learning_rate).minimize(loss_op)
# train_op = tf.train.GradientDescentOptimizer(learning_rate=learning_rate).minimize(loss_op)
# Initialize the variables (i.e. assign their default value)
init = tf.global_variables_initializer()
# Start training
with tf.Session() as sess:
# Run the initializer
sess.run(init)
# tf.train.Saver([v for v in tf.global_variables()]).restore(sess, 'models/model_pnu.ckpt')
for step in range(1, num_steps+1):
# Run optimization op (backprop)
sess.run(train_op, feed_dict={
X_pos: pos_x,
X_neg: neg_x,
X_unlabelled: unlabelled_data_x,
pos_pos_labels : gen_pos_labels(pos_x.shape[0]),
pos_neg_labels : gen_neg_labels(pos_x.shape[0]),
neg_pos_labels : gen_pos_labels(neg_x.shape[0]),
neg_neg_labels : gen_neg_labels(neg_x.shape[0]),
unlabelled_pos_labels : gen_pos_labels(unlabelled_data_x.shape[0]),
unlabelled_neg_labels : gen_neg_labels(unlabelled_data_x.shape[0])
})
if step % display_step == 0 or step == 1:
# Calculate batch loss and accuracy
[loss,acc] = sess.run([loss_op,accuracy], feed_dict={
X_pos: pos_x,
X_neg: neg_x,
X_unlabelled: unlabelled_data_x,
pos_pos_labels : gen_pos_labels(pos_x.shape[0]),
pos_neg_labels : gen_neg_labels(pos_x.shape[0]),
neg_pos_labels : gen_pos_labels(neg_x.shape[0]),
neg_neg_labels : gen_neg_labels(neg_x.shape[0]),
unlabelled_pos_labels : gen_pos_labels(unlabelled_data_x.shape[0]),
unlabelled_neg_labels : gen_neg_labels(unlabelled_data_x.shape[0]),
X: train_data_x,
Y: train_labels
})
test_acc = sess.run(accuracy, feed_dict={X: test_data_x,
Y: test_labels})
print("Step " + str(step) + ", Minibatch Loss= " + \
str(loss) + ", Training Accuracy= " + \
str(acc) + ", Test Accuracy = " + str(test_acc)
)
# print("Optimization Finished!")
# Calculate accuracy for MNIST test images
print(gamma,",", \
# sess.run(accuracy, feed_dict={X: test_pos,
# Y: test_neg_labels})
# , ",",
# sess.run(accuracy, feed_dict={X: test_neg,
# Y: test_pos_labels})
# , ",",
sess.run(accuracy, feed_dict={X: test_data_x,
Y: test_labels})
)
# save_path = tf.train.Saver([v for v in tf.global_variables()]).save(sess, 'models/model_pnu.ckpt')
# print("Model saved in file: %s" % save_path)
print("PN risk")
run(0,'pn')
# print("PUNU risk")
# eta = 0.0
# while eta <=1.04:
# run(eta,'punu')
# eta += 0.2
# print("PNU risk")
# eta = -0.2
# while eta <=0.24:
# run(eta,'pnu')
# eta += 0.2