-
Notifications
You must be signed in to change notification settings - Fork 5
/
kcnn.py
212 lines (181 loc) · 6.17 KB
/
kcnn.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
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import gzip
import os
import sys
import tarfile
import time
from six.moves import urllib
import tensorflow as tf
import google.protobuf
import numpy as np
from data_utils import *
from batch_norm import *
from summary import _activation_summary,_add_loss_summaries
from kmeans import kmeans,extract_features
batch_size = 64
initfact=10
learning_rate=.1
path='dataset'
path+='/cifar-100-python'
n_epochs = 800
NUM_CLASSES=20
valid_set=1000
time_per_epoch=10
repeat_layer=1
visual=False
ifbatchnorm=True
weight_d=0.001
ifDrop=False
features=256
def elapsed():
return (time.time()-t)/60
def _variable_on_cpu(name, shape, initializer):
with tf.device('/cpu:0'):
var = tf.get_variable(name, shape, initializer=initializer)
return var
def _variable_with_weight_decay(name, shape, stddev, wd):
var = _variable_on_cpu(name, shape,
tf.truncated_normal_initializer(stddev=stddev))
if wd is not None and wd !=0:
weight_decay = tf.mul(tf.nn.l2_loss(var), wd, name='weight_loss')
tf.add_to_collection('losses', weight_decay)
return var
def eval(xx,yy):
return str(sess.run(accuracy,
feed_dict={
x:xx,
y:yy,
is_training: False}))
def svd_orthonormal(shape):
flat_shape = (shape[0], np.prod(shape[1:]))
a = np.random.standard_normal(flat_shape)
u, _, v = np.linalg.svd(a, full_matrices=False)
q = u if u.shape == flat_shape else v
q = q.reshape(shape)
return q
######################architecture##########################################
trainWs=[]
x = tf.placeholder(tf.float32, [None,9,9,features])
y = tf.placeholder(tf.float32, [None])
is_training = tf.placeholder(tf.bool, name='is_training')
LUSV=tf.placeholder(tf.float32)
lr=tf.placeholder(tf.float32)
kernel = _variable_with_weight_decay('conv0',
shape=[3, 3, features, features],
stddev=np.sqrt(2.0/initfact/3)
, wd=weight_d)
ConvLayer0 = tf.nn.conv2d(x, kernel, [1, 1, 1, 1], padding='SAME')
net = tf.nn.relu(ConvLayer0)
if visual:
_activation_summary(net)
#Global avg pooling
net_shape = net.get_shape().as_list()
print(net_shape)
net = tf.nn.avg_pool(net,
ksize=[1, 3, 3, 1],
strides=[1, 3, 3, 1],
padding='VALID',name='global_pooling')
net_shape = net.get_shape().as_list()
hidden_inp=9*net_shape[1] * net_shape[2] * net_shape[3]
print(net_shape)
n_fc = 400
net = tf.reshape(net,
[-1, hidden_inp])
hw = _variable_with_weight_decay('hw',
[hidden_inp,n_fc],
stddev=1/hidden_inp,
wd=weight_d)
hb = _variable_on_cpu('hb',
[n_fc],
tf.constant_initializer(0.0))
# %% Create a fully-connected layer:
net = tf.nn.relu(tf.matmul(net, hw) + hb)
#softmax
weights = _variable_with_weight_decay('softmax_w',
[n_fc, NUM_CLASSES],
stddev=1/n_fc,
wd=weight_d)
biases = _variable_on_cpu('softmax_b',
[NUM_CLASSES],
tf.constant_initializer(0.0))
trainWs.append(weights)
trainWs.append(biases)
softmax_linear = tf.add(tf.matmul(net, weights), biases, name='softmax')
y = tf.cast(y, tf.int64)
cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(
softmax_linear, y, name='cross_entropy_per_example')
if visual:
_activation_summary(cross_entropy)
summary_op = tf.merge_all_summaries()
cross_entropy_mean = tf.reduce_mean(cross_entropy, name='cross_entropy')
tf.add_to_collection('losses', cross_entropy_mean)
cross_entropy_mean=tf.add_n(tf.get_collection('losses'))
# train
#train_step=tf.train.MomentumOptimizer(lr,.9).minimize(cross_entropy_mean)
train_step=tf.train.AdagradOptimizer(lr).minimize(cross_entropy_mean)
#predict
correct_prediction=tf.equal(tf.argmax(softmax_linear,1),y)
accuracy=tf.reduce_mean(tf.cast(correct_prediction,'float'))
tess=[cross_entropy_mean,train_step]
#########################preprocessing#######################################
t=time.time()
# load data
Xtr, Ytr, Xte, Yte=load_CIFAR100(path)
# simple preprocessing
mean_image = np.mean(Xtr, axis=0)
Xtr -= mean_image
Xte -= mean_image
Xtr=Xtr.swapaxes(1,3)
Xte=Xte.swapaxes(1,3)
centroids,a,b,c,d=kmeans(Xtr,1600,selected_feats=features)
print(elapsed())
Xtr=extract_features(Xtr,centroids,a,b,c,d)
Xte=extract_features(Xte,centroids,a,b,c,d)
print(elapsed())
##########################training###############################
def nextBatch():
idx=np.random.choice(numTrain,batch_size)
return Xtr[idx], Ytr[idx]
numTrain=len(Xtr)-valid_set
iter_per_epoch=numTrain // batch_size
# %% We now create a new session to actually perform the initialization the
# variables:
saver = tf.train.Saver()
sess=tf.Session()
#saver.restore(sess,'26.ckpt')
sess.run(tf.initialize_all_variables())
if visual:
summary_writer = tf.train.SummaryWriter("./summary", sess.graph)
for epoch_i in range(n_epochs):
avg_loss=0
for batch_i in range(iter_per_epoch):
batch_xs,batch_ys=nextBatch()
loss,_=sess.run([cross_entropy_mean,train_step],
feed_dict={x: batch_xs, y: batch_ys, is_training: True,lr:learning_rate})
avg_loss+=loss
if batch_i%int(iter_per_epoch/10)==0:
if visual:
summary_str = sess.run(summary_op,
feed_dict={x: batch_xs, y: batch_ys, is_training: False})
summary_writer.add_summary(summary_str, epoch_i*iter_per_epoch+batch_i)
print('loss '+str(loss)+',time '+str(elapsed()))
print("epoch"+str(epoch_i)+
" avg_loss:"+str(avg_loss/iter_per_epoch)+
" train acc:"+ eval( batch_xs,batch_ys )+
" val acc:"+ eval(Xtr[range(-valid_set,-1)],Ytr[range(-valid_set,-1)]))
save_path = saver.save(sess,'summary/'+str(repeat_layer)+'_'+ str(epoch_i)+".ckpt")
print("Model saved in file: %s" % save_path)
for w in trainWs:
a=w.eval(session=sess)
print(a.shape,a.mean(),a.std())
if epoch_i>83:
learning_rate=.01
if epoch_i>125:
learning_rate=.001
if epoch_i>162:
break
# if elapsed()>180-time_per_epoch:
# break
print("test acc:"+eval(Xte,Yte))