質問編集履歴
5
追記質問
title
CHANGED
|
File without changes
|
body
CHANGED
|
@@ -8,20 +8,24 @@
|
|
|
8
8
|
|
|
9
9
|
|
|
10
10
|
def trainable_inference(input):
|
|
11
|
-
hidden1_weight = tf.get_variable([INPUT_SIZE, HIDDEN_UNIT_SIZE], initializer = tf.random_normal_initializer(0, 0.1))
|
|
11
|
+
hidden1_weight = tf.get_variable("d_hidden1_weight",[INPUT_SIZE, HIDDEN_UNIT_SIZE], initializer = tf.random_normal_initializer(0, 0.1))
|
|
12
|
-
hidden1_bias = tf.get_variable([HIDDEN_UNIT_SIZE], initializer = tf.constant_initializer(0.1))
|
|
12
|
+
hidden1_bias = tf.get_variable("d_hidden1_bias",[HIDDEN_UNIT_SIZE], initializer = tf.constant_initializer(0.1))
|
|
13
|
-
output_weight = tf.get_variable([HIDDEN_UNIT_SIZE, 1],initializer = tf.random_normal_initializer(0, 0.1))
|
|
13
|
+
output_weight = tf.get_variable("d_output_bias",[HIDDEN_UNIT_SIZE, 1],initializer = tf.random_normal_initializer(0, 0.1))
|
|
14
|
-
output_bias = tf.get_variable([1],initializer = tf.constant_initializer(0.1))
|
|
14
|
+
output_bias = tf.get_variable("d_output_bias",[1],initializer = tf.constant_initializer(0.1))
|
|
15
15
|
hidden1_output = tf.nn.relu(tf.matmul(input, hidden1_weight) + hidden1_bias)
|
|
16
16
|
output = tf.sigmoid(tf.matmul(hidden1_output, output_weight) + output_bias)
|
|
17
17
|
return output
|
|
18
|
+
|
|
18
19
|
```
|
|
19
20
|
この関数に引数として以下の値を入れたのですが、エラーが発生してしまいました。
|
|
21
|
+
|
|
22
|
+
追記
|
|
23
|
+
名前を定義したのですが下記エラーが出てしまいました。
|
|
20
24
|
```
|
|
21
25
|
d_given_data_placeholder = tf.placeholder("float", [None, INPUT_SIZE])
|
|
22
26
|
|
|
23
27
|
trainable_inference(d_given_data_placeholder)
|
|
24
28
|
|
|
25
|
-
TypeError:
|
|
29
|
+
TypeError:Value passed to parameter 'shape' has DataType string not in list of allowed values: int32, int64
|
|
26
30
|
```
|
|
27
|
-
解決策
|
|
31
|
+
この解決策もお願いいたします
|
4
修正
title
CHANGED
|
File without changes
|
body
CHANGED
|
@@ -20,7 +20,7 @@
|
|
|
20
20
|
```
|
|
21
21
|
d_given_data_placeholder = tf.placeholder("float", [None, INPUT_SIZE])
|
|
22
22
|
|
|
23
|
-
trainable_inference(
|
|
23
|
+
trainable_inference(d_given_data_placeholder)
|
|
24
24
|
|
|
25
25
|
TypeError: unhashable type: 'list'
|
|
26
26
|
```
|
3
修正
title
CHANGED
|
File without changes
|
body
CHANGED
|
@@ -18,7 +18,7 @@
|
|
|
18
18
|
```
|
|
19
19
|
この関数に引数として以下の値を入れたのですが、エラーが発生してしまいました。
|
|
20
20
|
```
|
|
21
|
-
|
|
21
|
+
d_given_data_placeholder = tf.placeholder("float", [None, INPUT_SIZE])
|
|
22
22
|
|
|
23
23
|
trainable_inference(g_input_placeholder)
|
|
24
24
|
|
2
修正
title
CHANGED
|
File without changes
|
body
CHANGED
|
@@ -18,7 +18,7 @@
|
|
|
18
18
|
```
|
|
19
19
|
この関数に引数として以下の値を入れたのですが、エラーが発生してしまいました。
|
|
20
20
|
```
|
|
21
|
-
|
|
21
|
+
given_data_placeholder = tf.placeholder("float", [None, INPUT_SIZE])
|
|
22
22
|
|
|
23
23
|
trainable_inference(g_input_placeholder)
|
|
24
24
|
|
1
修正
title
CHANGED
|
@@ -1,1 +1,1 @@
|
|
|
1
|
-
|
|
1
|
+
エラー「unhashable type: 'list'」の解決方法
|
body
CHANGED
|
@@ -1,141 +1,27 @@
|
|
|
1
|
-
|
|
1
|
+
CNNの一部で次のような関数を定義しました
|
|
2
2
|
|
|
3
|
+
```
|
|
4
|
+
INPUT_SIZE = 1
|
|
3
|
-
|
|
5
|
+
HIDDEN_UNIT_SIZE = 64
|
|
4
|
-
|
|
6
|
+
TRAIN_DATA_SIZE = 100
|
|
7
|
+
OUTPUT_SIZE = 1
|
|
5
8
|
|
|
9
|
+
|
|
10
|
+
def trainable_inference(input):
|
|
11
|
+
hidden1_weight = tf.get_variable([INPUT_SIZE, HIDDEN_UNIT_SIZE], initializer = tf.random_normal_initializer(0, 0.1))
|
|
12
|
+
hidden1_bias = tf.get_variable([HIDDEN_UNIT_SIZE], initializer = tf.constant_initializer(0.1))
|
|
13
|
+
output_weight = tf.get_variable([HIDDEN_UNIT_SIZE, 1],initializer = tf.random_normal_initializer(0, 0.1))
|
|
14
|
+
output_bias = tf.get_variable([1],initializer = tf.constant_initializer(0.1))
|
|
15
|
+
hidden1_output = tf.nn.relu(tf.matmul(input, hidden1_weight) + hidden1_bias)
|
|
16
|
+
output = tf.sigmoid(tf.matmul(hidden1_output, output_weight) + output_bias)
|
|
17
|
+
return output
|
|
6
18
|
```
|
|
7
|
-
y.shape y_conv.shape
|
|
8
|
-
|
|
19
|
+
この関数に引数として以下の値を入れたのですが、エラーが発生してしまいました。
|
|
9
20
|
```
|
|
10
|
-
yはPlaceholderです。
|
|
11
|
-
しかしこのまま学習を行おうとすると以下のエラーメッセージが出てしまいます。
|
|
12
|
-
```
|
|
13
|
-
|
|
21
|
+
d_given_data_placeholder = tf.placeholder("float", [None, INPUT_SIZE])
|
|
14
|
-
```
|
|
15
|
-
これはおそらくラベルの形に等しいのでラベルをPlaceholderに入れることができないことを表しています。しかしここでPlaceholderの形を[None]にしてしまうと損失関数のlogits=y_convとlabels=yが一致しなくなってしまいます。
|
|
16
|
-
・損失関数のlogits=y_convとlabels=yが一致させる
|
|
17
|
-
・ラベル(y_trainとy_test)とy（Placeholder）を一致させる
|
|
18
22
|
|
|
19
|
-
|
|
23
|
+
trainable_inference(g_input_placeholder)
|
|
20
24
|
|
|
21
|
-
|
|
25
|
+
TypeError: unhashable type: 'list'
|
|
22
|
-
|
|
23
26
|
```
|
|
24
|
-
#データセット作成
|
|
25
|
-
import sys
|
|
26
|
-
import pickle
|
|
27
|
-
import numpy as np
|
|
28
|
-
|
|
29
|
-
|
|
30
|
-
def unpickle(file):
|
|
31
|
-
fp = open(file, 'rb')
|
|
32
|
-
if sys.version_info.major == 2:
|
|
33
|
-
data = pickle.load(fp)
|
|
34
|
-
elif sys.version_info.major == 3:
|
|
35
|
-
data = np.load(fp, encoding='latin1')
|
|
36
|
-
fp.close()
|
|
37
|
-
|
|
38
|
-
return data
|
|
39
|
-
|
|
40
|
-
X_train = None
|
|
41
|
-
|
|
27
|
+
解決策を教えてください？
|
|
42
|
-
|
|
43
|
-
for i in range(1,6):
|
|
44
|
-
data_dic = unpickle("data_batch_1".format(i))
|
|
45
|
-
if i == 1:
|
|
46
|
-
X_train = data_dict['data']
|
|
47
|
-
y_train = data_dict['labels']
|
|
48
|
-
else:
|
|
49
|
-
X_train = np.vstack((X_train, data_dict['data']))
|
|
50
|
-
y_train = np.hstack((y_train, data_dict['labels']))
|
|
51
|
-
|
|
52
|
-
test_data_dic = unpickle("test_batch")
|
|
53
|
-
X_test = test_data_dic['data']
|
|
54
|
-
y_test =np.array(test_data_dic['labels'])
|
|
55
|
-
|
|
56
|
-
#CNN
|
|
57
|
-
from __future__ import absolute_import
|
|
58
|
-
from __future__ import division
|
|
59
|
-
from __future__ import print_function
|
|
60
|
-
import tensorflow as tf
|
|
61
|
-
import os
|
|
62
|
-
import time
|
|
63
|
-
import numpy as np
|
|
64
|
-
NUM_CLASSES = 10
|
|
65
|
-
|
|
66
|
-
def weight_variable(shape):
|
|
67
|
-
initial = tf.truncated_normal(shape, stddev=0.1)
|
|
68
|
-
return tf.Variable(initial)
|
|
69
|
-
|
|
70
|
-
# バイアス変数
|
|
71
|
-
def bias_variable(shape):
|
|
72
|
-
initial = tf.constant(0.1, shape=shape)
|
|
73
|
-
return tf.Variable(initial)
|
|
74
|
-
|
|
75
|
-
# 畳み込み
|
|
76
|
-
def conv2d(x, W):
|
|
77
|
-
return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME')
|
|
78
|
-
|
|
79
|
-
# プーリング
|
|
80
|
-
def max_pool_2x2(x):
|
|
81
|
-
return tf.nn.max_pool(x, ksize=[1, 2, 2, 1],
|
|
82
|
-
strides=[1, 2, 2, 1], padding='SAME')
|
|
83
|
-
|
|
84
|
-
x = tf.placeholder(tf.float32, [None, 3*32*32])
|
|
85
|
-
y = tf.placeholder(tf.float32, [None,10])
|
|
86
|
-
|
|
87
|
-
# 畳み込み１層目(conv1)
|
|
88
|
-
W_conv1 = weight_variable([5, 5, 3, 32])
|
|
89
|
-
b_conv1 = bias_variable([32])
|
|
90
|
-
x_image = tf.reshape(x, [-1,32,32,3])
|
|
91
|
-
h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1)
|
|
92
|
-
# pool1
|
|
93
|
-
h_pool1 = max_pool_2x2(h_conv1)
|
|
94
|
-
|
|
95
|
-
|
|
96
|
-
# 畳み込み２層目(conv2)
|
|
97
|
-
W_conv2 = weight_variable([5, 5, 32, 64])
|
|
98
|
-
b_conv2 = bias_variable([64])
|
|
99
|
-
h_conv2 = tf.nn.relu(conv2d(h_pool1, W_conv2) + b_conv2)
|
|
100
|
-
# pool2
|
|
101
|
-
h_pool2 = max_pool_2x2(h_conv2)
|
|
102
|
-
|
|
103
|
-
|
|
104
|
-
# 全結合層
|
|
105
|
-
W_fc1 = weight_variable([8 * 8 * 64, 1024])
|
|
106
|
-
b_fc1 = bias_variable([1024])
|
|
107
|
-
h_pool2_flat = tf.reshape(h_pool2, [-1, 8 * 8 * 64])
|
|
108
|
-
h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1)
|
|
109
|
-
|
|
110
|
-
|
|
111
|
-
# ドロップアウト層
|
|
112
|
-
keep_prob = tf.placeholder(tf.float32)
|
|
113
|
-
h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)
|
|
114
|
-
|
|
115
|
-
# 出力層
|
|
116
|
-
W_fc2 = weight_variable([1024, 10])
|
|
117
|
-
b_fc2 = bias_variable([10])
|
|
118
|
-
y_conv = tf.matmul(h_fc1_drop, W_fc2) + b_fc2
|
|
119
|
-
|
|
120
|
-
# 損失関数（交差エントロピー誤差）
|
|
121
|
-
cross_entropy = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(logits=y_conv, labels=y))
|
|
122
|
-
|
|
123
|
-
# 勾配
|
|
124
|
-
train_step = tf.train.AdamOptimizer(1e-4).minimize(cross_entropy)
|
|
125
|
-
|
|
126
|
-
# 精度
|
|
127
|
-
correct_prediction = tf.equal(tf.argmax(y_conv, 1), tf.argmax(y, 1))
|
|
128
|
-
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
|
|
129
|
-
# セッション
|
|
130
|
-
sess = tf.InteractiveSession()
|
|
131
|
-
sess.run(tf.global_variables_initializer())
|
|
132
|
-
＃学習
|
|
133
|
-
i=0
|
|
134
|
-
for i in range(100):
|
|
135
|
-
i+=1
|
|
136
|
-
sess.run(train_step, feed_dict={x: X_train, y: y_train, keep_prob: 0.5})
|
|
137
|
-
if i % 10 == 0:
|
|
138
|
-
# 途中経過（10件ごと）
|
|
139
|
-
loss_val, acc_val = sess.run([loss, accuracy], feed_dict={x: X_test, y: y_test, keep_prob: 1.0})
|
|
140
|
-
print("step: %d, Loss: %f, Accuracy: %f" % (i, loss_val, acc_val))
|
|
141
|
-
```
|