質問編集履歴
5
修正
test
CHANGED
File without changes
|
test
CHANGED
@@ -56,7 +56,7 @@
|
|
56
56
|
|
57
57
|
|
58
58
|
|
59
|
-
isinstance(n, float)
|
59
|
+
isinstance(n, float) #or isinstance(X_image, float)
|
60
60
|
|
61
61
|
>>>False
|
62
62
|
|
4
修正
test
CHANGED
@@ -1 +1 @@
|
|
1
|
-
|
1
|
+
乱数をfloat32に変換する方法
|
test
CHANGED
@@ -1,87 +1,63 @@
|
|
1
|
-
|
1
|
+
以下で発生させた乱数とデータセットをfloat32に変換するにはどうしたら良いでしょうか?
|
2
|
-
|
3
|
-
|
4
|
-
|
5
|
-
```
|
6
|
-
|
7
|
-
GraphDef cannot be larger than 2GB.
|
8
2
|
|
9
3
|
```
|
10
4
|
|
5
|
+
import numpy as np
|
11
6
|
|
7
|
+
M = 1000
|
12
8
|
|
13
|
-
|
9
|
+
N = 3072
|
14
10
|
|
15
|
-
|
16
|
-
|
17
|
-
|
18
|
-
|
19
|
-
|
20
|
-
|
21
|
-
|
22
|
-
|
23
|
-
|
24
|
-
|
25
|
-
該当部分コード
|
26
|
-
|
27
|
-
```
|
28
|
-
|
29
|
-
# loss
|
30
|
-
|
31
|
-
d_loss_1, d_loss_2 = loss(d_output_from_given_data, d_output_from_noise_for_dtrain)
|
32
|
-
|
33
|
-
g_loss = tf.reduce_sum(tf.log(1 - d_output_from_noise_for_gtrain))
|
34
|
-
|
35
|
-
|
36
|
-
|
37
|
-
|
38
|
-
|
39
|
-
# training
|
40
|
-
|
41
|
-
d_train_step = tf.train.AdamOptimizer(learning_rate=0.0001, beta1=0.5).minimize(-(d_loss_1 + d_loss_2))
|
42
|
-
|
43
|
-
|
11
|
+
n = np.ones((1,N)) * np.linspace(0.0, 1.0, M).reshape(M,1)
|
44
|
-
|
45
|
-
|
46
|
-
|
47
|
-
|
48
|
-
|
49
|
-
|
50
|
-
|
51
|
-
sess=tf.Session()
|
52
|
-
|
53
|
-
sess.run(init)
|
54
|
-
|
55
|
-
num_steps = 512
|
56
|
-
|
57
|
-
for step in range(num_steps):
|
58
|
-
|
59
|
-
g_output_eval = sess.run(g_output, feed_dict = {g_input_placeholder: inputs})
|
60
|
-
|
61
|
-
sess.run(d_train_step, feed_dict = {
|
62
|
-
|
63
|
-
g_output_placeholder: g_output_eval,
|
64
|
-
|
65
|
-
d_given_data_placeholder: X_image})
|
66
|
-
|
67
|
-
sess.run(g_train_step, feed_dict = {g_input_placeholder: inputs})
|
68
|
-
|
69
|
-
if step % (num_steps / 8) == 0:
|
70
|
-
|
71
|
-
loss_1, loss_2 = sess.run([d_loss_1, d_loss_2], feed_dict = {
|
72
|
-
|
73
|
-
g_output_placeholder: g_output_eval,
|
74
|
-
|
75
|
-
g_input_placeholder: inputs,
|
76
|
-
|
77
|
-
d_given_data_placeholder: X_image})
|
78
|
-
|
79
|
-
print('step: %d, loss1: %f, loss2: %f'%(step, loss_1, loss_2))
|
80
|
-
|
81
|
-
|
82
|
-
|
83
|
-
ValueError: GraphDef cannot be larger than 2GB.
|
84
12
|
|
85
13
|
|
86
14
|
|
87
15
|
```
|
16
|
+
|
17
|
+
備考
|
18
|
+
|
19
|
+
ネット上で公開されている方法で変換してみましたが、instanceではFalseと出てしまいます。
|
20
|
+
|
21
|
+
```
|
22
|
+
|
23
|
+
X_image=np.array(x_train, dtype=np.float32)
|
24
|
+
|
25
|
+
X_image
|
26
|
+
|
27
|
+
|
28
|
+
|
29
|
+
>>>
|
30
|
+
|
31
|
+
array([[ 0.23137255, 0.16862746, 0.19607843, ..., 0.54901963,
|
32
|
+
|
33
|
+
0.32941177, 0.28235295],
|
34
|
+
|
35
|
+
[ 0.60392159, 0.49411765, 0.41176471, ..., 0.54509807,
|
36
|
+
|
37
|
+
0.55686277, 0.56470591],
|
38
|
+
|
39
|
+
[ 1. , 0.99215686, 0.99215686, ..., 0.32549021,
|
40
|
+
|
41
|
+
0.32549021, 0.32941177],
|
42
|
+
|
43
|
+
...,
|
44
|
+
|
45
|
+
[ 0.56862748, 0.51372552, 0.4509804 , ..., 0.34901962,
|
46
|
+
|
47
|
+
0.34509805, 0.35686275],
|
48
|
+
|
49
|
+
[ 0.57254905, 0.72549021, 0.97254902, ..., 0.57254905,
|
50
|
+
|
51
|
+
0.65098041, 0.93333334],
|
52
|
+
|
53
|
+
[ 0.79607844, 0.78823531, 0.81568629, ..., 0.44705883,
|
54
|
+
|
55
|
+
0.47058824, 0.49411765]], dtype=float32)
|
56
|
+
|
57
|
+
|
58
|
+
|
59
|
+
isinstance(n, float) #or isinstance(X_image, float)
|
60
|
+
|
61
|
+
>>>False
|
62
|
+
|
63
|
+
```
|
3
修正
test
CHANGED
File without changes
|
test
CHANGED
@@ -2,7 +2,11 @@
|
|
2
2
|
|
3
3
|
|
4
4
|
|
5
|
+
```
|
6
|
+
|
5
|
-
|
7
|
+
GraphDef cannot be larger than 2GB.
|
8
|
+
|
9
|
+
```
|
6
10
|
|
7
11
|
|
8
12
|
|
2
修正
test
CHANGED
File without changes
|
test
CHANGED
@@ -2,7 +2,7 @@
|
|
2
2
|
|
3
3
|
|
4
4
|
|
5
|
-
GraphDef cannot be larger than 2GB.
|
5
|
+
``` GraphDef cannot be larger than 2GB. ```
|
6
6
|
|
7
7
|
|
8
8
|
|
1
修正
test
CHANGED
@@ -1 +1 @@
|
|
1
|
-
|
1
|
+
エラー「GraphDef cannot be larger than 2GB.」の解決法
|
test
CHANGED
@@ -1,31 +1,83 @@
|
|
1
|
-
|
1
|
+
DCGANで生成器と識別器を学習させようとして、sees.runを実行しようとしたところ下記のようなエラーが出ました。
|
2
2
|
|
3
3
|
|
4
4
|
|
5
|
-
下記の変数①②③の内積をするためには変数②と③の形状をどのようにすれば良いでしょうか?
|
6
|
-
|
7
|
-
```
|
8
|
-
|
9
|
-
変数①tf.nn.tanh(out)
|
10
|
-
|
11
|
-
|
5
|
+
GraphDef cannot be larger than 2GB.
|
12
6
|
|
13
7
|
|
14
8
|
|
15
|
-
変数②d_hidden1_weight_placeholder = tf.placeholder("float", [None, 64, 64, 3])
|
16
|
-
|
17
|
-
|
9
|
+
これは何を意味しているのでしょうか?解決策を教えていただけないでしょうか?
|
18
10
|
|
19
11
|
|
20
12
|
|
21
13
|
|
22
14
|
|
23
|
-
tf.nn.relu(tf.matmul(tf.nn.tanh(out), d_hidden1_weight_placeholder) + d_hidden1_bias_placeholders)
|
24
15
|
|
25
16
|
|
26
17
|
|
27
|
-
#エラーメッセージ
|
28
18
|
|
19
|
+
|
20
|
+
|
29
|
-
|
21
|
+
該当部分コード
|
30
22
|
|
31
23
|
```
|
24
|
+
|
25
|
+
# loss
|
26
|
+
|
27
|
+
d_loss_1, d_loss_2 = loss(d_output_from_given_data, d_output_from_noise_for_dtrain)
|
28
|
+
|
29
|
+
g_loss = tf.reduce_sum(tf.log(1 - d_output_from_noise_for_gtrain))
|
30
|
+
|
31
|
+
|
32
|
+
|
33
|
+
|
34
|
+
|
35
|
+
# training
|
36
|
+
|
37
|
+
d_train_step = tf.train.AdamOptimizer(learning_rate=0.0001, beta1=0.5).minimize(-(d_loss_1 + d_loss_2))
|
38
|
+
|
39
|
+
g_train_step = tf.train.AdamOptimizer(learning_rate=0.0001, beta1=0.5).minimize(g_loss)
|
40
|
+
|
41
|
+
|
42
|
+
|
43
|
+
|
44
|
+
|
45
|
+
|
46
|
+
|
47
|
+
sess=tf.Session()
|
48
|
+
|
49
|
+
sess.run(init)
|
50
|
+
|
51
|
+
num_steps = 512
|
52
|
+
|
53
|
+
for step in range(num_steps):
|
54
|
+
|
55
|
+
g_output_eval = sess.run(g_output, feed_dict = {g_input_placeholder: inputs})
|
56
|
+
|
57
|
+
sess.run(d_train_step, feed_dict = {
|
58
|
+
|
59
|
+
g_output_placeholder: g_output_eval,
|
60
|
+
|
61
|
+
d_given_data_placeholder: X_image})
|
62
|
+
|
63
|
+
sess.run(g_train_step, feed_dict = {g_input_placeholder: inputs})
|
64
|
+
|
65
|
+
if step % (num_steps / 8) == 0:
|
66
|
+
|
67
|
+
loss_1, loss_2 = sess.run([d_loss_1, d_loss_2], feed_dict = {
|
68
|
+
|
69
|
+
g_output_placeholder: g_output_eval,
|
70
|
+
|
71
|
+
g_input_placeholder: inputs,
|
72
|
+
|
73
|
+
d_given_data_placeholder: X_image})
|
74
|
+
|
75
|
+
print('step: %d, loss1: %f, loss2: %f'%(step, loss_1, loss_2))
|
76
|
+
|
77
|
+
|
78
|
+
|
79
|
+
ValueError: GraphDef cannot be larger than 2GB.
|
80
|
+
|
81
|
+
|
82
|
+
|
83
|
+
```
|