質問編集履歴
3
書式の改善
test
CHANGED
|
File without changes
|
test
CHANGED
|
@@ -60,7 +60,83 @@
|
|
|
60
60
|
|
|
61
61
|
```
|
|
62
62
|
|
|
63
|
+
```Python
|
|
63
64
|
|
|
65
|
+
# ============================save============================
|
|
66
|
+
|
|
67
|
+
# var_to_start = tf.trainable_variables()
|
|
68
|
+
|
|
69
|
+
# saver = tf.train.Saver(var_to_start)
|
|
70
|
+
|
|
71
|
+
# saver = tf.train.Saver(tf.global_variables())
|
|
72
|
+
|
|
73
|
+
# saver = tf.train.Saver()
|
|
74
|
+
|
|
75
|
+
var_all = tf.all_variables()
|
|
76
|
+
|
|
77
|
+
saver = tf.train.Saver(var_all)
|
|
78
|
+
|
|
79
|
+
save_path = saver.save(sess, CKPT_PATH + '/model.ckpt', global_step=100)
|
|
80
|
+
|
|
81
|
+
print("Model saved in file: %s"%save_path)
|
|
82
|
+
|
|
83
|
+
k += 1
|
|
84
|
+
|
|
85
|
+
```
|
|
86
|
+
|
|
87
|
+
```Python
|
|
88
|
+
|
|
89
|
+
# ============================restore->test============================
|
|
90
|
+
|
|
91
|
+
init = tf.global_variables_initializer()
|
|
92
|
+
|
|
93
|
+
sess = tf.InteractiveSession()
|
|
94
|
+
|
|
95
|
+
sess.run(init)
|
|
96
|
+
|
|
97
|
+
# check
|
|
98
|
+
|
|
99
|
+
# var_to_start = tf.trainable_variables()
|
|
100
|
+
|
|
101
|
+
# saver = tf.train.Saver(var_to_start)
|
|
102
|
+
|
|
103
|
+
# saver = tf.train.Saver(tf.global_variables())
|
|
104
|
+
|
|
105
|
+
var_all = tf.all_variables()
|
|
106
|
+
|
|
107
|
+
saver = tf.train.Saver(var_all)
|
|
108
|
+
|
|
109
|
+
# saver = tf.train.Saver()
|
|
110
|
+
|
|
111
|
+
|
|
112
|
+
|
|
113
|
+
ckpt = tf.train.get_checkpoint_state(CKPT_PATH + '/')
|
|
114
|
+
|
|
115
|
+
if ckpt:
|
|
116
|
+
|
|
117
|
+
last_model = ckpt.model_checkpoint_path
|
|
118
|
+
|
|
119
|
+
print("load " + last_model)
|
|
120
|
+
|
|
121
|
+
saver.restore(sess, last_model)
|
|
122
|
+
|
|
123
|
+
print("Model restored.")
|
|
124
|
+
|
|
125
|
+
# print(sess.run(beta1))
|
|
126
|
+
|
|
127
|
+
|
|
128
|
+
|
|
129
|
+
print("test accuracy %g" % accuracy.eval(feed_dict={
|
|
130
|
+
|
|
131
|
+
x: data.test.images, y_: data.test.labels, keep_prob: 1.0}))
|
|
132
|
+
|
|
133
|
+
# print(W_conv1.eval())
|
|
134
|
+
|
|
135
|
+
else:
|
|
136
|
+
|
|
137
|
+
sess.run(init)
|
|
138
|
+
|
|
139
|
+
```
|
|
64
140
|
|
|
65
141
|
### 補足情報（FW/ツールのバージョンなど）
|
|
66
142
|
|
2
書式の改善
test
CHANGED
|
File without changes
|
test
CHANGED
|
@@ -10,7 +10,7 @@
|
|
|
10
10
|
|
|
11
11
|
バッチの正規化を行う関数(def)を作成してCNNを学習されているのだが、1枚のデータでテストを行う際にバッチの正規化を正常に行うことができない。
|
|
12
12
|
|
|
13
|
-
is_training=Falseにした際にpop_mean,pop_varが
|
|
13
|
+
is_training=Falseにした際にpop_mean,pop_varが学習によって更新された値(平均のようなもの)を出力するのではなく、初期値を返してしまう。(下のコード参照)
|
|
14
14
|
|
|
15
15
|
### 該当のソースコード
|
|
16
16
|
|
1
タグ追加
test
CHANGED
|
File without changes
|
test
CHANGED
|
File without changes
|