質問編集履歴
1
コード修正しました
title
CHANGED
|
@@ -1,1 +1,1 @@
|
|
|
1
|
-
tensorflow csvのデータ形式
|
|
1
|
+
tensorflow shapeの整え方をお尋ねしたいです。csvのデータ形式
|
body
CHANGED
|
@@ -1,12 +1,14 @@
|
|
|
1
1
|
tensorflowで12個の要素から学習を行い、restoreを行い実践データをセットした際に1×12の行列ではなくて、shape[1]のスカラー値を要求されエラーになります。shape[1]という要求は読み込みデータを
|
|
2
2
|
|
|
3
3
|
今のcsv
|
|
4
|
+
rank 1 shape [1,12]
|
|
4
5
|
````
|
|
5
6
|
[ 0.71428573, 0.85714287, 0.71428573, 0.5714286 , 0.5714286 ,
|
|
6
7
|
0.71428573, 0.5714286 , 0.71428573, 0.71428573, 0.71428573,
|
|
7
8
|
0.5714286 , 0.71428573]
|
|
8
9
|
````
|
|
10
|
+
こう変えることでshape[12,1,1] shape[1]という要求を満たせるですか？
|
|
9
|
-
|
|
11
|
+
お知恵を貸してください！
|
|
10
12
|
````
|
|
11
13
|
[ [0.71428573], [0.85714287], [0.71428573], [0.5714286] , [0.5714286] ,
|
|
12
14
|
[0.71428573], [0.5714286] , [0.71428573], [0.71428573], [0.71428573],
|
|
@@ -17,4 +19,117 @@
|
|
|
17
19
|
ValueError: Argument must be a dense tensor: [array([ 0.71428573, 0.85714287, 0.71428573, 0.5714286 , 0.5714286 ,
|
|
18
20
|
0.71428573, 0.5714286 , 0.71428573, 0.71428573, 0.71428573,
|
|
19
21
|
0.5714286 , 0.71428573], dtype=float32)] - got shape [1, 12], but wanted [1].
|
|
20
|
-
````
|
|
22
|
+
````
|
|
23
|
+
その他のコード（問題なのはsession2のほうです）
|
|
24
|
+
|
|
25
|
+
```
|
|
26
|
+
|
|
27
|
+
import tensorflow as tf
|
|
28
|
+
import numpy
|
|
29
|
+
import os
|
|
30
|
+
|
|
31
|
+
cwd = os.getcwd()
|
|
32
|
+
|
|
33
|
+
SCORE_SIZE = 12
|
|
34
|
+
HIDDEN_UNIT_SIZE = 40
|
|
35
|
+
TRAIN_DATA_SIZE = 45
|
|
36
|
+
TACK = 1
|
|
37
|
+
raw_input = numpy.loadtxt(open("test.csv"), delimiter=",")
|
|
38
|
+
[tensor, score] = numpy.hsplit(raw_input, [1])
|
|
39
|
+
[tensor_train, tensor_test] = numpy.vsplit(tensor, [TRAIN_DATA_SIZE])
|
|
40
|
+
[score_train, score_test] = numpy.vsplit(score, [TRAIN_DATA_SIZE])
|
|
41
|
+
print(score_test)
|
|
42
|
+
#tensorは正解データtrainは学習モデル、scoreは学習データ、testは実データ
|
|
43
|
+
|
|
44
|
+
def inference(score_placeholder):
|
|
45
|
+
with tf.name_scope('hidden1') as scope:
|
|
46
|
+
hidden1_weight = tf.Variable(tf.truncated_normal([SCORE_SIZE, HIDDEN_UNIT_SIZE], stddev=0.01), name="hidden1_weight")
|
|
47
|
+
hidden1_bias = tf.Variable(tf.constant(0.1, shape=[HIDDEN_UNIT_SIZE]), name="hidden1_bias")
|
|
48
|
+
hidden1_output = tf.nn.relu(tf.matmul(score_placeholder, hidden1_weight) + hidden1_bias)
|
|
49
|
+
with tf.name_scope('output') as scope:
|
|
50
|
+
output_weight = tf.Variable(tf.truncated_normal([HIDDEN_UNIT_SIZE, 1], stddev=0.01), name="output_weight")
|
|
51
|
+
output_bias = tf.Variable(tf.constant(0.1, shape=[1]), name="output_bias")
|
|
52
|
+
output = tf.matmul(hidden1_output, output_weight) + output_bias
|
|
53
|
+
if TACK != 1:
|
|
54
|
+
print("saku1")
|
|
55
|
+
print(output)
|
|
56
|
+
else:
|
|
57
|
+
print("saku2")
|
|
58
|
+
|
|
59
|
+
return tf.nn.l2_normalize(output, 0)
|
|
60
|
+
|
|
61
|
+
def loss(output, tensor_placeholder, loss_label_placeholder):
|
|
62
|
+
with tf.name_scope('loss') as scope:
|
|
63
|
+
loss = tf.nn.l2_loss(output - tf.nn.l2_normalize(tensor_placeholder, 0))
|
|
64
|
+
tf.summary.scalar('loss_label_placeholder', loss)
|
|
65
|
+
return loss
|
|
66
|
+
|
|
67
|
+
def training(loss):
|
|
68
|
+
with tf.name_scope('training') as scope:
|
|
69
|
+
train_step = tf.train.GradientDescentOptimizer(0.01).minimize(loss)
|
|
70
|
+
return train_step
|
|
71
|
+
|
|
72
|
+
|
|
73
|
+
|
|
74
|
+
with tf.Graph().as_default():
|
|
75
|
+
tensor_placeholder = tf.placeholder(tf.float32, [None, 1], name="tensor_placeholder")
|
|
76
|
+
score_placeholder = tf.placeholder(tf.float32, [None, SCORE_SIZE], name="score_placeholder")
|
|
77
|
+
loss_label_placeholder = tf.placeholder("string", name="loss_label_placeholder")
|
|
78
|
+
|
|
79
|
+
feed_dict_train={
|
|
80
|
+
tensor_placeholder: tensor_train,
|
|
81
|
+
score_placeholder: score_train,
|
|
82
|
+
loss_label_placeholder: "loss_train"
|
|
83
|
+
}
|
|
84
|
+
|
|
85
|
+
feed_dict_test={
|
|
86
|
+
tensor_placeholder: tensor_test,
|
|
87
|
+
score_placeholder: score_test,
|
|
88
|
+
loss_label_placeholder: "loss_test"
|
|
89
|
+
}
|
|
90
|
+
|
|
91
|
+
output = inference(score_placeholder)
|
|
92
|
+
loss = loss(output, tensor_placeholder, loss_label_placeholder)
|
|
93
|
+
training_op = training(loss)
|
|
94
|
+
summary_op = tf.summary.merge_all()
|
|
95
|
+
init = tf.global_variables_initializer()
|
|
96
|
+
best_loss = float("inf")
|
|
97
|
+
|
|
98
|
+
with tf.Session() as sess:
|
|
99
|
+
summary_writer = tf.summary.FileWriter('data', graph_def=sess.graph_def)
|
|
100
|
+
sess.run(init)
|
|
101
|
+
for step in range(10000):
|
|
102
|
+
sess.run(training_op, feed_dict=feed_dict_train)
|
|
103
|
+
loss_test = sess.run(loss, feed_dict=feed_dict_test)
|
|
104
|
+
if loss_test < best_loss:
|
|
105
|
+
best_loss = loss_test
|
|
106
|
+
best_match = sess.run(output, feed_dict=feed_dict_test)
|
|
107
|
+
#if step % 100 == 0:
|
|
108
|
+
# summary_str = sess.run(summary_op, feed_dict=feed_dict_test)
|
|
109
|
+
# summary_str += sess.run(summary_op, feed_dict=feed_dict_train)
|
|
110
|
+
# summary_writer.add_summary(summary_str, step)
|
|
111
|
+
|
|
112
|
+
saver=tf.train.Saver()
|
|
113
|
+
saver.save(sess,cwd+'/model.ckpt')
|
|
114
|
+
print(cwd)
|
|
115
|
+
print(best_match)
|
|
116
|
+
print('Saved a model.')
|
|
117
|
+
sess.close()
|
|
118
|
+
|
|
119
|
+
with tf.Session() as sess2:
|
|
120
|
+
#変数の読み込み
|
|
121
|
+
summary_writer = tf.summary.FileWriter('data', graph=sess2.graph)
|
|
122
|
+
#sess2.run(init)
|
|
123
|
+
#新しいデータ
|
|
124
|
+
TRAIN_DATA_SIZE2 = 0
|
|
125
|
+
test2 = numpy.loadtxt(open("one_record.csv"), delimiter=",").astype(numpy.float32)
|
|
126
|
+
score3 = [test2]
|
|
127
|
+
print(score3)
|
|
128
|
+
saver = tf.train.Saver()
|
|
129
|
+
cwd = os.getcwd()
|
|
130
|
+
saver.restore(sess2,cwd + "/model.ckpt")
|
|
131
|
+
best_match2 = sess2.run(inference(score3))
|
|
132
|
+
print(best_match2)
|
|
133
|
+
print("fin")
|
|
134
|
+
sess2.close()
|
|
135
|
+
```
|