質問編集履歴
4
修正
test
CHANGED
|
File without changes
|
test
CHANGED
|
@@ -32,6 +32,14 @@
|
|
|
32
32
|
|
|
33
33
|
|
|
34
34
|
|
|
35
|
+
データ形式は1×13行列になっています。最初の1×1の値は学習データでは正解値が入るため、また、モデルの再利用時にinterface()関数を呼び出しているために同じデータ形式にするために0を入れています。
|
|
36
|
+
|
|
37
|
+
````
|
|
38
|
+
|
|
39
|
+
0,0.714285714,0.857142857,0.714285714,0.571428571,0.571428571,0.714285714,0.571428571,0.714285714,0.714285714,0.714285714,0.571428571,0.714285714
|
|
40
|
+
|
|
41
|
+
````
|
|
42
|
+
|
|
35
43
|
|
|
36
44
|
|
|
37
45
|
````
|
3
修正
test
CHANGED
|
File without changes
|
test
CHANGED
|
@@ -32,6 +32,68 @@
|
|
|
32
32
|
|
|
33
33
|
|
|
34
34
|
|
|
35
|
+
|
|
36
|
+
|
|
37
|
+
````
|
|
38
|
+
|
|
39
|
+
with tf.Session() as sess2:
|
|
40
|
+
|
|
41
|
+
|
|
42
|
+
|
|
43
|
+
#TRAIN_DATA_SIZE2 = 0 複数行CSVの時にすべて実践データにまわすために0行目を分割している
|
|
44
|
+
|
|
45
|
+
test2 = numpy.loadtxt(open("one_record.csv"), delimiter=",")
|
|
46
|
+
|
|
47
|
+
[tensor2,score2] = numpy.hsplit(test2, [1])
|
|
48
|
+
|
|
49
|
+
#[tensor_train2,tensor_test2] = numpy.vsplit(tensor2, [TRAIN_DATA_SIZE2])
|
|
50
|
+
|
|
51
|
+
#[score_train2, score_test2] = numpy.vsplit(score2, [TRAIN_DATA_SIZE2])
|
|
52
|
+
|
|
53
|
+
print(tensor2)
|
|
54
|
+
|
|
55
|
+
print(score2)
|
|
56
|
+
|
|
57
|
+
#モデルつくり
|
|
58
|
+
|
|
59
|
+
feed_dict_test2 = {
|
|
60
|
+
|
|
61
|
+
|
|
62
|
+
|
|
63
|
+
複数行CSVの際にはvsplitして、tensor_test2,score_test2がそれぞれtensor,scoreにはいることで計算できていました。どう直せばいいでしょうか。
|
|
64
|
+
|
|
65
|
+
tensor_placeholder:tensor2,
|
|
66
|
+
|
|
67
|
+
score_placeholder:score2,
|
|
68
|
+
|
|
69
|
+
loss_label_placeholder:"loss_test2"
|
|
70
|
+
|
|
71
|
+
}
|
|
72
|
+
|
|
73
|
+
|
|
74
|
+
|
|
75
|
+
saver = tf.train.Saver()
|
|
76
|
+
|
|
77
|
+
cwd = os.getcwd()
|
|
78
|
+
|
|
79
|
+
saver.restore(sess2,cwd + "/model.ckpt")
|
|
80
|
+
|
|
81
|
+
|
|
82
|
+
|
|
83
|
+
print("recover")
|
|
84
|
+
|
|
85
|
+
best_match2 = sess2.run(output, feed_dict=feed_dict_test2)
|
|
86
|
+
|
|
87
|
+
print(best_match2)
|
|
88
|
+
|
|
89
|
+
print("fin")
|
|
90
|
+
|
|
91
|
+
sess2.close()
|
|
92
|
+
|
|
93
|
+
````
|
|
94
|
+
|
|
95
|
+
|
|
96
|
+
|
|
35
97
|
**
|
|
36
98
|
|
|
37
99
|
|
|
@@ -58,7 +120,231 @@
|
|
|
58
120
|
|
|
59
121
|
[https://teratail.com/questions/82450](https://teratail.com/questions/82450)
|
|
60
122
|
|
|
123
|
+
|
|
124
|
+
|
|
125
|
+
|
|
126
|
+
|
|
127
|
+
|
|
128
|
+
|
|
129
|
+
エラーコード
|
|
130
|
+
|
|
131
|
+
````
|
|
132
|
+
|
|
133
|
+
ValueError: Cannot feed value of shape (1,) for Tensor 'tensor_placeholder:0', which has shape '(?, 1)'
|
|
134
|
+
|
|
135
|
+
````
|
|
136
|
+
|
|
137
|
+
|
|
138
|
+
|
|
139
|
+
|
|
140
|
+
|
|
141
|
+
|
|
142
|
+
|
|
143
|
+
|
|
144
|
+
|
|
145
|
+
````
|
|
146
|
+
|
|
147
|
+
import tensorflow as tf
|
|
148
|
+
|
|
149
|
+
import numpy
|
|
150
|
+
|
|
151
|
+
import os
|
|
152
|
+
|
|
153
|
+
|
|
154
|
+
|
|
155
|
+
cwd = os.getcwd()
|
|
156
|
+
|
|
157
|
+
|
|
158
|
+
|
|
159
|
+
SCORE_SIZE = 12
|
|
160
|
+
|
|
161
|
+
HIDDEN_UNIT_SIZE = 70
|
|
162
|
+
|
|
163
|
+
TRAIN_DATA_SIZE = 45
|
|
164
|
+
|
|
165
|
+
|
|
166
|
+
|
|
167
|
+
raw_input = numpy.loadtxt(open("test.csv"), delimiter=",")
|
|
168
|
+
|
|
169
|
+
[tensor, score] = numpy.hsplit(raw_input, [1])
|
|
170
|
+
|
|
171
|
+
[tensor_train, tensor_test] = numpy.vsplit(tensor, [TRAIN_DATA_SIZE])
|
|
172
|
+
|
|
173
|
+
[score_train, score_test] = numpy.vsplit(score, [TRAIN_DATA_SIZE])
|
|
174
|
+
|
|
175
|
+
#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
|
|
176
|
+
|
|
177
|
+
|
|
178
|
+
|
|
179
|
+
def inference(score_placeholder):
|
|
180
|
+
|
|
181
|
+
with tf.name_scope('hidden1') as scope:
|
|
182
|
+
|
|
183
|
+
hidden1_weight = tf.Variable(tf.truncated_normal([SCORE_SIZE, HIDDEN_UNIT_SIZE], stddev=0.1), name="hidden1_weight")
|
|
184
|
+
|
|
185
|
+
hidden1_bias = tf.Variable(tf.constant(0.1, shape=[HIDDEN_UNIT_SIZE]), name="hidden1_bias")
|
|
186
|
+
|
|
187
|
+
hidden1_output = tf.nn.relu(tf.matmul(score_placeholder, hidden1_weight) + hidden1_bias)
|
|
188
|
+
|
|
189
|
+
with tf.name_scope('output') as scope:
|
|
190
|
+
|
|
191
|
+
output_weight = tf.Variable(tf.truncated_normal([HIDDEN_UNIT_SIZE, 1], stddev=0.1), name="output_weight")
|
|
192
|
+
|
|
193
|
+
output_bias = tf.Variable(tf.constant(0.1, shape=[1]), name="output_bias")
|
|
194
|
+
|
|
195
|
+
output = tf.matmul(hidden1_output, output_weight) + output_bias
|
|
196
|
+
|
|
197
|
+
print(output)
|
|
198
|
+
|
|
199
|
+
return tf.nn.l2_normalize(output, 0)
|
|
200
|
+
|
|
201
|
+
|
|
202
|
+
|
|
203
|
+
def loss(output, tensor_placeholder, loss_label_placeholder):
|
|
204
|
+
|
|
205
|
+
with tf.name_scope('loss') as scope:
|
|
206
|
+
|
|
207
|
+
loss = tf.nn.l2_loss(output - tf.nn.l2_normalize(tensor_placeholder, 0))
|
|
208
|
+
|
|
209
|
+
tf.summary.scalar('loss_label_placeholder', loss)
|
|
210
|
+
|
|
211
|
+
return loss
|
|
212
|
+
|
|
213
|
+
|
|
214
|
+
|
|
215
|
+
def training(loss):
|
|
216
|
+
|
|
217
|
+
with tf.name_scope('training') as scope:
|
|
218
|
+
|
|
219
|
+
train_step = tf.train.GradientDescentOptimizer(0.01).minimize(loss)
|
|
220
|
+
|
|
221
|
+
return train_step
|
|
222
|
+
|
|
223
|
+
|
|
224
|
+
|
|
225
|
+
|
|
226
|
+
|
|
227
|
+
|
|
228
|
+
|
|
229
|
+
with tf.Graph().as_default():
|
|
230
|
+
|
|
231
|
+
tensor_placeholder = tf.placeholder("float", [None,1], name="tensor_placeholder")
|
|
232
|
+
|
|
233
|
+
score_placeholder = tf.placeholder("float", [None, SCORE_SIZE], name="score_placeholder")
|
|
234
|
+
|
|
235
|
+
loss_label_placeholder = tf.placeholder("string", name="loss_label_placeholder")
|
|
236
|
+
|
|
237
|
+
|
|
238
|
+
|
|
239
|
+
feed_dict_train={
|
|
240
|
+
|
|
241
|
+
tensor_placeholder: tensor_train,
|
|
242
|
+
|
|
243
|
+
score_placeholder: score_train,
|
|
244
|
+
|
|
245
|
+
loss_label_placeholder: "loss_train"
|
|
246
|
+
|
|
247
|
+
}
|
|
248
|
+
|
|
249
|
+
#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
|
|
250
|
+
|
|
251
|
+
|
|
252
|
+
|
|
253
|
+
feed_dict_test={
|
|
254
|
+
|
|
255
|
+
tensor_placeholder: tensor_test,
|
|
256
|
+
|
|
257
|
+
score_placeholder: score_test,
|
|
258
|
+
|
|
259
|
+
loss_label_placeholder: "loss_test"
|
|
260
|
+
|
|
261
|
+
}
|
|
262
|
+
|
|
263
|
+
|
|
264
|
+
|
|
265
|
+
output = inference(score_placeholder)
|
|
266
|
+
|
|
267
|
+
loss = loss(output, tensor_placeholder, loss_label_placeholder)
|
|
268
|
+
|
|
269
|
+
training_op = training(loss)
|
|
270
|
+
|
|
271
|
+
summary_op = tf.summary.merge_all()
|
|
272
|
+
|
|
273
|
+
init = tf.global_variables_initializer()
|
|
274
|
+
|
|
61
|
-
|
|
275
|
+
best_loss = float("inf")
|
|
276
|
+
|
|
277
|
+
|
|
278
|
+
|
|
279
|
+
with tf.Session() as sess:
|
|
280
|
+
|
|
281
|
+
summary_writer = tf.summary.FileWriter('data', graph_def=sess.graph_def)
|
|
282
|
+
|
|
283
|
+
sess.run(init)
|
|
284
|
+
|
|
285
|
+
for step in range(10000):
|
|
286
|
+
|
|
287
|
+
sess.run(training_op, feed_dict=feed_dict_train)
|
|
288
|
+
|
|
289
|
+
loss_test = sess.run(loss, feed_dict=feed_dict_test)
|
|
290
|
+
|
|
291
|
+
if loss_test < best_loss:
|
|
292
|
+
|
|
293
|
+
best_loss = loss_test
|
|
294
|
+
|
|
295
|
+
best_match = sess.run(output, feed_dict=feed_dict_test)
|
|
296
|
+
|
|
297
|
+
if step % 100 == 0:
|
|
298
|
+
|
|
299
|
+
summary_str = sess.run(summary_op, feed_dict=feed_dict_test)
|
|
300
|
+
|
|
301
|
+
summary_str += sess.run(summary_op, feed_dict=feed_dict_train)
|
|
302
|
+
|
|
303
|
+
summary_writer.add_summary(summary_str, step)
|
|
304
|
+
|
|
305
|
+
|
|
306
|
+
|
|
307
|
+
saver=tf.train.Saver()
|
|
308
|
+
|
|
309
|
+
saver.save(sess,cwd+'/model.ckpt')
|
|
310
|
+
|
|
311
|
+
print(cwd)
|
|
312
|
+
|
|
313
|
+
print(best_match)
|
|
314
|
+
|
|
315
|
+
print('Saved a model.')
|
|
316
|
+
|
|
317
|
+
sess.close()
|
|
318
|
+
|
|
319
|
+
|
|
320
|
+
|
|
321
|
+
with tf.Session() as sess2:
|
|
322
|
+
|
|
323
|
+
#変数の読み込み
|
|
324
|
+
|
|
325
|
+
#新しいデータ
|
|
326
|
+
|
|
327
|
+
TRAIN_DATA_SIZE2 = 0
|
|
328
|
+
|
|
329
|
+
test2 = numpy.loadtxt(open("one_record.csv"), delimiter=",")
|
|
330
|
+
|
|
331
|
+
[tensor2,score2] = numpy.hsplit(test2, [1])
|
|
332
|
+
|
|
333
|
+
#[tensor_train2,tensor_test2] = numpy.vsplit(tensor2, [TRAIN_DATA_SIZE2])
|
|
334
|
+
|
|
335
|
+
#[score_train2, score_test2] = numpy.vsplit(score2, [TRAIN_DATA_SIZE2])
|
|
336
|
+
|
|
337
|
+
#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
|
|
338
|
+
|
|
339
|
+
print(tensor2)
|
|
340
|
+
|
|
341
|
+
print(score2)
|
|
342
|
+
|
|
343
|
+
#モデルつくり
|
|
344
|
+
|
|
345
|
+
feed_dict_test2 = {
|
|
346
|
+
|
|
347
|
+
tensor_placeholder:tensor2,
|
|
62
348
|
|
|
63
349
|
score_placeholder:score2,
|
|
64
350
|
|
|
@@ -66,7 +352,7 @@
|
|
|
66
352
|
|
|
67
353
|
}
|
|
68
354
|
|
|
69
|
-
|
|
355
|
+
#復元して、損失関数で定まった、重みをもとに予想を行う関数にいれる
|
|
70
356
|
|
|
71
357
|
saver = tf.train.Saver()
|
|
72
358
|
|
|
@@ -87,259 +373,3 @@
|
|
|
87
373
|
sess2.close()
|
|
88
374
|
|
|
89
375
|
````
|
|
90
|
-
|
|
91
|
-
|
|
92
|
-
|
|
93
|
-
|
|
94
|
-
|
|
95
|
-
エラーコード
|
|
96
|
-
|
|
97
|
-
````
|
|
98
|
-
|
|
99
|
-
ValueError: Cannot feed value of shape (1,) for Tensor 'tensor_placeholder:0', which has shape '(?, 1)'
|
|
100
|
-
|
|
101
|
-
````
|
|
102
|
-
|
|
103
|
-
|
|
104
|
-
|
|
105
|
-
|
|
106
|
-
|
|
107
|
-
|
|
108
|
-
|
|
109
|
-
|
|
110
|
-
|
|
111
|
-
````
|
|
112
|
-
|
|
113
|
-
|
|
114
|
-
|
|
115
|
-
|
|
116
|
-
|
|
117
|
-
import tensorflow as tf
|
|
118
|
-
|
|
119
|
-
import numpy
|
|
120
|
-
|
|
121
|
-
import os
|
|
122
|
-
|
|
123
|
-
|
|
124
|
-
|
|
125
|
-
cwd = os.getcwd()
|
|
126
|
-
|
|
127
|
-
|
|
128
|
-
|
|
129
|
-
SCORE_SIZE = 12
|
|
130
|
-
|
|
131
|
-
HIDDEN_UNIT_SIZE = 70
|
|
132
|
-
|
|
133
|
-
TRAIN_DATA_SIZE = 45
|
|
134
|
-
|
|
135
|
-
|
|
136
|
-
|
|
137
|
-
raw_input = numpy.loadtxt(open("test.csv"), delimiter=",")
|
|
138
|
-
|
|
139
|
-
[tensor, score] = numpy.hsplit(raw_input, [1])
|
|
140
|
-
|
|
141
|
-
[tensor_train, tensor_test] = numpy.vsplit(tensor, [TRAIN_DATA_SIZE])
|
|
142
|
-
|
|
143
|
-
[score_train, score_test] = numpy.vsplit(score, [TRAIN_DATA_SIZE])
|
|
144
|
-
|
|
145
|
-
#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
|
|
146
|
-
|
|
147
|
-
|
|
148
|
-
|
|
149
|
-
def inference(score_placeholder):
|
|
150
|
-
|
|
151
|
-
with tf.name_scope('hidden1') as scope:
|
|
152
|
-
|
|
153
|
-
hidden1_weight = tf.Variable(tf.truncated_normal([SCORE_SIZE, HIDDEN_UNIT_SIZE], stddev=0.1), name="hidden1_weight")
|
|
154
|
-
|
|
155
|
-
hidden1_bias = tf.Variable(tf.constant(0.1, shape=[HIDDEN_UNIT_SIZE]), name="hidden1_bias")
|
|
156
|
-
|
|
157
|
-
hidden1_output = tf.nn.relu(tf.matmul(score_placeholder, hidden1_weight) + hidden1_bias)
|
|
158
|
-
|
|
159
|
-
with tf.name_scope('output') as scope:
|
|
160
|
-
|
|
161
|
-
output_weight = tf.Variable(tf.truncated_normal([HIDDEN_UNIT_SIZE, 1], stddev=0.1), name="output_weight")
|
|
162
|
-
|
|
163
|
-
output_bias = tf.Variable(tf.constant(0.1, shape=[1]), name="output_bias")
|
|
164
|
-
|
|
165
|
-
output = tf.matmul(hidden1_output, output_weight) + output_bias
|
|
166
|
-
|
|
167
|
-
print(output)
|
|
168
|
-
|
|
169
|
-
return tf.nn.l2_normalize(output, 0)
|
|
170
|
-
|
|
171
|
-
|
|
172
|
-
|
|
173
|
-
def loss(output, tensor_placeholder, loss_label_placeholder):
|
|
174
|
-
|
|
175
|
-
with tf.name_scope('loss') as scope:
|
|
176
|
-
|
|
177
|
-
loss = tf.nn.l2_loss(output - tf.nn.l2_normalize(tensor_placeholder, 0))
|
|
178
|
-
|
|
179
|
-
tf.summary.scalar('loss_label_placeholder', loss)
|
|
180
|
-
|
|
181
|
-
return loss
|
|
182
|
-
|
|
183
|
-
|
|
184
|
-
|
|
185
|
-
def training(loss):
|
|
186
|
-
|
|
187
|
-
with tf.name_scope('training') as scope:
|
|
188
|
-
|
|
189
|
-
train_step = tf.train.GradientDescentOptimizer(0.01).minimize(loss)
|
|
190
|
-
|
|
191
|
-
return train_step
|
|
192
|
-
|
|
193
|
-
|
|
194
|
-
|
|
195
|
-
|
|
196
|
-
|
|
197
|
-
|
|
198
|
-
|
|
199
|
-
with tf.Graph().as_default():
|
|
200
|
-
|
|
201
|
-
tensor_placeholder = tf.placeholder("float", [None,1], name="tensor_placeholder")
|
|
202
|
-
|
|
203
|
-
score_placeholder = tf.placeholder("float", [None, SCORE_SIZE], name="score_placeholder")
|
|
204
|
-
|
|
205
|
-
loss_label_placeholder = tf.placeholder("string", name="loss_label_placeholder")
|
|
206
|
-
|
|
207
|
-
|
|
208
|
-
|
|
209
|
-
feed_dict_train={
|
|
210
|
-
|
|
211
|
-
tensor_placeholder: tensor_train,
|
|
212
|
-
|
|
213
|
-
score_placeholder: score_train,
|
|
214
|
-
|
|
215
|
-
loss_label_placeholder: "loss_train"
|
|
216
|
-
|
|
217
|
-
}
|
|
218
|
-
|
|
219
|
-
#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
|
|
220
|
-
|
|
221
|
-
|
|
222
|
-
|
|
223
|
-
feed_dict_test={
|
|
224
|
-
|
|
225
|
-
tensor_placeholder: tensor_test,
|
|
226
|
-
|
|
227
|
-
score_placeholder: score_test,
|
|
228
|
-
|
|
229
|
-
loss_label_placeholder: "loss_test"
|
|
230
|
-
|
|
231
|
-
}
|
|
232
|
-
|
|
233
|
-
|
|
234
|
-
|
|
235
|
-
output = inference(score_placeholder)
|
|
236
|
-
|
|
237
|
-
loss = loss(output, tensor_placeholder, loss_label_placeholder)
|
|
238
|
-
|
|
239
|
-
training_op = training(loss)
|
|
240
|
-
|
|
241
|
-
summary_op = tf.summary.merge_all()
|
|
242
|
-
|
|
243
|
-
init = tf.global_variables_initializer()
|
|
244
|
-
|
|
245
|
-
best_loss = float("inf")
|
|
246
|
-
|
|
247
|
-
|
|
248
|
-
|
|
249
|
-
with tf.Session() as sess:
|
|
250
|
-
|
|
251
|
-
summary_writer = tf.summary.FileWriter('data', graph_def=sess.graph_def)
|
|
252
|
-
|
|
253
|
-
sess.run(init)
|
|
254
|
-
|
|
255
|
-
for step in range(10000):
|
|
256
|
-
|
|
257
|
-
sess.run(training_op, feed_dict=feed_dict_train)
|
|
258
|
-
|
|
259
|
-
loss_test = sess.run(loss, feed_dict=feed_dict_test)
|
|
260
|
-
|
|
261
|
-
if loss_test < best_loss:
|
|
262
|
-
|
|
263
|
-
best_loss = loss_test
|
|
264
|
-
|
|
265
|
-
best_match = sess.run(output, feed_dict=feed_dict_test)
|
|
266
|
-
|
|
267
|
-
if step % 100 == 0:
|
|
268
|
-
|
|
269
|
-
summary_str = sess.run(summary_op, feed_dict=feed_dict_test)
|
|
270
|
-
|
|
271
|
-
summary_str += sess.run(summary_op, feed_dict=feed_dict_train)
|
|
272
|
-
|
|
273
|
-
summary_writer.add_summary(summary_str, step)
|
|
274
|
-
|
|
275
|
-
|
|
276
|
-
|
|
277
|
-
saver=tf.train.Saver()
|
|
278
|
-
|
|
279
|
-
saver.save(sess,cwd+'/model.ckpt')
|
|
280
|
-
|
|
281
|
-
print(cwd)
|
|
282
|
-
|
|
283
|
-
print(best_match)
|
|
284
|
-
|
|
285
|
-
print('Saved a model.')
|
|
286
|
-
|
|
287
|
-
sess.close()
|
|
288
|
-
|
|
289
|
-
|
|
290
|
-
|
|
291
|
-
with tf.Session() as sess2:
|
|
292
|
-
|
|
293
|
-
#変数の読み込み
|
|
294
|
-
|
|
295
|
-
#新しいデータ
|
|
296
|
-
|
|
297
|
-
TRAIN_DATA_SIZE2 = 0
|
|
298
|
-
|
|
299
|
-
test2 = numpy.loadtxt(open("one_record.csv"), delimiter=",")
|
|
300
|
-
|
|
301
|
-
[tensor2,score2] = numpy.hsplit(test2, [1])
|
|
302
|
-
|
|
303
|
-
#[tensor_train2,tensor_test2] = numpy.vsplit(tensor2, [TRAIN_DATA_SIZE2])
|
|
304
|
-
|
|
305
|
-
#[score_train2, score_test2] = numpy.vsplit(score2, [TRAIN_DATA_SIZE2])
|
|
306
|
-
|
|
307
|
-
#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
|
|
308
|
-
|
|
309
|
-
print(tensor2)
|
|
310
|
-
|
|
311
|
-
print(score2)
|
|
312
|
-
|
|
313
|
-
#モデルつくり
|
|
314
|
-
|
|
315
|
-
feed_dict_test2 = {
|
|
316
|
-
|
|
317
|
-
tensor_placeholder:tensor2,
|
|
318
|
-
|
|
319
|
-
score_placeholder:score2,
|
|
320
|
-
|
|
321
|
-
loss_label_placeholder:"loss_test2"
|
|
322
|
-
|
|
323
|
-
}
|
|
324
|
-
|
|
325
|
-
#復元して、損失関数で定まった、重みをもとに予想を行う関数にいれる
|
|
326
|
-
|
|
327
|
-
saver = tf.train.Saver()
|
|
328
|
-
|
|
329
|
-
cwd = os.getcwd()
|
|
330
|
-
|
|
331
|
-
saver.restore(sess2,cwd + "/model.ckpt")
|
|
332
|
-
|
|
333
|
-
|
|
334
|
-
|
|
335
|
-
print("recover")
|
|
336
|
-
|
|
337
|
-
best_match2 = sess2.run(output, feed_dict=feed_dict_test2)
|
|
338
|
-
|
|
339
|
-
print(best_match2)
|
|
340
|
-
|
|
341
|
-
print("fin")
|
|
342
|
-
|
|
343
|
-
sess2.close()
|
|
344
|
-
|
|
345
|
-
````
|
2
修正
test
CHANGED
|
File without changes
|
test
CHANGED
|
@@ -10,9 +10,13 @@
|
|
|
10
10
|
|
|
11
11
|
1,学習用csvファイル：13列×50行のデータになっており、最初の一列目に正解データが記入されています。
|
|
12
12
|
|
|
13
|
+
このようなデータ形式をとっています。（00001000のような2-13列はありません）
|
|
14
|
+
|
|
15
|
+
[https://gist.github.com/sergeant-wizard/b2c548fbd3b3a01b23ca](https://gist.github.com/sergeant-wizard/b2c548fbd3b3a01b23ca)
|
|
16
|
+
|
|
13
17
|
2,学習の際には一度hpsplitを行って分割して[tensor,score]に分ける
|
|
14
18
|
|
|
15
|
-
3,50行のうち、45行を学習にあて、5行をテストに分ける.
|
|
19
|
+
3,50行のうち、45行を学習にあて、5行をテスト予想に分ける.
|
|
16
20
|
|
|
17
21
|
**4,学習したのちに復元を行い、実践データを入れる。**
|
|
18
22
|
|
|
@@ -20,11 +24,9 @@
|
|
|
20
24
|
|
|
21
25
|
→これでエラーがでたので、実践データの一列目に０を入れて13列にして対応した。
|
|
22
26
|
|
|
23
|
-
→成功して復元可能。ユーザーの入力に対して予想したいと考えているので、１行のデータでできるようにしたい。
|
|
27
|
+
→成功して復元可能。**ユーザーの入力に対して予想したいと考えているので、１行のデータでできるようにしたい。**
|
|
24
|
-
|
|
28
|
+
|
|
25
|
-
*実践データが１行のみの時にvsplitが使えないので、
|
|
29
|
+
*実践データが１行のみの時にvsplitが使えないので、13列をを1:12にhpsplitを行って代入するも、行列の数がtensor_placeholderとcsvと対応していないみたいでエラーになります。このshapeの形を治すのに苦戦しており、お力をかしていただきたいです。
|
|
26
|
-
|
|
27
|
-
|
|
28
30
|
|
|
29
31
|
|
|
30
32
|
|
|
@@ -52,13 +54,241 @@
|
|
|
52
54
|
|
|
53
55
|
|
|
54
56
|
|
|
57
|
+
Tensorflow：一度学習したものの損失関数の値を保持・復元して、新たな入力に対して予測を行いたい
|
|
58
|
+
|
|
59
|
+
[https://teratail.com/questions/82450](https://teratail.com/questions/82450)
|
|
60
|
+
|
|
61
|
+
test2を入れることで計算できていました。どう直せばいいでしょうか。
|
|
62
|
+
|
|
63
|
+
score_placeholder:score2,
|
|
64
|
+
|
|
65
|
+
loss_label_placeholder:"loss_test2"
|
|
66
|
+
|
|
67
|
+
}
|
|
68
|
+
|
|
69
|
+
|
|
70
|
+
|
|
71
|
+
saver = tf.train.Saver()
|
|
72
|
+
|
|
73
|
+
cwd = os.getcwd()
|
|
74
|
+
|
|
55
|
-
|
|
75
|
+
saver.restore(sess2,cwd + "/model.ckpt")
|
|
76
|
+
|
|
77
|
+
|
|
78
|
+
|
|
56
|
-
|
|
79
|
+
print("recover")
|
|
80
|
+
|
|
81
|
+
best_match2 = sess2.run(output, feed_dict=feed_dict_test2)
|
|
82
|
+
|
|
83
|
+
print(best_match2)
|
|
84
|
+
|
|
85
|
+
print("fin")
|
|
86
|
+
|
|
87
|
+
sess2.close()
|
|
88
|
+
|
|
57
|
-
````
|
|
89
|
+
````
|
|
90
|
+
|
|
91
|
+
|
|
92
|
+
|
|
93
|
+
|
|
94
|
+
|
|
58
|
-
|
|
95
|
+
エラーコード
|
|
96
|
+
|
|
59
|
-
|
|
97
|
+
````
|
|
98
|
+
|
|
60
|
-
|
|
99
|
+
ValueError: Cannot feed value of shape (1,) for Tensor 'tensor_placeholder:0', which has shape '(?, 1)'
|
|
100
|
+
|
|
101
|
+
````
|
|
102
|
+
|
|
103
|
+
|
|
104
|
+
|
|
105
|
+
|
|
106
|
+
|
|
107
|
+
|
|
108
|
+
|
|
109
|
+
|
|
110
|
+
|
|
111
|
+
````
|
|
112
|
+
|
|
113
|
+
|
|
114
|
+
|
|
115
|
+
|
|
116
|
+
|
|
117
|
+
import tensorflow as tf
|
|
118
|
+
|
|
119
|
+
import numpy
|
|
120
|
+
|
|
121
|
+
import os
|
|
122
|
+
|
|
123
|
+
|
|
124
|
+
|
|
125
|
+
cwd = os.getcwd()
|
|
126
|
+
|
|
127
|
+
|
|
128
|
+
|
|
129
|
+
SCORE_SIZE = 12
|
|
130
|
+
|
|
131
|
+
HIDDEN_UNIT_SIZE = 70
|
|
132
|
+
|
|
133
|
+
TRAIN_DATA_SIZE = 45
|
|
134
|
+
|
|
135
|
+
|
|
136
|
+
|
|
137
|
+
raw_input = numpy.loadtxt(open("test.csv"), delimiter=",")
|
|
138
|
+
|
|
139
|
+
[tensor, score] = numpy.hsplit(raw_input, [1])
|
|
140
|
+
|
|
141
|
+
[tensor_train, tensor_test] = numpy.vsplit(tensor, [TRAIN_DATA_SIZE])
|
|
142
|
+
|
|
143
|
+
[score_train, score_test] = numpy.vsplit(score, [TRAIN_DATA_SIZE])
|
|
144
|
+
|
|
145
|
+
#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
|
|
146
|
+
|
|
147
|
+
|
|
148
|
+
|
|
149
|
+
def inference(score_placeholder):
|
|
150
|
+
|
|
151
|
+
with tf.name_scope('hidden1') as scope:
|
|
152
|
+
|
|
153
|
+
hidden1_weight = tf.Variable(tf.truncated_normal([SCORE_SIZE, HIDDEN_UNIT_SIZE], stddev=0.1), name="hidden1_weight")
|
|
154
|
+
|
|
155
|
+
hidden1_bias = tf.Variable(tf.constant(0.1, shape=[HIDDEN_UNIT_SIZE]), name="hidden1_bias")
|
|
156
|
+
|
|
157
|
+
hidden1_output = tf.nn.relu(tf.matmul(score_placeholder, hidden1_weight) + hidden1_bias)
|
|
158
|
+
|
|
159
|
+
with tf.name_scope('output') as scope:
|
|
160
|
+
|
|
161
|
+
output_weight = tf.Variable(tf.truncated_normal([HIDDEN_UNIT_SIZE, 1], stddev=0.1), name="output_weight")
|
|
162
|
+
|
|
163
|
+
output_bias = tf.Variable(tf.constant(0.1, shape=[1]), name="output_bias")
|
|
164
|
+
|
|
165
|
+
output = tf.matmul(hidden1_output, output_weight) + output_bias
|
|
166
|
+
|
|
167
|
+
print(output)
|
|
168
|
+
|
|
169
|
+
return tf.nn.l2_normalize(output, 0)
|
|
170
|
+
|
|
171
|
+
|
|
172
|
+
|
|
173
|
+
def loss(output, tensor_placeholder, loss_label_placeholder):
|
|
174
|
+
|
|
175
|
+
with tf.name_scope('loss') as scope:
|
|
176
|
+
|
|
177
|
+
loss = tf.nn.l2_loss(output - tf.nn.l2_normalize(tensor_placeholder, 0))
|
|
178
|
+
|
|
179
|
+
tf.summary.scalar('loss_label_placeholder', loss)
|
|
180
|
+
|
|
181
|
+
return loss
|
|
182
|
+
|
|
183
|
+
|
|
184
|
+
|
|
185
|
+
def training(loss):
|
|
186
|
+
|
|
187
|
+
with tf.name_scope('training') as scope:
|
|
188
|
+
|
|
189
|
+
train_step = tf.train.GradientDescentOptimizer(0.01).minimize(loss)
|
|
190
|
+
|
|
191
|
+
return train_step
|
|
192
|
+
|
|
193
|
+
|
|
194
|
+
|
|
195
|
+
|
|
196
|
+
|
|
197
|
+
|
|
198
|
+
|
|
199
|
+
with tf.Graph().as_default():
|
|
200
|
+
|
|
201
|
+
tensor_placeholder = tf.placeholder("float", [None,1], name="tensor_placeholder")
|
|
202
|
+
|
|
203
|
+
score_placeholder = tf.placeholder("float", [None, SCORE_SIZE], name="score_placeholder")
|
|
204
|
+
|
|
205
|
+
loss_label_placeholder = tf.placeholder("string", name="loss_label_placeholder")
|
|
206
|
+
|
|
207
|
+
|
|
208
|
+
|
|
209
|
+
feed_dict_train={
|
|
210
|
+
|
|
211
|
+
tensor_placeholder: tensor_train,
|
|
212
|
+
|
|
213
|
+
score_placeholder: score_train,
|
|
214
|
+
|
|
215
|
+
loss_label_placeholder: "loss_train"
|
|
216
|
+
|
|
217
|
+
}
|
|
218
|
+
|
|
219
|
+
#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
|
|
220
|
+
|
|
221
|
+
|
|
222
|
+
|
|
223
|
+
feed_dict_test={
|
|
224
|
+
|
|
225
|
+
tensor_placeholder: tensor_test,
|
|
226
|
+
|
|
227
|
+
score_placeholder: score_test,
|
|
228
|
+
|
|
229
|
+
loss_label_placeholder: "loss_test"
|
|
230
|
+
|
|
231
|
+
}
|
|
232
|
+
|
|
233
|
+
|
|
234
|
+
|
|
235
|
+
output = inference(score_placeholder)
|
|
236
|
+
|
|
237
|
+
loss = loss(output, tensor_placeholder, loss_label_placeholder)
|
|
238
|
+
|
|
239
|
+
training_op = training(loss)
|
|
240
|
+
|
|
241
|
+
summary_op = tf.summary.merge_all()
|
|
242
|
+
|
|
243
|
+
init = tf.global_variables_initializer()
|
|
244
|
+
|
|
245
|
+
best_loss = float("inf")
|
|
246
|
+
|
|
247
|
+
|
|
248
|
+
|
|
249
|
+
with tf.Session() as sess:
|
|
250
|
+
|
|
251
|
+
summary_writer = tf.summary.FileWriter('data', graph_def=sess.graph_def)
|
|
252
|
+
|
|
253
|
+
sess.run(init)
|
|
254
|
+
|
|
255
|
+
for step in range(10000):
|
|
256
|
+
|
|
257
|
+
sess.run(training_op, feed_dict=feed_dict_train)
|
|
258
|
+
|
|
259
|
+
loss_test = sess.run(loss, feed_dict=feed_dict_test)
|
|
260
|
+
|
|
261
|
+
if loss_test < best_loss:
|
|
262
|
+
|
|
263
|
+
best_loss = loss_test
|
|
264
|
+
|
|
265
|
+
best_match = sess.run(output, feed_dict=feed_dict_test)
|
|
266
|
+
|
|
267
|
+
if step % 100 == 0:
|
|
268
|
+
|
|
269
|
+
summary_str = sess.run(summary_op, feed_dict=feed_dict_test)
|
|
270
|
+
|
|
271
|
+
summary_str += sess.run(summary_op, feed_dict=feed_dict_train)
|
|
272
|
+
|
|
273
|
+
summary_writer.add_summary(summary_str, step)
|
|
274
|
+
|
|
275
|
+
|
|
276
|
+
|
|
277
|
+
saver=tf.train.Saver()
|
|
278
|
+
|
|
279
|
+
saver.save(sess,cwd+'/model.ckpt')
|
|
280
|
+
|
|
281
|
+
print(cwd)
|
|
282
|
+
|
|
283
|
+
print(best_match)
|
|
284
|
+
|
|
285
|
+
print('Saved a model.')
|
|
286
|
+
|
|
287
|
+
sess.close()
|
|
288
|
+
|
|
289
|
+
|
|
290
|
+
|
|
61
|
-
with tf.Session() as sess2:
|
|
291
|
+
with tf.Session() as sess2:
|
|
62
292
|
|
|
63
293
|
#変数の読み込み
|
|
64
294
|
|
|
@@ -66,8 +296,6 @@
|
|
|
66
296
|
|
|
67
297
|
TRAIN_DATA_SIZE2 = 0
|
|
68
298
|
|
|
69
|
-
|
|
70
|
-
|
|
71
299
|
test2 = numpy.loadtxt(open("one_record.csv"), delimiter=",")
|
|
72
300
|
|
|
73
301
|
[tensor2,score2] = numpy.hsplit(test2, [1])
|
|
@@ -115,259 +343,3 @@
|
|
|
115
343
|
sess2.close()
|
|
116
344
|
|
|
117
345
|
````
|
|
118
|
-
|
|
119
|
-
|
|
120
|
-
|
|
121
|
-
|
|
122
|
-
|
|
123
|
-
エラーコード
|
|
124
|
-
|
|
125
|
-
````
|
|
126
|
-
|
|
127
|
-
ValueError: Cannot feed value of shape (1,) for Tensor 'tensor_placeholder:0', which has shape '(?, 1)'
|
|
128
|
-
|
|
129
|
-
````
|
|
130
|
-
|
|
131
|
-
|
|
132
|
-
|
|
133
|
-
|
|
134
|
-
|
|
135
|
-
|
|
136
|
-
|
|
137
|
-
|
|
138
|
-
|
|
139
|
-
````
|
|
140
|
-
|
|
141
|
-
|
|
142
|
-
|
|
143
|
-
|
|
144
|
-
|
|
145
|
-
import tensorflow as tf
|
|
146
|
-
|
|
147
|
-
import numpy
|
|
148
|
-
|
|
149
|
-
import os
|
|
150
|
-
|
|
151
|
-
|
|
152
|
-
|
|
153
|
-
cwd = os.getcwd()
|
|
154
|
-
|
|
155
|
-
|
|
156
|
-
|
|
157
|
-
SCORE_SIZE = 12
|
|
158
|
-
|
|
159
|
-
HIDDEN_UNIT_SIZE = 70
|
|
160
|
-
|
|
161
|
-
TRAIN_DATA_SIZE = 45
|
|
162
|
-
|
|
163
|
-
|
|
164
|
-
|
|
165
|
-
raw_input = numpy.loadtxt(open("test.csv"), delimiter=",")
|
|
166
|
-
|
|
167
|
-
[tensor, score] = numpy.hsplit(raw_input, [1])
|
|
168
|
-
|
|
169
|
-
[tensor_train, tensor_test] = numpy.vsplit(tensor, [TRAIN_DATA_SIZE])
|
|
170
|
-
|
|
171
|
-
[score_train, score_test] = numpy.vsplit(score, [TRAIN_DATA_SIZE])
|
|
172
|
-
|
|
173
|
-
#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
|
|
174
|
-
|
|
175
|
-
|
|
176
|
-
|
|
177
|
-
def inference(score_placeholder):
|
|
178
|
-
|
|
179
|
-
with tf.name_scope('hidden1') as scope:
|
|
180
|
-
|
|
181
|
-
hidden1_weight = tf.Variable(tf.truncated_normal([SCORE_SIZE, HIDDEN_UNIT_SIZE], stddev=0.1), name="hidden1_weight")
|
|
182
|
-
|
|
183
|
-
hidden1_bias = tf.Variable(tf.constant(0.1, shape=[HIDDEN_UNIT_SIZE]), name="hidden1_bias")
|
|
184
|
-
|
|
185
|
-
hidden1_output = tf.nn.relu(tf.matmul(score_placeholder, hidden1_weight) + hidden1_bias)
|
|
186
|
-
|
|
187
|
-
with tf.name_scope('output') as scope:
|
|
188
|
-
|
|
189
|
-
output_weight = tf.Variable(tf.truncated_normal([HIDDEN_UNIT_SIZE, 1], stddev=0.1), name="output_weight")
|
|
190
|
-
|
|
191
|
-
output_bias = tf.Variable(tf.constant(0.1, shape=[1]), name="output_bias")
|
|
192
|
-
|
|
193
|
-
output = tf.matmul(hidden1_output, output_weight) + output_bias
|
|
194
|
-
|
|
195
|
-
print(output)
|
|
196
|
-
|
|
197
|
-
return tf.nn.l2_normalize(output, 0)
|
|
198
|
-
|
|
199
|
-
|
|
200
|
-
|
|
201
|
-
def loss(output, tensor_placeholder, loss_label_placeholder):
|
|
202
|
-
|
|
203
|
-
with tf.name_scope('loss') as scope:
|
|
204
|
-
|
|
205
|
-
loss = tf.nn.l2_loss(output - tf.nn.l2_normalize(tensor_placeholder, 0))
|
|
206
|
-
|
|
207
|
-
tf.summary.scalar('loss_label_placeholder', loss)
|
|
208
|
-
|
|
209
|
-
return loss
|
|
210
|
-
|
|
211
|
-
|
|
212
|
-
|
|
213
|
-
def training(loss):
|
|
214
|
-
|
|
215
|
-
with tf.name_scope('training') as scope:
|
|
216
|
-
|
|
217
|
-
train_step = tf.train.GradientDescentOptimizer(0.01).minimize(loss)
|
|
218
|
-
|
|
219
|
-
return train_step
|
|
220
|
-
|
|
221
|
-
|
|
222
|
-
|
|
223
|
-
|
|
224
|
-
|
|
225
|
-
|
|
226
|
-
|
|
227
|
-
with tf.Graph().as_default():
|
|
228
|
-
|
|
229
|
-
tensor_placeholder = tf.placeholder("float", [None,1], name="tensor_placeholder")
|
|
230
|
-
|
|
231
|
-
score_placeholder = tf.placeholder("float", [None, SCORE_SIZE], name="score_placeholder")
|
|
232
|
-
|
|
233
|
-
loss_label_placeholder = tf.placeholder("string", name="loss_label_placeholder")
|
|
234
|
-
|
|
235
|
-
|
|
236
|
-
|
|
237
|
-
feed_dict_train={
|
|
238
|
-
|
|
239
|
-
tensor_placeholder: tensor_train,
|
|
240
|
-
|
|
241
|
-
score_placeholder: score_train,
|
|
242
|
-
|
|
243
|
-
loss_label_placeholder: "loss_train"
|
|
244
|
-
|
|
245
|
-
}
|
|
246
|
-
|
|
247
|
-
#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
|
|
248
|
-
|
|
249
|
-
|
|
250
|
-
|
|
251
|
-
feed_dict_test={
|
|
252
|
-
|
|
253
|
-
tensor_placeholder: tensor_test,
|
|
254
|
-
|
|
255
|
-
score_placeholder: score_test,
|
|
256
|
-
|
|
257
|
-
loss_label_placeholder: "loss_test"
|
|
258
|
-
|
|
259
|
-
}
|
|
260
|
-
|
|
261
|
-
|
|
262
|
-
|
|
263
|
-
output = inference(score_placeholder)
|
|
264
|
-
|
|
265
|
-
loss = loss(output, tensor_placeholder, loss_label_placeholder)
|
|
266
|
-
|
|
267
|
-
training_op = training(loss)
|
|
268
|
-
|
|
269
|
-
summary_op = tf.summary.merge_all()
|
|
270
|
-
|
|
271
|
-
init = tf.global_variables_initializer()
|
|
272
|
-
|
|
273
|
-
best_loss = float("inf")
|
|
274
|
-
|
|
275
|
-
|
|
276
|
-
|
|
277
|
-
with tf.Session() as sess:
|
|
278
|
-
|
|
279
|
-
summary_writer = tf.summary.FileWriter('data', graph_def=sess.graph_def)
|
|
280
|
-
|
|
281
|
-
sess.run(init)
|
|
282
|
-
|
|
283
|
-
for step in range(10000):
|
|
284
|
-
|
|
285
|
-
sess.run(training_op, feed_dict=feed_dict_train)
|
|
286
|
-
|
|
287
|
-
loss_test = sess.run(loss, feed_dict=feed_dict_test)
|
|
288
|
-
|
|
289
|
-
if loss_test < best_loss:
|
|
290
|
-
|
|
291
|
-
best_loss = loss_test
|
|
292
|
-
|
|
293
|
-
best_match = sess.run(output, feed_dict=feed_dict_test)
|
|
294
|
-
|
|
295
|
-
if step % 100 == 0:
|
|
296
|
-
|
|
297
|
-
summary_str = sess.run(summary_op, feed_dict=feed_dict_test)
|
|
298
|
-
|
|
299
|
-
summary_str += sess.run(summary_op, feed_dict=feed_dict_train)
|
|
300
|
-
|
|
301
|
-
summary_writer.add_summary(summary_str, step)
|
|
302
|
-
|
|
303
|
-
|
|
304
|
-
|
|
305
|
-
saver=tf.train.Saver()
|
|
306
|
-
|
|
307
|
-
saver.save(sess,cwd+'/model.ckpt')
|
|
308
|
-
|
|
309
|
-
print(cwd)
|
|
310
|
-
|
|
311
|
-
print(best_match)
|
|
312
|
-
|
|
313
|
-
print('Saved a model.')
|
|
314
|
-
|
|
315
|
-
sess.close()
|
|
316
|
-
|
|
317
|
-
|
|
318
|
-
|
|
319
|
-
with tf.Session() as sess2:
|
|
320
|
-
|
|
321
|
-
#変数の読み込み
|
|
322
|
-
|
|
323
|
-
#新しいデータ
|
|
324
|
-
|
|
325
|
-
TRAIN_DATA_SIZE2 = 0
|
|
326
|
-
|
|
327
|
-
test2 = numpy.loadtxt(open("one_record.csv"), delimiter=",")
|
|
328
|
-
|
|
329
|
-
[tensor2,score2] = numpy.hsplit(test2, [1])
|
|
330
|
-
|
|
331
|
-
#[tensor_train2,tensor_test2] = numpy.vsplit(tensor2, [TRAIN_DATA_SIZE2])
|
|
332
|
-
|
|
333
|
-
#[score_train2, score_test2] = numpy.vsplit(score2, [TRAIN_DATA_SIZE2])
|
|
334
|
-
|
|
335
|
-
#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
|
|
336
|
-
|
|
337
|
-
print(tensor2)
|
|
338
|
-
|
|
339
|
-
print(score2)
|
|
340
|
-
|
|
341
|
-
#モデルつくり
|
|
342
|
-
|
|
343
|
-
feed_dict_test2 = {
|
|
344
|
-
|
|
345
|
-
tensor_placeholder:tensor2,
|
|
346
|
-
|
|
347
|
-
score_placeholder:score2,
|
|
348
|
-
|
|
349
|
-
loss_label_placeholder:"loss_test2"
|
|
350
|
-
|
|
351
|
-
}
|
|
352
|
-
|
|
353
|
-
#復元して、損失関数で定まった、重みをもとに予想を行う関数にいれる
|
|
354
|
-
|
|
355
|
-
saver = tf.train.Saver()
|
|
356
|
-
|
|
357
|
-
cwd = os.getcwd()
|
|
358
|
-
|
|
359
|
-
saver.restore(sess2,cwd + "/model.ckpt")
|
|
360
|
-
|
|
361
|
-
|
|
362
|
-
|
|
363
|
-
print("recover")
|
|
364
|
-
|
|
365
|
-
best_match2 = sess2.run(output, feed_dict=feed_dict_test2)
|
|
366
|
-
|
|
367
|
-
print(best_match2)
|
|
368
|
-
|
|
369
|
-
print("fin")
|
|
370
|
-
|
|
371
|
-
sess2.close()
|
|
372
|
-
|
|
373
|
-
````
|
1
修正
test
CHANGED
|
File without changes
|
test
CHANGED
|
@@ -8,6 +8,32 @@
|
|
|
8
8
|
|
|
9
9
|
|
|
10
10
|
|
|
11
|
+
1,学習用csvファイル：13列×50行のデータになっており、最初の一列目に正解データが記入されています。
|
|
12
|
+
|
|
13
|
+
2,学習の際には一度hpsplitを行って分割して[tensor,score]に分ける
|
|
14
|
+
|
|
15
|
+
3,50行のうち、45行を学習にあて、5行をテストに分ける.
|
|
16
|
+
|
|
17
|
+
**4,学習したのちに復元を行い、実践データを入れる。**
|
|
18
|
+
|
|
19
|
+
*実践データは正解値を含まないので学習用から一つ列の減った12列×？行になる。
|
|
20
|
+
|
|
21
|
+
→これでエラーがでたので、実践データの一列目に０を入れて13列にして対応した。
|
|
22
|
+
|
|
23
|
+
→成功して復元可能。ユーザーの入力に対して予想したいと考えているので、１行のデータでできるようにしたい。
|
|
24
|
+
|
|
25
|
+
*実践データが１行のみの時にvsplitが使えないので、今のコードのコメントをつけさせていただいています。
|
|
26
|
+
|
|
27
|
+
|
|
28
|
+
|
|
29
|
+
|
|
30
|
+
|
|
31
|
+
|
|
32
|
+
|
|
33
|
+
**
|
|
34
|
+
|
|
35
|
+
|
|
36
|
+
|
|
11
37
|
|
|
12
38
|
|
|
13
39
|
・参考リンク
|
|
@@ -28,40 +54,6 @@
|
|
|
28
54
|
|
|
29
55
|
・実行環境はwindows10+Anaconda+python3.5+tensorflow1.0~になります
|
|
30
56
|
|
|
31
|
-
|
|
32
|
-
|
|
33
|
-
現状の理解・躓いていると思っている箇所
|
|
34
|
-
|
|
35
|
-
*修正7/1 17:02
|
|
36
|
-
|
|
37
|
-
|
|
38
|
-
|
|
39
|
-
|
|
40
|
-
|
|
41
|
-
色々な点を修正させていただき新たな疑問として
|
|
42
|
-
|
|
43
|
-
|
|
44
|
-
|
|
45
|
-
復元したモデルを再利用する流れについては
|
|
46
|
-
|
|
47
|
-
1.変数の読み込み(csvファイル)
|
|
48
|
-
|
|
49
|
-
2.使えるデータ形式に変化（）
|
|
50
|
-
|
|
51
|
-
3.学習したときに使ったinterface(score_placeholder)で使える形にモデルを作る
|
|
52
|
-
|
|
53
|
-
4.meta.ckptを復元して、重みをもとに予想を行う関数(interface)にいれる
|
|
54
|
-
|
|
55
|
-
5,sess.runを行い予想値をプリントする。
|
|
56
|
-
|
|
57
|
-
|
|
58
|
-
|
|
59
|
-
**ただ、実践データは学習データと異なり正解のデータがないのでcsvファイルの列も一つ少ないので、モデルの組み方が異なると、最初に学習の時に使ったinterface（）関数をそのまま使えないのでwith tf.Session() as sess2:ないで新たに関数を定義したらmodel.ckptの値が反映できないのではないかとそのあたりで躓いています。
|
|
60
|
-
|
|
61
|
-
現状の実践データ正解データも込みで行っていますがエラーが発生しておりなかなかうまくいきません。よろしくお願いいたします
|
|
62
|
-
|
|
63
|
-
**
|
|
64
|
-
|
|
65
57
|
````
|
|
66
58
|
|
|
67
59
|
|
|
@@ -74,6 +66,8 @@
|
|
|
74
66
|
|
|
75
67
|
TRAIN_DATA_SIZE2 = 0
|
|
76
68
|
|
|
69
|
+
|
|
70
|
+
|
|
77
71
|
test2 = numpy.loadtxt(open("one_record.csv"), delimiter=",")
|
|
78
72
|
|
|
79
73
|
[tensor2,score2] = numpy.hsplit(test2, [1])
|