teratail
回答率
85.35%
teratail header banner
teratail header banner
質問するログイン新規登録
トップに関する質問python keras モデル構築について

編集履歴

質問編集履歴

2

モデルをクラスにしたところ新たな問題が出ました

2020/10/21 06:28

投稿

vurie
vurie

スコア0

test CHANGED
File without changes
test CHANGED
@@ -1,18 +1,16 @@
1
1
  ### 前提・実現したいこと
2
2
 
3
- 分類を行う人工知能を作成したく、kerasを使って構築をしていますが、現在以下のようなエラーが発生しています
3
+ 分類を行う人工知能の出力を時系列データとして扱う人工知能モデルの開発
4
4
 
5
5
 
6
6
 
7
7
  ### 発生している問題・エラーメッセージ
8
8
 
9
- batch_tの方をdense_3_に投げてエラーが図れているように思うのですが、原因がわかりません
9
+ モデルを連結して学習を行おうとすると下記のようなエラーが出ます
10
-
11
- どなたか、ご助力お願いいたします
12
10
 
13
11
  ```
14
12
 
15
- ValueError: Error when checking target: expected dense_3_ to have 2 dimensions, but got array with shape (40, 200, 10)
13
+ ValueError: Graph disconnected: cannot obtain value for tensor Tensor("inputs_2_:0", shape=(None, 4, 10), dtype=float32) at layer "inputs_2_". The following previous layers were accessed without issue: []
16
14
 
17
15
  ```
18
16
 
@@ -22,11 +20,183 @@
22
20
 
23
21
 
24
22
 
23
+ ```constmodel
24
+
25
+ from keras.layers import Input, Dense, LSTM, Concatenate, Activation
26
+
27
+ from keras.models import Model
28
+
29
+
30
+
31
+
32
+
33
+ class Mymodel(Model):
34
+
35
+ def __init__(self, hidden_finger_size, output_finger_size, output_hand_size, hidden_LSTM_size):
36
+
37
+ self.hidden_fing = hidden_finger_size
38
+
39
+ self.output_fing = output_finger_size
40
+
41
+ self.output_hand = output_hand_size
42
+
43
+ self.hidden_lstm = hidden_LSTM_size
44
+
45
+
46
+
47
+ def static_model(self, inputs):
48
+
49
+ #input
50
+
51
+ inputs[0] = Input(shape=(12, ), name='inputs_0')
52
+
53
+ inputs[1] = Input(shape=(15, ), name='inputs_1')
54
+
55
+ inputs[2] = Input(shape=(15, ), name='inputs_2')
56
+
57
+ inputs[3] = Input(shape=(15, ), name='inputs_3')
58
+
59
+ inputs[4] = Input(shape=(15, ), name='inputs_4')
60
+
61
+
62
+
63
+ #dense_1
64
+
65
+ dense_1_0 = Dense(self.hidden_fing, name='dense_1_0')(inputs[0])
66
+
67
+ dense_1_1 = Dense(self.hidden_fing, name='dense_1_1')(inputs[1])
68
+
69
+ dense_1_2 = Dense(self.hidden_fing, name='dense_1_2')(inputs[2])
70
+
71
+ dense_1_3 = Dense(self.hidden_fing, name='dense_1_3')(inputs[3])
72
+
73
+ dense_1_4 = Dense(self.hidden_fing, name='dense_1_4')(inputs[4])
74
+
75
+
76
+
77
+ #activate_1
78
+
79
+ activation_1_0 = Activation(activation='sigmoid', name='activation_1_0')(dense_1_0)
80
+
81
+ activation_1_1 = Activation(activation='sigmoid', name='activation_1_1')(dense_1_1)
82
+
83
+ activation_1_2 = Activation(activation='sigmoid', name='activation_1_2')(dense_1_2)
84
+
85
+ activation_1_3 = Activation(activation='sigmoid', name='activation_1_3')(dense_1_3)
86
+
87
+ activation_1_4 = Activation(activation='sigmoid', name='activation_1_4')(dense_1_4)
88
+
89
+
90
+
91
+ #dense_2
92
+
93
+ dense_2_0 = Dense(self.output_fing, name='dense_2_0')(activation_1_0)
94
+
95
+ dense_2_1 = Dense(self.output_fing, name='dense_2_1')(activation_1_1)
96
+
97
+ dense_2_2 = Dense(self.output_fing, name='dense_2_2')(activation_1_2)
98
+
99
+ dense_2_3 = Dense(self.output_fing, name='dense_2_3')(activation_1_3)
100
+
101
+ dense_2_4 = Dense(self.output_fing, name='dense_2_4')(activation_1_4)
102
+
103
+
104
+
105
+ #activate_2
106
+
107
+ activation_2_0 = Activation(activation='sigmoid', name='activation_2_0')(dense_2_0)
108
+
109
+ activation_2_1 = Activation(activation='sigmoid', name='activation_2_1')(dense_2_1)
110
+
111
+ activation_2_2 = Activation(activation='sigmoid', name='activation_2_2')(dense_2_2)
112
+
113
+ activation_2_3 = Activation(activation='sigmoid', name='activation_2_3')(dense_2_3)
114
+
115
+ activation_2_4 = Activation(activation='sigmoid', name='activation_2_4')(dense_2_4)
116
+
117
+
118
+
119
+ #concatenate
120
+
121
+ concatenate = Concatenate()([activation_2_0, activation_2_1, activation_2_2, activation_2_3, activation_2_4])
122
+
123
+
124
+
125
+ #dense_3
126
+
127
+ dense_3_ = Dense(self.output_hand, name='dense_3_')(concatenate)
128
+
129
+ activation_3_ = Activation(activation='sigmoid', name='activation_3_')(dense_3_)
130
+
131
+
132
+
133
+ return activation_3_
134
+
135
+
136
+
137
+ def dynamic_model(self, inputs):
138
+
139
+ inputs_2_ = Input(shape=(4, 10, ), name='inputs_2_')
140
+
141
+
142
+
143
+ lstm_0 = LSTM(self.hidden_lstm, activation='sigmoid', name='lstm_0')(inputs_2_)
144
+
145
+ dense_4_ = Dense(self.output_hand, name='dense_4_')(lstm_0)
146
+
147
+ activation_4_ = Activation(activation='sigmoid', name='activation_4_')(dense_4_)
148
+
149
+
150
+
151
+ return activation_4_
152
+
153
+
154
+
155
+ ```
156
+
157
+
158
+
25
- ```python
159
+ ```train
160
+
161
+ import sys, time, csv, os
162
+
163
+
164
+
165
+ my_path = ".."
166
+
167
+ sys.path.append(my_path)
168
+
169
+
170
+
171
+ from my_dataset.load_hand import load_hand_data
172
+
173
+ from Myfunction import Myfunction
174
+
175
+ from const_model import Mymodel
176
+
177
+
178
+
179
+ from keras.layers import Input, Dense, LSTM, Flatten, Concatenate, Activation
180
+
181
+ from keras.models import Model
182
+
183
+ from keras.utils import plot_model
184
+
185
+
186
+
187
+ import matplotlib.pyplot as plt
188
+
189
+ import numpy as np
190
+
191
+ import tensorflow as tf
192
+
193
+
194
+
195
+
26
196
 
27
197
  I = [12, 15, 15, 15, 15]
28
198
 
29
- epochs = 100
199
+ epochs = 1
30
200
 
31
201
  batch_size = 40
32
202
 
@@ -38,7 +208,7 @@
38
208
 
39
209
  hidden_finger_size = 10 #指の中間層サイズ
40
210
 
41
- output_finger_size = 8 #指の出力層サイズ
211
+ output_finger_size = 5 #指の出力層サイズ
42
212
 
43
213
  output_hand_size = 10 #手の出力層サイズ
44
214
 
@@ -46,161 +216,75 @@
46
216
 
47
217
 
48
218
 
219
+ mymodel = Mymodel(hidden_finger_size, output_finger_size, output_hand_size, hidden_LSTM_size)
220
+
49
221
  func = Myfunction(batch_size)
50
222
 
51
223
 
52
224
 
225
+
226
+
53
227
  train_x, train_t, test_x, test_t = load_hand_data(["a.txt", "i.txt", "u.txt", "e.txt", "ku.txt", "se.txt", "so.txt", "ma.txt", "ru.txt", "ya.txt"],
54
228
 
55
229
  [0, 1, 2, 3, 4, 5, 6, 7, 8, 9])
56
230
 
57
-
58
-
59
- #reshape_train_data_to_batch_data
231
+ #reshape_train_data_to_batch
60
-
232
+
61
- batch_x, batch_t = func.get_batch(train_x, train_t)
233
+ tr_batch_x, tr_batch_t = func.get_batch(train_x, train_t)
62
-
234
+
63
- #batch_x = (40, 200, 72)
235
+ te_batch_x, te_batch_t = func.get_test_batch(test_x, test_t)
64
-
65
- #batch_t = (40, 200, 10)
66
236
 
67
237
 
68
238
 
69
239
  #train_data
70
240
 
71
- h_1_0 = batch_x[:, :, :I[0]] #親指
241
+ h_1_0 = tr_batch_x[:, :I[0]] #親指
72
-
242
+
73
- h_1_1 = batch_x[:, :, I[0]:I[0]+I[1]] #人差し指
243
+ h_1_1 = tr_batch_x[:, I[0]:I[0]+I[1]] #人差し指
74
-
244
+
75
- h_1_2 = batch_x[:, :, I[0]+I[1]:I[0]+I[1]+I[2]] #中指
245
+ h_1_2 = tr_batch_x[:, I[0]+I[1]:I[0]+I[1]+I[2]] #中指
76
-
246
+
77
- h_1_3 = batch_x[:, :, I[0]+I[1]+I[2]:I[0]+I[1]+I[2]+I[3]] #薬指
247
+ h_1_3 = tr_batch_x[:, I[0]+I[1]+I[2]:I[0]+I[1]+I[2]+I[3]] #薬指
78
-
248
+
79
- h_1_4 = batch_x[:, :, I[0]+I[1]+I[2]+I[3]:] #小指
249
+ h_1_4 = tr_batch_x[:, I[0]+I[1]+I[2]+I[3]:] #小指
80
-
81
-
82
250
 
83
251
  #test_data
84
252
 
85
- h_2_0 = test_x[:, :I[0]] #親指
253
+ h_2_0 = te_batch_x[:, :I[0]] #親指
86
-
254
+
87
- h_2_1 = test_x[:, I[0]:I[0]+I[1]] #人差し指
255
+ h_2_1 = te_batch_x[:, I[0]:I[0]+I[1]] #人差し指
88
-
256
+
89
- h_2_2 = test_x[:, I[0]+I[1]:I[0]+I[1]+I[2]] #中指
257
+ h_2_2 = te_batch_x[:, I[0]+I[1]:I[0]+I[1]+I[2]] #中指
90
-
258
+
91
- h_2_3 = test_x[:, I[0]+I[1]+I[2]:I[0]+I[1]+I[2]+I[3]] #薬指
259
+ h_2_3 = te_batch_x[:, I[0]+I[1]+I[2]:I[0]+I[1]+I[2]+I[3]] #薬指
92
-
260
+
93
- h_2_4 = test_x[:, I[0]+I[1]+I[2]+I[3]:] #小指
261
+ h_2_4 = te_batch_x[:, I[0]+I[1]+I[2]+I[3]:] #小指
94
-
95
- ################################################################
262
+
96
-
97
- #
263
+
98
-
99
- #
264
+
100
-
101
- #モデル:静的モデル
265
+
102
-
103
- #
104
-
105
- #
106
-
107
- ################################################################
108
266
 
109
267
  #input
110
268
 
111
- inputs_0 = Input(shape=(func.batch_frame, 12), name='inputs_0')
269
+ inputs_0 = Input(shape=(12, ), name='inputs_0')
112
-
270
+
113
- inputs_1 = Input(shape=(func.batch_frame, 15), name='inputs_1')
271
+ inputs_1 = Input(shape=(15, ), name='inputs_1')
114
-
272
+
115
- inputs_2 = Input(shape=(func.batch_frame, 15), name='inputs_2')
273
+ inputs_2 = Input(shape=(15, ), name='inputs_2')
116
-
274
+
117
- inputs_3 = Input(shape=(func.batch_frame, 15), name='inputs_3')
275
+ inputs_3 = Input(shape=(15, ), name='inputs_3')
118
-
276
+
119
- inputs_4 = Input(shape=(func.batch_frame, 15), name='inputs_4')
277
+ inputs_4 = Input(shape=(15, ), name='inputs_4')
120
-
121
-
122
-
123
- #dense_1
278
+
124
-
125
- dense_1_0 = Dense(hidden_finger_size, name='dense_1_0')(inputs_0)
126
-
127
- dense_1_1 = Dense(hidden_finger_size, name='dense_1_1')(inputs_1)
128
-
129
- dense_1_2 = Dense(hidden_finger_size, name='dense_1_2')(inputs_2)
130
-
131
- dense_1_3 = Dense(hidden_finger_size, name='dense_1_3')(inputs_3)
132
-
133
- dense_1_4 = Dense(hidden_finger_size, name='dense_1_4')(inputs_4)
134
-
135
-
136
-
137
- #activate_1
138
-
139
- activation_1_0 = Activation(activation='sigmoid', name='activation_1_0')(dense_1_0)
140
-
141
- activation_1_1 = Activation(activation='sigmoid', name='activation_1_1')(dense_1_1)
142
-
143
- activation_1_2 = Activation(activation='sigmoid', name='activation_1_2')(dense_1_2)
144
-
145
- activation_1_3 = Activation(activation='sigmoid', name='activation_1_3')(dense_1_3)
146
-
147
- activation_1_4 = Activation(activation='sigmoid', name='activation_1_4')(dense_1_4)
148
-
149
-
150
-
151
- #dense_2
152
-
153
- dense_2_0 = Dense(output_finger_size, name='dense_2_0')(activation_1_0)
154
-
155
- dense_2_1 = Dense(output_finger_size, name='dense_2_1')(activation_1_1)
156
-
157
- dense_2_2 = Dense(output_finger_size, name='dense_2_2')(activation_1_2)
158
-
159
- dense_2_3 = Dense(output_finger_size, name='dense_2_3')(activation_1_3)
160
-
161
- dense_2_4 = Dense(output_finger_size, name='dense_2_4')(activation_1_4)
162
-
163
-
164
-
165
- #activate_2
166
-
167
- activation_2_0 = Activation(activation='sigmoid', name='activation_2_0')(dense_2_0)
168
-
169
- activation_2_1 = Activation(activation='sigmoid', name='activation_2_1')(dense_2_1)
170
-
171
- activation_2_2 = Activation(activation='sigmoid', name='activation_2_2')(dense_2_2)
172
-
173
- activation_2_3 = Activation(activation='sigmoid', name='activation_2_3')(dense_2_3)
174
-
175
- activation_2_4 = Activation(activation='sigmoid', name='activation_2_4')(dense_2_4)
176
-
177
-
178
-
179
- #flatten
180
-
181
- flatten_0 = Flatten()(activation_2_0)
182
-
183
- flatten_1 = Flatten()(activation_2_1)
184
-
185
- flatten_2 = Flatten()(activation_2_2)
186
-
187
- flatten_3 = Flatten()(activation_2_3)
188
-
189
- flatten_4 = Flatten()(activation_2_4)
190
-
191
-
192
-
193
-
194
-
195
- #concatenate
196
-
197
- concatenate = Concatenate()([flatten_0, flatten_1, flatten_2, flatten_3, flatten_4])
279
+ inputs = [inputs_0, inputs_1, inputs_2, inputs_3, inputs_4]
198
-
199
-
200
-
280
+
281
+
282
+
283
+
284
+
201
- #dense_3
285
+ static_outputs = mymodel.static_model(inputs)
202
-
286
+
203
- dense_3_ = Dense(output_hand_size, activation='sigmoid', name='dense_3_')(concatenate)
287
+ dynamic_outputs = mymodel.dynamic_model(static_outputs)
204
288
 
205
289
 
206
290
 
@@ -208,11 +292,7 @@
208
292
 
209
293
  #train
210
294
 
211
- model = Model(inputs=[inputs_0, inputs_1, inputs_2, inputs_3, inputs_4], outputs=dense_3_)
295
+ model = Model(inputs=inputs, outputs=dynamic_outputs)
212
-
213
- model.summary()
214
-
215
-
216
296
 
217
297
  model.compile(optimizer='adam',
218
298
 
@@ -220,7 +300,7 @@
220
300
 
221
301
  metrics=['accuracy'])
222
302
 
223
- history = model.fit([h_1_0, h_1_1, h_1_2, h_1_3, h_1_4], batch_t,
303
+ history = model.fit([h_1_0, h_1_1, h_1_2, h_1_3, h_1_4], tr_batch_t,
224
304
 
225
305
  batch_size=batch_size,
226
306
 
@@ -228,17 +308,21 @@
228
308
 
229
309
  verbose=1,
230
310
 
231
- validation_data=([h_2_0, h_2_1, h_2_2, h_2_3, h_2_4], test_t))
311
+ validation_data=([h_2_0, h_2_1, h_2_2, h_2_3, h_2_4], te_batch_t))
312
+
313
+
314
+
315
+
232
316
 
233
317
  ```
234
318
 
235
-
236
-
237
319
  ### 試したこと
238
320
 
321
+ 前回の投稿から、モデルをクラス化しました
322
+
239
- 同様なエラーメッセージについていろいろ調べてみましたが、解決まで至りませんでした
323
+ クラス化した一つめのモデルの方は学習まで持って行けたのですが、二つ目のモデルの方を繋げて学習を行おうとするとエラーが
324
+
240
-
325
+ 出ました
241
-
242
326
 
243
327
  ### 補足情報（FW/ツールのバージョンなど）
244
328
 

1

エラー部分について追記

2020/10/21 06:28

投稿

vurie
vurie

スコア0

test CHANGED
File without changes
test CHANGED
@@ -1,12 +1,14 @@
1
1
  ### 前提・実現したいこと
2
2
 
3
- 分類を行う人工知能を作成したく、kerasを使って構築をしています
3
+ 分類を行う人工知能を作成したく、kerasを使って構築をしていますが、現在以下のようなエラーが発生しています
4
4
 
5
5
 
6
6
 
7
7
  ### 発生している問題・エラーメッセージ
8
8
 
9
-
9
+ batch_tの方をdense_3_に投げてエラーが図れているように思うのですが、原因がわかりません
10
+
11
+ どなたか、ご助力お願いいたします
10
12
 
11
13
  ```
12
14