質問編集履歴
1
keras_cnn.h5の保存コードを追加いたしました。よろしくお願い致します。
test
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
kerasで学習させたh5モデルからmlmodelへの変換
|
|
1
|
+
kerasで学習させたh5モデルからmlmodelへの変換しようとするとエラーがでる
|
test
CHANGED
|
@@ -88,6 +88,152 @@
|
|
|
88
88
|
|
|
89
89
|
|
|
90
90
|
|
|
91
|
+
|
|
92
|
+
|
|
93
|
+
```
|
|
94
|
+
|
|
95
|
+
from keras.models import Sequential
|
|
96
|
+
|
|
97
|
+
|
|
98
|
+
|
|
99
|
+
from keras.layers import Conv2D, MaxPooling2D
|
|
100
|
+
|
|
101
|
+
from keras.layers import Activation, Dropout, Flatten, Dense
|
|
102
|
+
|
|
103
|
+
from keras.utils import np_utils
|
|
104
|
+
|
|
105
|
+
import keras
|
|
106
|
+
|
|
107
|
+
import numpy as np
|
|
108
|
+
|
|
109
|
+
|
|
110
|
+
|
|
111
|
+
|
|
112
|
+
|
|
113
|
+
classes = ["perfect", "normal", "bad"]
|
|
114
|
+
|
|
115
|
+
num_classes = len(classes)
|
|
116
|
+
|
|
117
|
+
image_size = 75
|
|
118
|
+
|
|
119
|
+
|
|
120
|
+
|
|
121
|
+
|
|
122
|
+
|
|
123
|
+
def main():
|
|
124
|
+
|
|
125
|
+
X_train, X_test, y_train, y_test = np.load("./Judgment.npy")
|
|
126
|
+
|
|
127
|
+
X_train = X_train.astype("float") / 256
|
|
128
|
+
|
|
129
|
+
X_test = X_test.astype("float") / 256
|
|
130
|
+
|
|
131
|
+
y_train = np_utils.to_categorical(y_train, num_classes)
|
|
132
|
+
|
|
133
|
+
y_test = np_utils.to_categorical(y_test, num_classes)
|
|
134
|
+
|
|
135
|
+
|
|
136
|
+
|
|
137
|
+
model = model_train(X_train, y_train)
|
|
138
|
+
|
|
139
|
+
model_eval(model, X_test, y_test)
|
|
140
|
+
|
|
141
|
+
|
|
142
|
+
|
|
143
|
+
|
|
144
|
+
|
|
145
|
+
def model_train(X, y):
|
|
146
|
+
|
|
147
|
+
model = Sequential()
|
|
148
|
+
|
|
149
|
+
model.add(Conv2D(32, (3, 3), padding='same', input_shape=X.shape[1:]))
|
|
150
|
+
|
|
151
|
+
model.add(Activation('relu'))
|
|
152
|
+
|
|
153
|
+
model.add(Conv2D(32, (3, 3)))
|
|
154
|
+
|
|
155
|
+
model.add(Activation('relu'))
|
|
156
|
+
|
|
157
|
+
model.add(MaxPooling2D(pool_size=(2, 2)))
|
|
158
|
+
|
|
159
|
+
model.add(Dropout(0.25))
|
|
160
|
+
|
|
161
|
+
|
|
162
|
+
|
|
163
|
+
model.add(Conv2D(64, (3, 3), padding='same'))
|
|
164
|
+
|
|
165
|
+
model.add(Activation('relu'))
|
|
166
|
+
|
|
167
|
+
model.add(Conv2D(64, (3, 3)))
|
|
168
|
+
|
|
169
|
+
model.add(Activation('relu'))
|
|
170
|
+
|
|
171
|
+
model.add(MaxPooling2D(pool_size=(2, 2)))
|
|
172
|
+
|
|
173
|
+
model.add(Dropout(0.25))
|
|
174
|
+
|
|
175
|
+
|
|
176
|
+
|
|
177
|
+
model.add(Flatten())
|
|
178
|
+
|
|
179
|
+
model.add(Dense(512))
|
|
180
|
+
|
|
181
|
+
model.add(Activation('relu'))
|
|
182
|
+
|
|
183
|
+
model.add(Dropout(0.5))
|
|
184
|
+
|
|
185
|
+
model.add(Dense(3))
|
|
186
|
+
|
|
187
|
+
model.add(Activation('softmax'))
|
|
188
|
+
|
|
189
|
+
|
|
190
|
+
|
|
191
|
+
opt = keras.optimizers.rmsprop(lr=0.0001, decay=1e-6)
|
|
192
|
+
|
|
193
|
+
model.compile(
|
|
194
|
+
|
|
195
|
+
loss='categorical_crossentropy', optimizer=opt, metrics=['accuracy'])
|
|
196
|
+
|
|
197
|
+
model.fit(X, y, batch_size=32, epochs=100)
|
|
198
|
+
|
|
199
|
+
|
|
200
|
+
|
|
201
|
+
|
|
202
|
+
|
|
203
|
+
model.save('./keras_cnn.h5')
|
|
204
|
+
|
|
205
|
+
|
|
206
|
+
|
|
207
|
+
return model
|
|
208
|
+
|
|
209
|
+
|
|
210
|
+
|
|
211
|
+
|
|
212
|
+
|
|
213
|
+
def model_eval(model, X, y):
|
|
214
|
+
|
|
215
|
+
scores = model.evaluate(X, y, verbose=1)
|
|
216
|
+
|
|
217
|
+
print('Test Loss: ', scores[0])
|
|
218
|
+
|
|
219
|
+
print('Test Accuracy: ', scores[1])
|
|
220
|
+
|
|
221
|
+
|
|
222
|
+
|
|
223
|
+
|
|
224
|
+
|
|
225
|
+
if __name__ == "__main__":
|
|
226
|
+
|
|
227
|
+
main()
|
|
228
|
+
|
|
229
|
+
```
|
|
230
|
+
|
|
231
|
+
|
|
232
|
+
|
|
233
|
+
|
|
234
|
+
|
|
235
|
+
|
|
236
|
+
|
|
91
237
|
### 試したこと
|
|
92
238
|
|
|
93
239
|
|