質問編集履歴
2
追記
test
CHANGED
|
File without changes
|
test
CHANGED
|
@@ -313,3 +313,25 @@
|
|
|
313
313
|
|
|
314
314
|
|
|
315
315
|
その他、batchサイズやepoch数を大きくしたり小さくしたりしてみたりしたのですが、偏りはあまり解消されませんでした。
|
|
316
|
+
|
|
317
|
+
|
|
318
|
+
|
|
319
|
+
###追記
|
|
320
|
+
|
|
321
|
+
テストデータに分けてから、訓練データのみ水増し処理するように修正しました。
|
|
322
|
+
|
|
323
|
+
結果は以下のようになりました。
|
|
324
|
+
|
|
325
|
+
ただ、特定のメンバーが確率高く出る偏りは解消されないようでした。
|
|
326
|
+
|
|
327
|
+
|
|
328
|
+
|
|
329
|
+
```
|
|
330
|
+
|
|
331
|
+
validation loss:0.9875603318214417
|
|
332
|
+
|
|
333
|
+
validation accuracy:0.7211538553237915
|
|
334
|
+
|
|
335
|
+
```
|
|
336
|
+
|
|
337
|
+
|
1
summary追加
test
CHANGED
|
File without changes
|
test
CHANGED
|
@@ -48,6 +48,16 @@
|
|
|
48
48
|
|
|
49
49
|
```python
|
|
50
50
|
|
|
51
|
+
members = ['A','B','C','D','E','F','G','H','I']
|
|
52
|
+
|
|
53
|
+
|
|
54
|
+
|
|
55
|
+
TRAIN_FOLDER_PATH = 'D:\train'
|
|
56
|
+
|
|
57
|
+
TEST_FOLDER_PATH = 'D:\test'
|
|
58
|
+
|
|
59
|
+
|
|
60
|
+
|
|
51
61
|
# 教師データのラベル付け
|
|
52
62
|
|
|
53
63
|
X_train = []
|
|
@@ -180,6 +190,100 @@
|
|
|
180
190
|
|
|
181
191
|
|
|
182
192
|
|
|
193
|
+
###model.summary()
|
|
194
|
+
|
|
195
|
+
```Model: "sequential"
|
|
196
|
+
|
|
197
|
+
_________________________________________________________________
|
|
198
|
+
|
|
199
|
+
Layer (type) Output Shape Param #
|
|
200
|
+
|
|
201
|
+
=================================================================
|
|
202
|
+
|
|
203
|
+
conv2d (Conv2D) (None, 64, 64, 32) 896
|
|
204
|
+
|
|
205
|
+
|
|
206
|
+
|
|
207
|
+
max_pooling2d (MaxPooling2D (None, 32, 32, 32) 0
|
|
208
|
+
|
|
209
|
+
)
|
|
210
|
+
|
|
211
|
+
|
|
212
|
+
|
|
213
|
+
dropout (Dropout) (None, 32, 32, 32) 0
|
|
214
|
+
|
|
215
|
+
|
|
216
|
+
|
|
217
|
+
conv2d_1 (Conv2D) (None, 30, 30, 64) 18496
|
|
218
|
+
|
|
219
|
+
|
|
220
|
+
|
|
221
|
+
max_pooling2d_1 (MaxPooling (None, 15, 15, 64) 0
|
|
222
|
+
|
|
223
|
+
2D)
|
|
224
|
+
|
|
225
|
+
|
|
226
|
+
|
|
227
|
+
dropout_1 (Dropout) (None, 15, 15, 64) 0
|
|
228
|
+
|
|
229
|
+
|
|
230
|
+
|
|
231
|
+
conv2d_2 (Conv2D) (None, 13, 13, 128) 73856
|
|
232
|
+
|
|
233
|
+
|
|
234
|
+
|
|
235
|
+
max_pooling2d_2 (MaxPooling (None, 6, 6, 128) 0
|
|
236
|
+
|
|
237
|
+
2D)
|
|
238
|
+
|
|
239
|
+
|
|
240
|
+
|
|
241
|
+
dropout_2 (Dropout) (None, 6, 6, 128) 0
|
|
242
|
+
|
|
243
|
+
|
|
244
|
+
|
|
245
|
+
conv2d_3 (Conv2D) (None, 4, 4, 128) 147584
|
|
246
|
+
|
|
247
|
+
|
|
248
|
+
|
|
249
|
+
max_pooling2d_3 (MaxPooling (None, 2, 2, 128) 0
|
|
250
|
+
|
|
251
|
+
2D)
|
|
252
|
+
|
|
253
|
+
|
|
254
|
+
|
|
255
|
+
dropout_3 (Dropout) (None, 2, 2, 128) 0
|
|
256
|
+
|
|
257
|
+
|
|
258
|
+
|
|
259
|
+
flatten (Flatten) (None, 512) 0
|
|
260
|
+
|
|
261
|
+
|
|
262
|
+
|
|
263
|
+
dense (Dense) (None, 512) 262656
|
|
264
|
+
|
|
265
|
+
|
|
266
|
+
|
|
267
|
+
dropout_4 (Dropout) (None, 512) 0
|
|
268
|
+
|
|
269
|
+
|
|
270
|
+
|
|
271
|
+
dense_1 (Dense) (None, 9) 4617
|
|
272
|
+
|
|
273
|
+
|
|
274
|
+
|
|
275
|
+
=================================================================
|
|
276
|
+
|
|
277
|
+
Total params: 508,105
|
|
278
|
+
|
|
279
|
+
Trainable params: 508,105
|
|
280
|
+
|
|
281
|
+
Non-trainable params: 0
|
|
282
|
+
|
|
283
|
+
```
|
|
284
|
+
|
|
285
|
+
|
|
286
|
+
|
|
183
287
|
###結果
|
|
184
288
|
|
|
185
289
|
```
|