質問編集履歴
1
作成途中
test
CHANGED
|
File without changes
|
test
CHANGED
|
@@ -24,6 +24,10 @@
|
|
|
24
24
|
|
|
25
25
|
from keras.callbacks import EarlyStopping, ModelCheckpoint, CSVLogger
|
|
26
26
|
|
|
27
|
+
from sklearn.model_selection import train_test_split
|
|
28
|
+
|
|
29
|
+
from sklearn import datasets
|
|
30
|
+
|
|
27
31
|
import numpy as np
|
|
28
32
|
|
|
29
33
|
from PIL import Image
|
|
@@ -48,9 +52,9 @@
|
|
|
48
52
|
|
|
49
53
|
#それぞれの画像の枚数を入力
|
|
50
54
|
|
|
51
|
-
A =
|
|
55
|
+
A = 50
|
|
52
|
-
|
|
56
|
+
|
|
53
|
-
B =
|
|
57
|
+
B = 50
|
|
54
58
|
|
|
55
59
|
sum =A+B
|
|
56
60
|
|
|
@@ -72,7 +76,7 @@
|
|
|
72
76
|
|
|
73
77
|
#エポック数
|
|
74
78
|
|
|
75
|
-
E = 5
|
|
79
|
+
E = 5
|
|
76
80
|
|
|
77
81
|
#バッチサイズ
|
|
78
82
|
|
|
@@ -142,10 +146,6 @@
|
|
|
142
146
|
|
|
143
147
|
|
|
144
148
|
|
|
145
|
-
|
|
146
|
-
|
|
147
|
-
|
|
148
|
-
|
|
149
149
|
print("画像の読み込み 終了")
|
|
150
150
|
|
|
151
151
|
|
|
@@ -156,8 +156,6 @@
|
|
|
156
156
|
|
|
157
157
|
|
|
158
158
|
|
|
159
|
-
|
|
160
|
-
|
|
161
159
|
#最大応力の位置＿読み込み＿表示
|
|
162
160
|
|
|
163
161
|
location = np.loadtxt("data/value/max_stress_value_a.csv",delimiter=",",skiprows=0)
|
|
@@ -188,7 +186,7 @@
|
|
|
188
186
|
|
|
189
187
|
|
|
190
188
|
|
|
191
|
-
model.add(Dense(
|
|
189
|
+
model.add(Dense(5000, input_dim=Z,kernel_initializer='random_uniform',bias_initializer='zeros'))
|
|
192
190
|
|
|
193
191
|
#model.add(Activation("LeakyReLU"))
|
|
194
192
|
|
|
@@ -198,6 +196,8 @@
|
|
|
198
196
|
|
|
199
197
|
|
|
200
198
|
|
|
199
|
+
|
|
200
|
+
|
|
201
201
|
model.add(Dense(100,kernel_initializer='random_uniform',bias_initializer='zeros'))
|
|
202
202
|
|
|
203
203
|
model.add(LeakyReLU())
|
|
@@ -206,15 +206,17 @@
|
|
|
206
206
|
|
|
207
207
|
|
|
208
208
|
|
|
209
|
+
|
|
210
|
+
|
|
209
|
-
model.add(Dense(
|
|
211
|
+
model.add(Dense(10,kernel_initializer='random_uniform',bias_initializer='zeros'))
|
|
210
212
|
|
|
211
213
|
model.add(LeakyReLU())
|
|
212
214
|
|
|
213
|
-
model.add(Dropout(0.0
|
|
215
|
+
model.add(Dropout(0.0))
|
|
214
|
-
|
|
215
|
-
|
|
216
|
-
|
|
216
|
+
|
|
217
|
+
|
|
218
|
+
|
|
217
|
-
model.add(Dense(
|
|
219
|
+
model.add(Dense(5,kernel_initializer='random_uniform',bias_initializer='zeros'))
|
|
218
220
|
|
|
219
221
|
model.add(LeakyReLU())
|
|
220
222
|
|
|
@@ -290,6 +292,102 @@
|
|
|
290
292
|
|
|
291
293
|
|
|
292
294
|
|
|
295
|
+
image_list = np.array(image_list)
|
|
296
|
+
|
|
297
|
+
location_list = np.array(location_list)
|
|
298
|
+
|
|
299
|
+
print(image_list.shape, image_list.dtype) # (300, 11250) float64
|
|
300
|
+
|
|
301
|
+
print(location_list.shape, location_list.dtype) # (300,) float64
|
|
302
|
+
|
|
303
|
+
|
|
304
|
+
|
|
305
|
+
|
|
306
|
+
|
|
307
|
+
def get_batch(image_list, location_list, batch_size, shuffle=False):
|
|
308
|
+
|
|
309
|
+
'''ミニバッチを生成するジェネレーター関数
|
|
310
|
+
|
|
311
|
+
'''
|
|
312
|
+
|
|
313
|
+
num_samples = location_list # サンプル数
|
|
314
|
+
|
|
315
|
+
if shuffle:# シャッフルする場合
|
|
316
|
+
|
|
317
|
+
indices = np.random.permutation(num_samples)
|
|
318
|
+
|
|
319
|
+
else: # シャッフルしない場合
|
|
320
|
+
|
|
321
|
+
indices = np.random.arange(num_samples)
|
|
322
|
+
|
|
323
|
+
num_steps = np.ceil(num_samples / batch_size).astype(int)
|
|
324
|
+
|
|
325
|
+
print(num_steps)
|
|
326
|
+
|
|
327
|
+
print(type(num_steps))
|
|
328
|
+
|
|
329
|
+
|
|
330
|
+
|
|
331
|
+
for itr in range(num_steps):
|
|
332
|
+
|
|
333
|
+
start = batch_size * itr
|
|
334
|
+
|
|
335
|
+
excerpt = indices[start:start + batch_size]
|
|
336
|
+
|
|
337
|
+
yield x[excerpt], y[excerpt]
|
|
338
|
+
|
|
339
|
+
|
|
340
|
+
|
|
341
|
+
# 保存用ディレクトリ
|
|
342
|
+
|
|
343
|
+
out_dirpath = 'prediction'
|
|
344
|
+
|
|
345
|
+
os.makedirs(out_dirpath, exist_ok=True)
|
|
346
|
+
|
|
347
|
+
|
|
348
|
+
|
|
349
|
+
x_train, x_test, y_train, y_test = train_test_split(image_list, location_list, test_size=0.2)
|
|
350
|
+
|
|
351
|
+
|
|
352
|
+
|
|
353
|
+
# 学習する。
|
|
354
|
+
|
|
355
|
+
epochs = E
|
|
356
|
+
|
|
357
|
+
for i in range(epochs):
|
|
358
|
+
|
|
359
|
+
for x_batch, y_batch in get_batch(x_train, y_train, batch_size=BATCH_SIZE, shuffle=True):
|
|
360
|
+
|
|
361
|
+
# x_batch, y_batch が生成されたミニバッチ
|
|
362
|
+
|
|
363
|
+
|
|
364
|
+
|
|
365
|
+
# 1バッチ分学習する
|
|
366
|
+
|
|
367
|
+
model.train_on_batch(x_batch, y_batch)
|
|
368
|
+
|
|
369
|
+
|
|
370
|
+
|
|
371
|
+
# エポックごとにテストデータで推論する。
|
|
372
|
+
|
|
373
|
+
y_pred = model.predict_classes(x_train)
|
|
374
|
+
|
|
375
|
+
result = np.c_[y_pred, y_train]
|
|
376
|
+
|
|
377
|
+
|
|
378
|
+
|
|
379
|
+
# 推論結果を保存する。
|
|
380
|
+
|
|
381
|
+
filepath = os.path.join(out_dirpath, 'prediction_{}.csv'.format(i))
|
|
382
|
+
|
|
383
|
+
np.savetxt(filepath, result, fmt='%.0f')
|
|
384
|
+
|
|
385
|
+
|
|
386
|
+
|
|
387
|
+
|
|
388
|
+
|
|
389
|
+
|
|
390
|
+
|
|
293
391
|
end_time = time.time()
|
|
294
392
|
|
|
295
393
|
print("\n終了時刻: ",end_time)
|
|
@@ -308,4 +406,8 @@
|
|
|
308
406
|
|
|
309
407
|
|
|
310
408
|
|
|
409
|
+
|
|
410
|
+
|
|
411
|
+
|
|
412
|
+
|
|
311
413
|
```
|