質問編集履歴
3
解決したので解決コード書いた。
test
CHANGED
|
File without changes
|
test
CHANGED
|
@@ -299,3 +299,249 @@
|
|
|
299
299
|
|
|
300
300
|
|
|
301
301
|
デスクトップ上ではできます。
|
|
302
|
+
|
|
303
|
+
|
|
304
|
+
|
|
305
|
+
|
|
306
|
+
|
|
307
|
+
できました、ありがとうございます、以下にコード書きます、皆さん使ってみて下さい。
|
|
308
|
+
|
|
309
|
+
```python
|
|
310
|
+
|
|
311
|
+
import sys
|
|
312
|
+
|
|
313
|
+
|
|
314
|
+
|
|
315
|
+
from google.colab import drive
|
|
316
|
+
|
|
317
|
+
drive.mount('/content/drive')
|
|
318
|
+
|
|
319
|
+
|
|
320
|
+
|
|
321
|
+
sys.path.append('/content/drive/My Drive')
|
|
322
|
+
|
|
323
|
+
|
|
324
|
+
|
|
325
|
+
import numpy as np
|
|
326
|
+
|
|
327
|
+
import ActivationFunction as AF
|
|
328
|
+
|
|
329
|
+
|
|
330
|
+
|
|
331
|
+
# 3層ニューラルネットワーク
|
|
332
|
+
|
|
333
|
+
class ThreeLayerNetwork:
|
|
334
|
+
|
|
335
|
+
# コンストラクタ
|
|
336
|
+
|
|
337
|
+
def __init__(self, inodes, hnodes, onodes, lr):
|
|
338
|
+
|
|
339
|
+
# 各レイヤーのノード数
|
|
340
|
+
|
|
341
|
+
self.inodes = inodes
|
|
342
|
+
|
|
343
|
+
self.hnodes = hnodes
|
|
344
|
+
|
|
345
|
+
self.onodes = onodes
|
|
346
|
+
|
|
347
|
+
|
|
348
|
+
|
|
349
|
+
# 学習率
|
|
350
|
+
|
|
351
|
+
self.lr = lr
|
|
352
|
+
|
|
353
|
+
|
|
354
|
+
|
|
355
|
+
# 重みの初期化
|
|
356
|
+
|
|
357
|
+
self.w_ih = np.random.normal(0.0, 1.0, (self.hnodes, self.inodes))
|
|
358
|
+
|
|
359
|
+
self.w_ho = np.random.normal(0.0, 1.0, (self.onodes, self.hnodes))
|
|
360
|
+
|
|
361
|
+
|
|
362
|
+
|
|
363
|
+
# 活性化関数
|
|
364
|
+
|
|
365
|
+
self.af = AF.sigmoid
|
|
366
|
+
|
|
367
|
+
self.daf = AF.derivative_sigmoid
|
|
368
|
+
|
|
369
|
+
|
|
370
|
+
|
|
371
|
+
# 誤差逆伝搬
|
|
372
|
+
|
|
373
|
+
def backprop(self, idata, tdata):
|
|
374
|
+
|
|
375
|
+
# 縦ベクトルに変換
|
|
376
|
+
|
|
377
|
+
o_i = np.array(idata, ndmin=2).T
|
|
378
|
+
|
|
379
|
+
t = np.array(tdata, ndmin=2).T
|
|
380
|
+
|
|
381
|
+
|
|
382
|
+
|
|
383
|
+
# 隠れ層
|
|
384
|
+
|
|
385
|
+
x_h = np.dot(self.w_ih, o_i)
|
|
386
|
+
|
|
387
|
+
o_h = self.af(x_h)
|
|
388
|
+
|
|
389
|
+
|
|
390
|
+
|
|
391
|
+
# 出力層
|
|
392
|
+
|
|
393
|
+
x_o = np.dot(self.w_ho, o_h)
|
|
394
|
+
|
|
395
|
+
o_o = self.af(x_o)
|
|
396
|
+
|
|
397
|
+
|
|
398
|
+
|
|
399
|
+
# 誤差計算
|
|
400
|
+
|
|
401
|
+
e_o = (t - o_o)
|
|
402
|
+
|
|
403
|
+
e_h = np.dot(self.w_ho.T, e_o)
|
|
404
|
+
|
|
405
|
+
|
|
406
|
+
|
|
407
|
+
# 重みの更新
|
|
408
|
+
|
|
409
|
+
self.w_ho += self.lr * np.dot((e_o * self.daf(o_o)), o_h.T)
|
|
410
|
+
|
|
411
|
+
self.w_ih += self.lr * np.dot((e_h * self.daf(o_h)), o_i.T)
|
|
412
|
+
|
|
413
|
+
|
|
414
|
+
|
|
415
|
+
|
|
416
|
+
|
|
417
|
+
# 順伝搬
|
|
418
|
+
|
|
419
|
+
def feedforward(self, idata):
|
|
420
|
+
|
|
421
|
+
# 入力のリストを縦ベクトルに変換
|
|
422
|
+
|
|
423
|
+
o_i = np.array(idata, ndmin=2).T
|
|
424
|
+
|
|
425
|
+
|
|
426
|
+
|
|
427
|
+
# 隠れ層
|
|
428
|
+
|
|
429
|
+
x_h = np.dot(self.w_ih, o_i)
|
|
430
|
+
|
|
431
|
+
o_h = self.af(x_h)
|
|
432
|
+
|
|
433
|
+
|
|
434
|
+
|
|
435
|
+
# 出力層
|
|
436
|
+
|
|
437
|
+
x_o = np.dot(self.w_ho, o_h)
|
|
438
|
+
|
|
439
|
+
o_o = self.af(x_o)
|
|
440
|
+
|
|
441
|
+
|
|
442
|
+
|
|
443
|
+
return o_o
|
|
444
|
+
|
|
445
|
+
|
|
446
|
+
|
|
447
|
+
if __name__=='__main__':
|
|
448
|
+
|
|
449
|
+
# パラメータ
|
|
450
|
+
|
|
451
|
+
inodes = 784
|
|
452
|
+
|
|
453
|
+
hnodes = 100
|
|
454
|
+
|
|
455
|
+
onodes = 10
|
|
456
|
+
|
|
457
|
+
lr = 0.3
|
|
458
|
+
|
|
459
|
+
|
|
460
|
+
|
|
461
|
+
# ニューラルネットワークの初期化
|
|
462
|
+
|
|
463
|
+
nn = ThreeLayerNetwork(inodes, hnodes, onodes, lr)
|
|
464
|
+
|
|
465
|
+
|
|
466
|
+
|
|
467
|
+
# トレーニングデータのロード
|
|
468
|
+
|
|
469
|
+
training_data_file = open('drive/My Drive/mnist_dataset/mnist_train.csv', 'r')
|
|
470
|
+
|
|
471
|
+
training_data_list = training_data_file.readlines()
|
|
472
|
+
|
|
473
|
+
training_data_file.close()
|
|
474
|
+
|
|
475
|
+
|
|
476
|
+
|
|
477
|
+
# テストデータのロード
|
|
478
|
+
|
|
479
|
+
test_data_file = open('drive/My Drive/mnist_dataset/mnist_test.csv')
|
|
480
|
+
|
|
481
|
+
test_data_list = test_data_file.readlines()
|
|
482
|
+
|
|
483
|
+
test_data_file.close()
|
|
484
|
+
|
|
485
|
+
|
|
486
|
+
|
|
487
|
+
# 学習
|
|
488
|
+
|
|
489
|
+
epoch = 10
|
|
490
|
+
|
|
491
|
+
for e in range(epoch):
|
|
492
|
+
|
|
493
|
+
print('#epoch ', e)
|
|
494
|
+
|
|
495
|
+
data_size = len(training_data_list)
|
|
496
|
+
|
|
497
|
+
for i in range(data_size):
|
|
498
|
+
|
|
499
|
+
if i % 1000 == 0:
|
|
500
|
+
|
|
501
|
+
print(' train: {0:>5d} / {1:>5d}'.format(i, data_size))
|
|
502
|
+
|
|
503
|
+
val = training_data_list[i].split(',')
|
|
504
|
+
|
|
505
|
+
idata = (np.asfarray(val[1:]) / 255.0 * 0.99) + 0.01
|
|
506
|
+
|
|
507
|
+
tdata = np.zeros(onodes) + 0.01
|
|
508
|
+
|
|
509
|
+
tdata[int(val[0])] = 0.99
|
|
510
|
+
|
|
511
|
+
nn.backprop(idata, tdata)
|
|
512
|
+
|
|
513
|
+
pass
|
|
514
|
+
|
|
515
|
+
pass
|
|
516
|
+
|
|
517
|
+
|
|
518
|
+
|
|
519
|
+
# テスト
|
|
520
|
+
|
|
521
|
+
scoreboard = []
|
|
522
|
+
|
|
523
|
+
for record in test_data_list:
|
|
524
|
+
|
|
525
|
+
val = record.split(',')
|
|
526
|
+
|
|
527
|
+
idata = (np.asfarray(val[1:]) / 255.0 * 0.99) + 0.01
|
|
528
|
+
|
|
529
|
+
tlabel = int(val[0])
|
|
530
|
+
|
|
531
|
+
predict = nn.feedforward(idata)
|
|
532
|
+
|
|
533
|
+
plabel = np.argmax(predict)
|
|
534
|
+
|
|
535
|
+
scoreboard.append(tlabel == plabel)
|
|
536
|
+
|
|
537
|
+
pass
|
|
538
|
+
|
|
539
|
+
|
|
540
|
+
|
|
541
|
+
scoreboard_array = np.asarray(scoreboard)
|
|
542
|
+
|
|
543
|
+
print('performance: ', scoreboard_array.sum() / scoreboard_array.size)
|
|
544
|
+
|
|
545
|
+
コード
|
|
546
|
+
|
|
547
|
+
```
|
2
修正します。
test
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
自分でニューラルネットワークを作ろう
|
|
1
|
+
自分でニューラルネットワークを作ろうのプログラムをGoogleCollaboratoryで動かしたい。
|
test
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
https://qiita.com/takahiro_itazuri/items/d2bea1c643d7cca11352
|
|
2
2
|
|
|
3
|
-
ここを参考にしたのですが、以下そのまま転載。
|
|
3
|
+
プログラムはここを参考にしたのですが、以下そのまま転載。
|
|
4
4
|
|
|
5
5
|
|
|
6
6
|
|
|
@@ -287,3 +287,15 @@
|
|
|
287
287
|
|
|
288
288
|
|
|
289
289
|
と出ます、どうすれば良いでしょうか？
|
|
290
|
+
|
|
291
|
+
|
|
292
|
+
|
|
293
|
+
また、パスは
|
|
294
|
+
|
|
295
|
+
https://blog.mktia.com/file-access-from-colab-to-google-drive/
|
|
296
|
+
|
|
297
|
+
ここを参考に、そのまま書きました、そのままなので間違ってると思います。
|
|
298
|
+
|
|
299
|
+
|
|
300
|
+
|
|
301
|
+
デスクトップ上ではできます。
|
1
一部、編集。
test
CHANGED
|
File without changes
|
test
CHANGED
|
@@ -6,6 +6,18 @@
|
|
|
6
6
|
|
|
7
7
|
```python
|
|
8
8
|
|
|
9
|
+
from google.colab import drive
|
|
10
|
+
|
|
11
|
+
drive.mount('/content/drive')
|
|
12
|
+
|
|
13
|
+
|
|
14
|
+
|
|
15
|
+
import sys
|
|
16
|
+
|
|
17
|
+
sys.path.append('/content/drive/My Drive/colab')
|
|
18
|
+
|
|
19
|
+
|
|
20
|
+
|
|
9
21
|
import numpy as np
|
|
10
22
|
|
|
11
23
|
import ActivationFunction as AF
|
|
@@ -234,19 +246,23 @@
|
|
|
234
246
|
|
|
235
247
|
|
|
236
248
|
|
|
249
|
+
Drive already mounted at /content/drive; to attempt to forcibly remount, call drive.mount("/content/drive", force_remount=True).
|
|
250
|
+
|
|
251
|
+
---------------------------------------------------------------------------
|
|
252
|
+
|
|
237
253
|
ModuleNotFoundError Traceback (most recent call last)
|
|
238
254
|
|
|
239
|
-
<ipython-input-
|
|
255
|
+
<ipython-input-5-21a5c5cca7e6> in <module>()
|
|
256
|
+
|
|
240
|
-
|
|
257
|
+
6
|
|
258
|
+
|
|
241
|
-
|
|
259
|
+
7 import numpy as np
|
|
242
|
-
|
|
260
|
+
|
|
243
|
-
---->
|
|
261
|
+
----> 8 import ActivationFunction as AF
|
|
244
|
-
|
|
262
|
+
|
|
245
|
-
|
|
263
|
+
9
|
|
246
|
-
|
|
264
|
+
|
|
247
|
-
|
|
265
|
+
10 # 3層ニューラルネットワーク
|
|
248
|
-
|
|
249
|
-
5 class ThreeLayerNetwork:
|
|
250
266
|
|
|
251
267
|
|
|
252
268
|
|
|
@@ -254,4 +270,20 @@
|
|
|
254
270
|
|
|
255
271
|
|
|
256
272
|
|
|
273
|
+
---------------------------------------------------------------------------
|
|
274
|
+
|
|
275
|
+
NOTE: If your import is failing due to a missing package, you can
|
|
276
|
+
|
|
277
|
+
manually install dependencies using either !pip or !apt.
|
|
278
|
+
|
|
279
|
+
|
|
280
|
+
|
|
281
|
+
To view examples of installing some common dependencies, click the
|
|
282
|
+
|
|
283
|
+
"Open Examples" button below.
|
|
284
|
+
|
|
285
|
+
---------------------------------------------------------------------------
|
|
286
|
+
|
|
287
|
+
|
|
288
|
+
|
|
257
289
|
と出ます、どうすれば良いでしょうか？
|