キカガク（KIKAGAKU）で「畳み込みニューラルネットワークの実装」を学ぶ ~ rteak

Photo by Owen Beard on Unsplash

今回は「畳み込みニューラルネットワークの実装 - KIKAGAKU」を学ぶ。

Table of Contents [Disable]

1.学習内容
2.ソースコード
3.出力結果
4.実行ログ
5.関連記事

学習内容

データセットの準備

Tensorflow で使用できる形式に変換

CNN モデルの定義

目的関数と最適化手法の選択
モデルの学習
予測精度の評価

CNN モデルの順伝播の流れ

Convolution 層の計算
Pooling 層の計算
ベクトル化

ソースコード

	import numpy as np
	import pandas as pd
	import matplotlib.pyplot as plt

	import tensorflow as tf

	# GPU が使用可能であることを確認
	from tensorflow.python.client import device_lib
	print(device_lib.list_local_devices())

	from tensorflow.keras.datasets import mnist

	# データセットの取得
	train, test = mnist.load_data()

	print(len(train))

	# 1 つ目の要素の確認
	print(type(train[0]))
	print(train[0])

	# 1 目の要素の形を確認
	print(train[0].shape)

	img = train[0][0] # 画像データセットの 1 サンプル目を抽出
	plt.imshow(img, cmap='gray')
	plt.show()

	# 2 つ目の要素の確認
	print(type(train[1]))
	print(train[1])
	print(train[1].shape)

	# height, width, channel への変換と正規化
	x_train = train[0].reshape(60000, 28, 28, 1) / 255
	x_test = test[0].reshape(10000, 28, 28, 1) / 255

	# チャネルが追加されていることを確認
	print(x_train[0].shape)

	# 正規化されていることを確認
	print(x_train[0].min(), x_train[0].max())

	# 目標値を学習用とテスト用に分割
	t_train = train[1]
	t_test = test[1]

	# データ型変換
	x_train, x_test = x_train.astype('float32'), x_test.astype('float32')
	t_train, t_test = t_train.astype('int32'), t_test.astype('int32')

	# CNN モデルの定義
	import os, random

	def reset_seed(seed=0):
	os.environ['PYTHONHASHSEED'] = '0'
	random.seed(seed)
	np.random.seed(seed)
	tf.random.set_seed(seed)

	# CNN モデルの構築
	from tensorflow.keras import models,layers

	# シードの固定
	reset_seed(0)

	# モデルの構築
	model = models.Sequential([
	# 特徴量抽出
	layers.Conv2D(filters=3, kernel_size=(3, 3), activation='relu', input_shape=(28, 28, 1)),
	layers.MaxPool2D(pool_size=(2, 2)),
	# ベクトル化
	layers.Flatten(),
	# 識別
	layers.Dense(100, activation='relu'),
	layers.Dense(10, activation='softmax')
	])

	# パラメータの確認
	print(model.summary())

	# 構造のプロット
	from tensorflow.keras.utils import plot_model
	print(plot_model(model))

	# 目的関数と最適化手法の選択
	# optimizer の設定
	optimizer = tf.keras.optimizers.Adam(lr=0.01)

	# モデルのコンパイル
	model.compile(optimizer=optimizer,
	loss='sparse_categorical_crossentropy',
	metrics=['accuracy'])

	# モデルの学習
	batch_size = 4096
	epochs = 30

	# 学習の実行
	history = model.fit(x_train, t_train,
	batch_size=batch_size,
	epochs=epochs, verbose=1,
	validation_data=(x_test, t_test))

	# 予測精度の評価
	# 学習結果の表示
	results = pd.DataFrame(history.history)
	print(results.tail(3))

	# 損失を可視化
	results[['loss', 'val_loss']].plot(title='loss')
	plt.xlabel('epochs')
	plt.show()

	# 正解率を可視化
	results[['accuracy', 'val_accuracy']].plot(title='accuracy')
	plt.xlabel('epochs')
	plt.show()

	# CNN モデルの順伝播の流れ
	# 推論に使用するデータを切り出し + バッチサイズの追加
	x_sample = np.array([x_train[0]])
	print(x_sample.shape)

	# 学習済みモデルの層
	print(model.layers)

	# 切り出した重みの取得
	print(model.layers[0].get_weights())

	# Convolution 層の計算
	output = model.layers[0](x_sample) # convolution 層の計算
	output = output[0].numpy() # NumPy の ndarray オブジェクトに変換

	print(output.shape)

	# 1 つ目の出力
	plt.imshow(output[:, :, 0], cmap='gray')
	plt.show()

	# 2 つ目の出力
	plt.imshow(output[:, :, 1], cmap='gray')
	plt.show()

	# 3 つ目の出力
	plt.imshow(output[:, :, 2], cmap='gray')
	plt.show()

	# Pooling 層の計算
	output = model.layers[0](x_sample) # convolution 層の計算
	output = model.layers[1](output) # pooling 層の計算（サイズを 1/2 に変換）
	output = output[0].numpy()

	print(output.shape)

	# 1 つ目の出力
	plt.imshow(output[:, :, 0], cmap='gray')
	plt.show()

	# 2 つ目の出力
	plt.imshow(output[:, :, 1], cmap='gray')
	plt.show()

	# 3 つ目の出力
	plt.imshow(output[:, :, 2], cmap='gray')
	plt.show()

	# ベクトル化
	output = model.layers[0](x_sample) # convolution 層の計算
	output = model.layers[1](output) # pooling 層の計算（サイズを 1/2 に変換）
	output = model.layers[2](output) # ベクトル化
	output = output[0].numpy()

	print(output.shape)

view raw image_recognition_2.py hosted with ❤ by GitHub

出力結果

データセット読み込み後、1つ目の画像を抽出。

予測精度の評価。学習結果を確認するために損失を可視化。

正解率を可視化。

Convolution 層のフィルタを通して出力されるデータ毎に可視化。1つ目。

2つ目。

3つ目。

Pooling 層を通して出力されるデータ毎に可視化。1つ目。

2つ目。

3つ目。

実行ログ

	[name: "/device:CPU:0"
	device_type: "CPU"
	memory_limit: 268435456
	locality {
	}
	incarnation: 13236399410543303316
	]
	2
	<class 'numpy.ndarray'>
	[[[0 0 0 ... 0 0 0]
	[0 0 0 ... 0 0 0]
	[0 0 0 ... 0 0 0]
	...
	[0 0 0 ... 0 0 0]
	[0 0 0 ... 0 0 0]
	[0 0 0 ... 0 0 0]]

	[[0 0 0 ... 0 0 0]
	[0 0 0 ... 0 0 0]
	[0 0 0 ... 0 0 0]
	...
	[0 0 0 ... 0 0 0]
	[0 0 0 ... 0 0 0]
	[0 0 0 ... 0 0 0]]

	[[0 0 0 ... 0 0 0]
	[0 0 0 ... 0 0 0]
	[0 0 0 ... 0 0 0]
	...
	[0 0 0 ... 0 0 0]
	[0 0 0 ... 0 0 0]
	[0 0 0 ... 0 0 0]]

	...

	[[0 0 0 ... 0 0 0]
	[0 0 0 ... 0 0 0]
	[0 0 0 ... 0 0 0]
	...
	[0 0 0 ... 0 0 0]
	[0 0 0 ... 0 0 0]
	[0 0 0 ... 0 0 0]]

	[[0 0 0 ... 0 0 0]
	[0 0 0 ... 0 0 0]
	[0 0 0 ... 0 0 0]
	...
	[0 0 0 ... 0 0 0]
	[0 0 0 ... 0 0 0]
	[0 0 0 ... 0 0 0]]

	[[0 0 0 ... 0 0 0]
	[0 0 0 ... 0 0 0]
	[0 0 0 ... 0 0 0]
	...
	[0 0 0 ... 0 0 0]
	[0 0 0 ... 0 0 0]
	[0 0 0 ... 0 0 0]]]
	(60000, 28, 28)
	<class 'numpy.ndarray'>
	[5 0 4 ... 5 6 8]
	(60000,)
	(28, 28, 1)
	0.0 1.0
	Model: "sequential"
	_________________________________________________________________
	Layer (type) Output Shape Param #
	=================================================================
	conv2d (Conv2D) (None, 26, 26, 3) 30
	_________________________________________________________________
	max_pooling2d (MaxPooling2D) (None, 13, 13, 3) 0
	_________________________________________________________________
	flatten (Flatten) (None, 507) 0
	_________________________________________________________________
	dense (Dense) (None, 100) 50800
	_________________________________________________________________
	dense_1 (Dense) (None, 10) 1010
	=================================================================
	Total params: 51,840
	Trainable params: 51,840
	Non-trainable params: 0
	_________________________________________________________________
	None
	None
	Train on 60000 samples, validate on 10000 samples
	Epoch 1/30

	4096/60000 [=>............................] - ETA: 22s - loss: 2.4218 - accuracy: 0.0925
	8192/60000 [===>..........................] - ETA: 12s - loss: 2.1954 - accuracy: 0.2791
	12288/60000 [=====>........................] - ETA: 9s - loss: 1.9989 - accuracy: 0.3930
	16384/60000 [=======>......................] - ETA: 7s - loss: 1.8076 - accuracy: 0.4697
	20480/60000 [=========>....................] - ETA: 6s - loss: 1.6403 - accuracy: 0.5237
	24576/60000 [===========>..................] - ETA: 5s - loss: 1.4936 - accuracy: 0.5659
	28672/60000 [=============>................] - ETA: 4s - loss: 1.3750 - accuracy: 0.5969
	32768/60000 [===============>..............] - ETA: 3s - loss: 1.2725 - accuracy: 0.6250
	36864/60000 [=================>............] - ETA: 2s - loss: 1.1860 - accuracy: 0.6492
	40960/60000 [===================>..........] - ETA: 2s - loss: 1.1180 - accuracy: 0.6685
	45056/60000 [=====================>........] - ETA: 1s - loss: 1.0581 - accuracy: 0.6862
	49152/60000 [=======================>......] - ETA: 1s - loss: 1.0071 - accuracy: 0.7018
	53248/60000 [=========================>....] - ETA: 0s - loss: 0.9646 - accuracy: 0.7141
	57344/60000 [===========================>..] - ETA: 0s - loss: 0.9236 - accuracy: 0.7260
	60000/60000 [==============================] - 7s 117us/sample - loss: 0.9005 - accuracy: 0.7333 - val_loss: 0.3815 - val_accuracy: 0.8948
	Epoch 2/30

	4096/60000 [=>............................] - ETA: 4s - loss: 0.3643 - accuracy: 0.8975
	8192/60000 [===>..........................] - ETA: 4s - loss: 0.3985 - accuracy: 0.8896
	12288/60000 [=====>........................] - ETA: 4s - loss: 0.3881 - accuracy: 0.8893
	16384/60000 [=======>......................] - ETA: 3s - loss: 0.3851 - accuracy: 0.8896
	20480/60000 [=========>....................] - ETA: 3s - loss: 0.3761 - accuracy: 0.8926
	24576/60000 [===========>..................] - ETA: 3s - loss: 0.3705 - accuracy: 0.8948
	28672/60000 [=============>................] - ETA: 2s - loss: 0.3625 - accuracy: 0.8964
	32768/60000 [===============>..............] - ETA: 2s - loss: 0.3587 - accuracy: 0.8974
	36864/60000 [=================>............] - ETA: 1s - loss: 0.3554 - accuracy: 0.8981
	40960/60000 [===================>..........] - ETA: 1s - loss: 0.3493 - accuracy: 0.9003
	45056/60000 [=====================>........] - ETA: 1s - loss: 0.3462 - accuracy: 0.9011
	49152/60000 [=======================>......] - ETA: 0s - loss: 0.3416 - accuracy: 0.9021
	53248/60000 [=========================>....] - ETA: 0s - loss: 0.3364 - accuracy: 0.9035
	57344/60000 [===========================>..] - ETA: 0s - loss: 0.3316 - accuracy: 0.9046
	60000/60000 [==============================] - 5s 92us/sample - loss: 0.3285 - accuracy: 0.9055 - val_loss: 0.2628 - val_accuracy: 0.9209
	Epoch 3/30

	4096/60000 [=>............................] - ETA: 4s - loss: 0.2686 - accuracy: 0.9197
	8192/60000 [===>..........................] - ETA: 4s - loss: 0.2810 - accuracy: 0.9185
	12288/60000 [=====>........................] - ETA: 4s - loss: 0.2694 - accuracy: 0.9211
	16384/60000 [=======>......................] - ETA: 3s - loss: 0.2659 - accuracy: 0.9231
	20480/60000 [=========>....................] - ETA: 3s - loss: 0.2645 - accuracy: 0.9232
	24576/60000 [===========>..................] - ETA: 3s - loss: 0.2602 - accuracy: 0.9246
	28672/60000 [=============>................] - ETA: 2s - loss: 0.2569 - accuracy: 0.9252
	32768/60000 [===============>..............] - ETA: 2s - loss: 0.2546 - accuracy: 0.9265
	36864/60000 [=================>............] - ETA: 2s - loss: 0.2514 - accuracy: 0.9271
	40960/60000 [===================>..........] - ETA: 1s - loss: 0.2491 - accuracy: 0.9274
	45056/60000 [=====================>........] - ETA: 1s - loss: 0.2449 - accuracy: 0.9288
	49152/60000 [=======================>......] - ETA: 0s - loss: 0.2420 - accuracy: 0.9296
	53248/60000 [=========================>....] - ETA: 0s - loss: 0.2407 - accuracy: 0.9297
	57344/60000 [===========================>..] - ETA: 0s - loss: 0.2383 - accuracy: 0.9304
	60000/60000 [==============================] - 6s 93us/sample - loss: 0.2376 - accuracy: 0.9307 - val_loss: 0.2018 - val_accuracy: 0.9366
	Epoch 4/30

	4096/60000 [=>............................] - ETA: 4s - loss: 0.2013 - accuracy: 0.9448
	8192/60000 [===>..........................] - ETA: 4s - loss: 0.2076 - accuracy: 0.9415
	12288/60000 [=====>........................] - ETA: 4s - loss: 0.2024 - accuracy: 0.9408
	16384/60000 [=======>......................] - ETA: 3s - loss: 0.2013 - accuracy: 0.9413
	20480/60000 [=========>....................] - ETA: 3s - loss: 0.2025 - accuracy: 0.9405
	24576/60000 [===========>..................] - ETA: 3s - loss: 0.2023 - accuracy: 0.9406
	28672/60000 [=============>................] - ETA: 2s - loss: 0.1978 - accuracy: 0.9422
	32768/60000 [===============>..............] - ETA: 2s - loss: 0.1951 - accuracy: 0.9426
	36864/60000 [=================>............] - ETA: 1s - loss: 0.1928 - accuracy: 0.9431
	40960/60000 [===================>..........] - ETA: 1s - loss: 0.1892 - accuracy: 0.9441
	45056/60000 [=====================>........] - ETA: 1s - loss: 0.1894 - accuracy: 0.9439
	49152/60000 [=======================>......] - ETA: 0s - loss: 0.1878 - accuracy: 0.9442
	53248/60000 [=========================>....] - ETA: 0s - loss: 0.1885 - accuracy: 0.9444
	57344/60000 [===========================>..] - ETA: 0s - loss: 0.1891 - accuracy: 0.9442
	60000/60000 [==============================] - 5s 91us/sample - loss: 0.1883 - accuracy: 0.9445 - val_loss: 0.1668 - val_accuracy: 0.9465
	Epoch 5/30

	4096/60000 [=>............................] - ETA: 4s - loss: 0.1623 - accuracy: 0.9546
	8192/60000 [===>..........................] - ETA: 4s - loss: 0.1689 - accuracy: 0.9509
	12288/60000 [=====>........................] - ETA: 4s - loss: 0.1715 - accuracy: 0.9492
	16384/60000 [=======>......................] - ETA: 3s - loss: 0.1663 - accuracy: 0.9503
	20480/60000 [=========>....................] - ETA: 3s - loss: 0.1664 - accuracy: 0.9501
	24576/60000 [===========>..................] - ETA: 2s - loss: 0.1653 - accuracy: 0.9506
	28672/60000 [=============>................] - ETA: 2s - loss: 0.1648 - accuracy: 0.9506
	32768/60000 [===============>..............] - ETA: 2s - loss: 0.1635 - accuracy: 0.9510
	36864/60000 [=================>............] - ETA: 1s - loss: 0.1614 - accuracy: 0.9516
	40960/60000 [===================>..........] - ETA: 1s - loss: 0.1611 - accuracy: 0.9514
	45056/60000 [=====================>........] - ETA: 1s - loss: 0.1603 - accuracy: 0.9517
	49152/60000 [=======================>......] - ETA: 0s - loss: 0.1594 - accuracy: 0.9522
	53248/60000 [=========================>....] - ETA: 0s - loss: 0.1590 - accuracy: 0.9520
	57344/60000 [===========================>..] - ETA: 0s - loss: 0.1567 - accuracy: 0.9529
	60000/60000 [==============================] - 5s 91us/sample - loss: 0.1562 - accuracy: 0.9531 - val_loss: 0.1433 - val_accuracy: 0.9547
	Epoch 6/30

	4096/60000 [=>............................] - ETA: 4s - loss: 0.1276 - accuracy: 0.9639
	8192/60000 [===>..........................] - ETA: 4s - loss: 0.1416 - accuracy: 0.9597
	12288/60000 [=====>........................] - ETA: 3s - loss: 0.1406 - accuracy: 0.9598
	16384/60000 [=======>......................] - ETA: 3s - loss: 0.1412 - accuracy: 0.9595
	20480/60000 [=========>....................] - ETA: 3s - loss: 0.1405 - accuracy: 0.9591
	24576/60000 [===========>..................] - ETA: 2s - loss: 0.1395 - accuracy: 0.9599
	28672/60000 [=============>................] - ETA: 2s - loss: 0.1382 - accuracy: 0.9600
	32768/60000 [===============>..............] - ETA: 2s - loss: 0.1365 - accuracy: 0.9599
	36864/60000 [=================>............] - ETA: 1s - loss: 0.1370 - accuracy: 0.9595
	40960/60000 [===================>..........] - ETA: 1s - loss: 0.1366 - accuracy: 0.9599
	45056/60000 [=====================>........] - ETA: 1s - loss: 0.1353 - accuracy: 0.9601
	49152/60000 [=======================>......] - ETA: 0s - loss: 0.1338 - accuracy: 0.9605
	53248/60000 [=========================>....] - ETA: 0s - loss: 0.1340 - accuracy: 0.9604
	57344/60000 [===========================>..] - ETA: 0s - loss: 0.1333 - accuracy: 0.9607
	60000/60000 [==============================] - 5s 92us/sample - loss: 0.1334 - accuracy: 0.9605 - val_loss: 0.1293 - val_accuracy: 0.9592
	Epoch 7/30

	4096/60000 [=>............................] - ETA: 4s - loss: 0.1419 - accuracy: 0.9602
	8192/60000 [===>..........................] - ETA: 4s - loss: 0.1305 - accuracy: 0.9618
	12288/60000 [=====>........................] - ETA: 4s - loss: 0.1245 - accuracy: 0.9644
	16384/60000 [=======>......................] - ETA: 3s - loss: 0.1216 - accuracy: 0.9650
	20480/60000 [=========>....................] - ETA: 3s - loss: 0.1172 - accuracy: 0.9663
	24576/60000 [===========>..................] - ETA: 3s - loss: 0.1161 - accuracy: 0.9661
	28672/60000 [=============>................] - ETA: 2s - loss: 0.1164 - accuracy: 0.9655
	32768/60000 [===============>..............] - ETA: 2s - loss: 0.1170 - accuracy: 0.9654
	36864/60000 [=================>............] - ETA: 1s - loss: 0.1168 - accuracy: 0.9654
	40960/60000 [===================>..........] - ETA: 1s - loss: 0.1173 - accuracy: 0.9654
	45056/60000 [=====================>........] - ETA: 1s - loss: 0.1169 - accuracy: 0.9653
	49152/60000 [=======================>......] - ETA: 0s - loss: 0.1164 - accuracy: 0.9653
	53248/60000 [=========================>....] - ETA: 0s - loss: 0.1165 - accuracy: 0.9650
	57344/60000 [===========================>..] - ETA: 0s - loss: 0.1157 - accuracy: 0.9654
	60000/60000 [==============================] - 6s 93us/sample - loss: 0.1155 - accuracy: 0.9654 - val_loss: 0.1108 - val_accuracy: 0.9654
	Epoch 8/30

	4096/60000 [=>............................] - ETA: 4s - loss: 0.1004 - accuracy: 0.9678
	8192/60000 [===>..........................] - ETA: 4s - loss: 0.0994 - accuracy: 0.9688
	12288/60000 [=====>........................] - ETA: 4s - loss: 0.0976 - accuracy: 0.9700
	16384/60000 [=======>......................] - ETA: 3s - loss: 0.0981 - accuracy: 0.9702
	20480/60000 [=========>....................] - ETA: 3s - loss: 0.0970 - accuracy: 0.9703
	24576/60000 [===========>..................] - ETA: 3s - loss: 0.0971 - accuracy: 0.9698
	28672/60000 [=============>................] - ETA: 2s - loss: 0.0985 - accuracy: 0.9700
	32768/60000 [===============>..............] - ETA: 2s - loss: 0.0967 - accuracy: 0.9704
	36864/60000 [=================>............] - ETA: 2s - loss: 0.0962 - accuracy: 0.9706
	40960/60000 [===================>..........] - ETA: 1s - loss: 0.0967 - accuracy: 0.9707
	45056/60000 [=====================>........] - ETA: 1s - loss: 0.0958 - accuracy: 0.9709
	49152/60000 [=======================>......] - ETA: 0s - loss: 0.0968 - accuracy: 0.9706
	53248/60000 [=========================>....] - ETA: 0s - loss: 0.0971 - accuracy: 0.9707
	57344/60000 [===========================>..] - ETA: 0s - loss: 0.0985 - accuracy: 0.9705
	60000/60000 [==============================] - 6s 95us/sample - loss: 0.0985 - accuracy: 0.9706 - val_loss: 0.0990 - val_accuracy: 0.9686
	Epoch 9/30

	4096/60000 [=>............................] - ETA: 4s - loss: 0.0931 - accuracy: 0.9734
	8192/60000 [===>..........................] - ETA: 4s - loss: 0.0986 - accuracy: 0.9707
	12288/60000 [=====>........................] - ETA: 4s - loss: 0.0982 - accuracy: 0.9713
	16384/60000 [=======>......................] - ETA: 3s - loss: 0.0943 - accuracy: 0.9720
	20480/60000 [=========>....................] - ETA: 3s - loss: 0.0926 - accuracy: 0.9723
	24576/60000 [===========>..................] - ETA: 3s - loss: 0.0915 - accuracy: 0.9727
	28672/60000 [=============>................] - ETA: 2s - loss: 0.0936 - accuracy: 0.9719
	32768/60000 [===============>..............] - ETA: 2s - loss: 0.0935 - accuracy: 0.9716
	36864/60000 [=================>............] - ETA: 2s - loss: 0.0921 - accuracy: 0.9723
	40960/60000 [===================>..........] - ETA: 1s - loss: 0.0909 - accuracy: 0.9726
	45056/60000 [=====================>........] - ETA: 1s - loss: 0.0909 - accuracy: 0.9726
	49152/60000 [=======================>......] - ETA: 0s - loss: 0.0905 - accuracy: 0.9725
	53248/60000 [=========================>....] - ETA: 0s - loss: 0.0912 - accuracy: 0.9723
	57344/60000 [===========================>..] - ETA: 0s - loss: 0.0906 - accuracy: 0.9724
	60000/60000 [==============================] - 6s 95us/sample - loss: 0.0900 - accuracy: 0.9727 - val_loss: 0.1048 - val_accuracy: 0.9661
	Epoch 10/30

	4096/60000 [=>............................] - ETA: 5s - loss: 0.0806 - accuracy: 0.9744
	8192/60000 [===>..........................] - ETA: 4s - loss: 0.0882 - accuracy: 0.9730
	12288/60000 [=====>........................] - ETA: 4s - loss: 0.0862 - accuracy: 0.9733
	16384/60000 [=======>......................] - ETA: 3s - loss: 0.0855 - accuracy: 0.9738
	20480/60000 [=========>....................] - ETA: 3s - loss: 0.0842 - accuracy: 0.9746
	24576/60000 [===========>..................] - ETA: 3s - loss: 0.0828 - accuracy: 0.9749
	28672/60000 [=============>................] - ETA: 2s - loss: 0.0813 - accuracy: 0.9754
	32768/60000 [===============>..............] - ETA: 2s - loss: 0.0817 - accuracy: 0.9753
	36864/60000 [=================>............] - ETA: 2s - loss: 0.0818 - accuracy: 0.9752
	40960/60000 [===================>..........] - ETA: 1s - loss: 0.0819 - accuracy: 0.9753
	45056/60000 [=====================>........] - ETA: 1s - loss: 0.0814 - accuracy: 0.9751
	49152/60000 [=======================>......] - ETA: 0s - loss: 0.0811 - accuracy: 0.9752
	53248/60000 [=========================>....] - ETA: 0s - loss: 0.0812 - accuracy: 0.9751
	57344/60000 [===========================>..] - ETA: 0s - loss: 0.0820 - accuracy: 0.9748
	60000/60000 [==============================] - 6s 96us/sample - loss: 0.0817 - accuracy: 0.9748 - val_loss: 0.0892 - val_accuracy: 0.9714
	Epoch 11/30

	4096/60000 [=>............................] - ETA: 5s - loss: 0.0764 - accuracy: 0.9749
	8192/60000 [===>..........................] - ETA: 4s - loss: 0.0846 - accuracy: 0.9722
	12288/60000 [=====>........................] - ETA: 4s - loss: 0.0788 - accuracy: 0.9757
	16384/60000 [=======>......................] - ETA: 3s - loss: 0.0775 - accuracy: 0.9761
	20480/60000 [=========>....................] - ETA: 3s - loss: 0.0791 - accuracy: 0.9756
	24576/60000 [===========>..................] - ETA: 3s - loss: 0.0776 - accuracy: 0.9759
	28672/60000 [=============>................] - ETA: 2s - loss: 0.0786 - accuracy: 0.9755
	32768/60000 [===============>..............] - ETA: 2s - loss: 0.0782 - accuracy: 0.9760
	36864/60000 [=================>............] - ETA: 2s - loss: 0.0777 - accuracy: 0.9762
	40960/60000 [===================>..........] - ETA: 1s - loss: 0.0768 - accuracy: 0.9764
	45056/60000 [=====================>........] - ETA: 1s - loss: 0.0770 - accuracy: 0.9762
	49152/60000 [=======================>......] - ETA: 0s - loss: 0.0775 - accuracy: 0.9761
	53248/60000 [=========================>....] - ETA: 0s - loss: 0.0774 - accuracy: 0.9761
	57344/60000 [===========================>..] - ETA: 0s - loss: 0.0767 - accuracy: 0.9763
	60000/60000 [==============================] - 6s 97us/sample - loss: 0.0769 - accuracy: 0.9761 - val_loss: 0.0852 - val_accuracy: 0.9745
	Epoch 12/30

	4096/60000 [=>............................] - ETA: 5s - loss: 0.0669 - accuracy: 0.9785
	8192/60000 [===>..........................] - ETA: 4s - loss: 0.0698 - accuracy: 0.9780
	12288/60000 [=====>........................] - ETA: 4s - loss: 0.0695 - accuracy: 0.9787
	16384/60000 [=======>......................] - ETA: 3s - loss: 0.0687 - accuracy: 0.9783
	20480/60000 [=========>....................] - ETA: 3s - loss: 0.0696 - accuracy: 0.9781
	24576/60000 [===========>..................] - ETA: 3s - loss: 0.0691 - accuracy: 0.9781
	28672/60000 [=============>................] - ETA: 2s - loss: 0.0684 - accuracy: 0.9783
	32768/60000 [===============>..............] - ETA: 2s - loss: 0.0668 - accuracy: 0.9788
	36864/60000 [=================>............] - ETA: 2s - loss: 0.0684 - accuracy: 0.9785
	40960/60000 [===================>..........] - ETA: 1s - loss: 0.0685 - accuracy: 0.9787
	45056/60000 [=====================>........] - ETA: 1s - loss: 0.0687 - accuracy: 0.9785
	49152/60000 [=======================>......] - ETA: 0s - loss: 0.0690 - accuracy: 0.9784
	53248/60000 [=========================>....] - ETA: 0s - loss: 0.0687 - accuracy: 0.9785
	57344/60000 [===========================>..] - ETA: 0s - loss: 0.0689 - accuracy: 0.9786
	60000/60000 [==============================] - 6s 96us/sample - loss: 0.0690 - accuracy: 0.9786 - val_loss: 0.0824 - val_accuracy: 0.9749
	Epoch 13/30

	4096/60000 [=>............................] - ETA: 4s - loss: 0.0714 - accuracy: 0.9768
	8192/60000 [===>..........................] - ETA: 4s - loss: 0.0639 - accuracy: 0.9806
	12288/60000 [=====>........................] - ETA: 4s - loss: 0.0648 - accuracy: 0.9798
	16384/60000 [=======>......................] - ETA: 3s - loss: 0.0639 - accuracy: 0.9806
	20480/60000 [=========>....................] - ETA: 3s - loss: 0.0629 - accuracy: 0.9813
	24576/60000 [===========>..................] - ETA: 3s - loss: 0.0629 - accuracy: 0.9815
	28672/60000 [=============>................] - ETA: 2s - loss: 0.0633 - accuracy: 0.9810
	32768/60000 [===============>..............] - ETA: 2s - loss: 0.0637 - accuracy: 0.9809
	36864/60000 [=================>............] - ETA: 1s - loss: 0.0636 - accuracy: 0.9809
	40960/60000 [===================>..........] - ETA: 1s - loss: 0.0635 - accuracy: 0.9809
	45056/60000 [=====================>........] - ETA: 1s - loss: 0.0628 - accuracy: 0.9810
	49152/60000 [=======================>......] - ETA: 0s - loss: 0.0621 - accuracy: 0.9814
	53248/60000 [=========================>....] - ETA: 0s - loss: 0.0621 - accuracy: 0.9813
	57344/60000 [===========================>..] - ETA: 0s - loss: 0.0629 - accuracy: 0.9810
	60000/60000 [==============================] - 6s 92us/sample - loss: 0.0629 - accuracy: 0.9810 - val_loss: 0.0771 - val_accuracy: 0.9760
	Epoch 14/30

	4096/60000 [=>............................] - ETA: 4s - loss: 0.0471 - accuracy: 0.9851
	8192/60000 [===>..........................] - ETA: 4s - loss: 0.0542 - accuracy: 0.9838
	12288/60000 [=====>........................] - ETA: 4s - loss: 0.0528 - accuracy: 0.9832
	16384/60000 [=======>......................] - ETA: 3s - loss: 0.0568 - accuracy: 0.9826
	20480/60000 [=========>....................] - ETA: 3s - loss: 0.0556 - accuracy: 0.9832
	24576/60000 [===========>..................] - ETA: 3s - loss: 0.0553 - accuracy: 0.9832
	28672/60000 [=============>................] - ETA: 2s - loss: 0.0539 - accuracy: 0.9838
	32768/60000 [===============>..............] - ETA: 2s - loss: 0.0537 - accuracy: 0.9837
	36864/60000 [=================>............] - ETA: 1s - loss: 0.0531 - accuracy: 0.9842
	40960/60000 [===================>..........] - ETA: 1s - loss: 0.0529 - accuracy: 0.9842
	45056/60000 [=====================>........] - ETA: 1s - loss: 0.0533 - accuracy: 0.9840
	49152/60000 [=======================>......] - ETA: 0s - loss: 0.0538 - accuracy: 0.9837
	53248/60000 [=========================>....] - ETA: 0s - loss: 0.0536 - accuracy: 0.9837
	57344/60000 [===========================>..] - ETA: 0s - loss: 0.0541 - accuracy: 0.9836
	60000/60000 [==============================] - 6s 92us/sample - loss: 0.0549 - accuracy: 0.9834 - val_loss: 0.0779 - val_accuracy: 0.9756
	Epoch 15/30

	4096/60000 [=>............................] - ETA: 5s - loss: 0.0568 - accuracy: 0.9829
	8192/60000 [===>..........................] - ETA: 4s - loss: 0.0537 - accuracy: 0.9844
	12288/60000 [=====>........................] - ETA: 4s - loss: 0.0530 - accuracy: 0.9848
	16384/60000 [=======>......................] - ETA: 3s - loss: 0.0539 - accuracy: 0.9843
	20480/60000 [=========>....................] - ETA: 3s - loss: 0.0522 - accuracy: 0.9849
	24576/60000 [===========>..................] - ETA: 3s - loss: 0.0521 - accuracy: 0.9848
	28672/60000 [=============>................] - ETA: 2s - loss: 0.0517 - accuracy: 0.9851
	32768/60000 [===============>..............] - ETA: 2s - loss: 0.0516 - accuracy: 0.9848
	36864/60000 [=================>............] - ETA: 2s - loss: 0.0511 - accuracy: 0.9848
	40960/60000 [===================>..........] - ETA: 1s - loss: 0.0509 - accuracy: 0.9847
	45056/60000 [=====================>........] - ETA: 1s - loss: 0.0515 - accuracy: 0.9847
	49152/60000 [=======================>......] - ETA: 0s - loss: 0.0510 - accuracy: 0.9848
	53248/60000 [=========================>....] - ETA: 0s - loss: 0.0508 - accuracy: 0.9849
	57344/60000 [===========================>..] - ETA: 0s - loss: 0.0511 - accuracy: 0.9845
	60000/60000 [==============================] - 6s 96us/sample - loss: 0.0511 - accuracy: 0.9844 - val_loss: 0.0777 - val_accuracy: 0.9773
	Epoch 16/30

	4096/60000 [=>............................] - ETA: 5s - loss: 0.0478 - accuracy: 0.9844
	8192/60000 [===>..........................] - ETA: 4s - loss: 0.0466 - accuracy: 0.9845
	12288/60000 [=====>........................] - ETA: 4s - loss: 0.0430 - accuracy: 0.9865
	16384/60000 [=======>......................] - ETA: 3s - loss: 0.0438 - accuracy: 0.9863
	20480/60000 [=========>....................] - ETA: 3s - loss: 0.0454 - accuracy: 0.9854
	24576/60000 [===========>..................] - ETA: 3s - loss: 0.0447 - accuracy: 0.9857
	28672/60000 [=============>................] - ETA: 2s - loss: 0.0459 - accuracy: 0.9857
	32768/60000 [===============>..............] - ETA: 2s - loss: 0.0460 - accuracy: 0.9854
	36864/60000 [=================>............] - ETA: 2s - loss: 0.0463 - accuracy: 0.9857
	40960/60000 [===================>..........] - ETA: 1s - loss: 0.0463 - accuracy: 0.9857
	45056/60000 [=====================>........] - ETA: 1s - loss: 0.0468 - accuracy: 0.9855
	49152/60000 [=======================>......] - ETA: 0s - loss: 0.0477 - accuracy: 0.9852
	53248/60000 [=========================>....] - ETA: 0s - loss: 0.0480 - accuracy: 0.9851
	57344/60000 [===========================>..] - ETA: 0s - loss: 0.0478 - accuracy: 0.9852
	60000/60000 [==============================] - 6s 96us/sample - loss: 0.0476 - accuracy: 0.9853 - val_loss: 0.0734 - val_accuracy: 0.9774
	Epoch 17/30

	4096/60000 [=>............................] - ETA: 4s - loss: 0.0489 - accuracy: 0.9893
	8192/60000 [===>..........................] - ETA: 4s - loss: 0.0461 - accuracy: 0.9884
	12288/60000 [=====>........................] - ETA: 4s - loss: 0.0457 - accuracy: 0.9874
	16384/60000 [=======>......................] - ETA: 3s - loss: 0.0439 - accuracy: 0.9877
	20480/60000 [=========>....................] - ETA: 3s - loss: 0.0451 - accuracy: 0.9866
	24576/60000 [===========>..................] - ETA: 3s - loss: 0.0450 - accuracy: 0.9864
	28672/60000 [=============>................] - ETA: 2s - loss: 0.0442 - accuracy: 0.9865
	32768/60000 [===============>..............] - ETA: 2s - loss: 0.0442 - accuracy: 0.9864
	36864/60000 [=================>............] - ETA: 2s - loss: 0.0439 - accuracy: 0.9862
	40960/60000 [===================>..........] - ETA: 1s - loss: 0.0437 - accuracy: 0.9862
	45056/60000 [=====================>........] - ETA: 1s - loss: 0.0438 - accuracy: 0.9861
	49152/60000 [=======================>......] - ETA: 0s - loss: 0.0444 - accuracy: 0.9860
	53248/60000 [=========================>....] - ETA: 0s - loss: 0.0449 - accuracy: 0.9858
	57344/60000 [===========================>..] - ETA: 0s - loss: 0.0455 - accuracy: 0.9855
	60000/60000 [==============================] - 6s 96us/sample - loss: 0.0460 - accuracy: 0.9854 - val_loss: 0.0745 - val_accuracy: 0.9779
	Epoch 18/30

	4096/60000 [=>............................] - ETA: 4s - loss: 0.0341 - accuracy: 0.9883
	8192/60000 [===>..........................] - ETA: 4s - loss: 0.0436 - accuracy: 0.9845
	12288/60000 [=====>........................] - ETA: 4s - loss: 0.0420 - accuracy: 0.9857
	16384/60000 [=======>......................] - ETA: 3s - loss: 0.0445 - accuracy: 0.9854
	20480/60000 [=========>....................] - ETA: 3s - loss: 0.0454 - accuracy: 0.9853
	24576/60000 [===========>..................] - ETA: 3s - loss: 0.0457 - accuracy: 0.9851
	28672/60000 [=============>................] - ETA: 2s - loss: 0.0450 - accuracy: 0.9855
	32768/60000 [===============>..............] - ETA: 2s - loss: 0.0452 - accuracy: 0.9854
	36864/60000 [=================>............] - ETA: 2s - loss: 0.0447 - accuracy: 0.9856
	40960/60000 [===================>..........] - ETA: 1s - loss: 0.0445 - accuracy: 0.9856
	45056/60000 [=====================>........] - ETA: 1s - loss: 0.0441 - accuracy: 0.9857
	49152/60000 [=======================>......] - ETA: 0s - loss: 0.0439 - accuracy: 0.9857
	53248/60000 [=========================>....] - ETA: 0s - loss: 0.0440 - accuracy: 0.9858
	57344/60000 [===========================>..] - ETA: 0s - loss: 0.0437 - accuracy: 0.9860
	60000/60000 [==============================] - 6s 94us/sample - loss: 0.0436 - accuracy: 0.9862 - val_loss: 0.0712 - val_accuracy: 0.9779
	Epoch 19/30

	4096/60000 [=>............................] - ETA: 5s - loss: 0.0329 - accuracy: 0.9895
	8192/60000 [===>..........................] - ETA: 4s - loss: 0.0314 - accuracy: 0.9901
	12288/60000 [=====>........................] - ETA: 4s - loss: 0.0327 - accuracy: 0.9889
	16384/60000 [=======>......................] - ETA: 3s - loss: 0.0345 - accuracy: 0.9885
	20480/60000 [=========>....................] - ETA: 3s - loss: 0.0355 - accuracy: 0.9883
	24576/60000 [===========>..................] - ETA: 3s - loss: 0.0360 - accuracy: 0.9883
	28672/60000 [=============>................] - ETA: 2s - loss: 0.0366 - accuracy: 0.9883
	32768/60000 [===============>..............] - ETA: 2s - loss: 0.0374 - accuracy: 0.9882
	36864/60000 [=================>............] - ETA: 2s - loss: 0.0375 - accuracy: 0.9883
	40960/60000 [===================>..........] - ETA: 1s - loss: 0.0380 - accuracy: 0.9880
	45056/60000 [=====================>........] - ETA: 1s - loss: 0.0382 - accuracy: 0.9880
	49152/60000 [=======================>......] - ETA: 0s - loss: 0.0387 - accuracy: 0.9879
	53248/60000 [=========================>....] - ETA: 0s - loss: 0.0390 - accuracy: 0.9875
	57344/60000 [===========================>..] - ETA: 0s - loss: 0.0384 - accuracy: 0.9878
	60000/60000 [==============================] - 6s 97us/sample - loss: 0.0384 - accuracy: 0.9877 - val_loss: 0.0830 - val_accuracy: 0.9753
	Epoch 20/30

	4096/60000 [=>............................] - ETA: 4s - loss: 0.0465 - accuracy: 0.9829
	8192/60000 [===>..........................] - ETA: 4s - loss: 0.0405 - accuracy: 0.9855
	12288/60000 [=====>........................] - ETA: 4s - loss: 0.0374 - accuracy: 0.9870
	16384/60000 [=======>......................] - ETA: 3s - loss: 0.0385 - accuracy: 0.9870
	20480/60000 [=========>....................] - ETA: 3s - loss: 0.0374 - accuracy: 0.9875
	24576/60000 [===========>..................] - ETA: 3s - loss: 0.0377 - accuracy: 0.9876
	28672/60000 [=============>................] - ETA: 2s - loss: 0.0380 - accuracy: 0.9875
	32768/60000 [===============>..............] - ETA: 2s - loss: 0.0385 - accuracy: 0.9874
	36864/60000 [=================>............] - ETA: 2s - loss: 0.0384 - accuracy: 0.9875
	40960/60000 [===================>..........] - ETA: 1s - loss: 0.0377 - accuracy: 0.9877
	45056/60000 [=====================>........] - ETA: 1s - loss: 0.0371 - accuracy: 0.9879
	49152/60000 [=======================>......] - ETA: 0s - loss: 0.0371 - accuracy: 0.9880
	53248/60000 [=========================>....] - ETA: 0s - loss: 0.0368 - accuracy: 0.9882
	57344/60000 [===========================>..] - ETA: 0s - loss: 0.0372 - accuracy: 0.9881
	60000/60000 [==============================] - 6s 96us/sample - loss: 0.0376 - accuracy: 0.9879 - val_loss: 0.0730 - val_accuracy: 0.9798
	Epoch 21/30

	4096/60000 [=>............................] - ETA: 4s - loss: 0.0259 - accuracy: 0.9934
	8192/60000 [===>..........................] - ETA: 4s - loss: 0.0305 - accuracy: 0.9911
	12288/60000 [=====>........................] - ETA: 4s - loss: 0.0302 - accuracy: 0.9915
	16384/60000 [=======>......................] - ETA: 3s - loss: 0.0293 - accuracy: 0.9914
	20480/60000 [=========>....................] - ETA: 3s - loss: 0.0318 - accuracy: 0.9902
	24576/60000 [===========>..................] - ETA: 3s - loss: 0.0335 - accuracy: 0.9896
	28672/60000 [=============>................] - ETA: 2s - loss: 0.0340 - accuracy: 0.9893
	32768/60000 [===============>..............] - ETA: 2s - loss: 0.0344 - accuracy: 0.9890
	36864/60000 [=================>............] - ETA: 2s - loss: 0.0346 - accuracy: 0.9889
	40960/60000 [===================>..........] - ETA: 1s - loss: 0.0343 - accuracy: 0.9890
	45056/60000 [=====================>........] - ETA: 1s - loss: 0.0345 - accuracy: 0.9890
	49152/60000 [=======================>......] - ETA: 0s - loss: 0.0350 - accuracy: 0.9889
	53248/60000 [=========================>....] - ETA: 0s - loss: 0.0343 - accuracy: 0.9891
	57344/60000 [===========================>..] - ETA: 0s - loss: 0.0347 - accuracy: 0.9890
	60000/60000 [==============================] - 6s 95us/sample - loss: 0.0353 - accuracy: 0.9889 - val_loss: 0.0710 - val_accuracy: 0.9784
	Epoch 22/30

	4096/60000 [=>............................] - ETA: 4s - loss: 0.0273 - accuracy: 0.9922
	8192/60000 [===>..........................] - ETA: 4s - loss: 0.0301 - accuracy: 0.9911
	12288/60000 [=====>........................] - ETA: 4s - loss: 0.0306 - accuracy: 0.9904
	16384/60000 [=======>......................] - ETA: 3s - loss: 0.0294 - accuracy: 0.9907
	20480/60000 [=========>....................] - ETA: 3s - loss: 0.0305 - accuracy: 0.9902
	24576/60000 [===========>..................] - ETA: 3s - loss: 0.0308 - accuracy: 0.9903
	28672/60000 [=============>................] - ETA: 2s - loss: 0.0305 - accuracy: 0.9902
	32768/60000 [===============>..............] - ETA: 2s - loss: 0.0305 - accuracy: 0.9901
	36864/60000 [=================>............] - ETA: 2s - loss: 0.0307 - accuracy: 0.9901
	40960/60000 [===================>..........] - ETA: 1s - loss: 0.0310 - accuracy: 0.9900
	45056/60000 [=====================>........] - ETA: 1s - loss: 0.0309 - accuracy: 0.9901
	49152/60000 [=======================>......] - ETA: 0s - loss: 0.0307 - accuracy: 0.9901
	53248/60000 [=========================>....] - ETA: 0s - loss: 0.0307 - accuracy: 0.9903
	57344/60000 [===========================>..] - ETA: 0s - loss: 0.0311 - accuracy: 0.9901
	60000/60000 [==============================] - 6s 94us/sample - loss: 0.0309 - accuracy: 0.9902 - val_loss: 0.0695 - val_accuracy: 0.9799
	Epoch 23/30

	4096/60000 [=>............................] - ETA: 4s - loss: 0.0247 - accuracy: 0.9927
	8192/60000 [===>..........................] - ETA: 4s - loss: 0.0268 - accuracy: 0.9918
	12288/60000 [=====>........................] - ETA: 4s - loss: 0.0276 - accuracy: 0.9912
	16384/60000 [=======>......................] - ETA: 3s - loss: 0.0258 - accuracy: 0.9920
	20480/60000 [=========>....................] - ETA: 3s - loss: 0.0269 - accuracy: 0.9919
	24576/60000 [===========>..................] - ETA: 3s - loss: 0.0264 - accuracy: 0.9920
	28672/60000 [=============>................] - ETA: 2s - loss: 0.0266 - accuracy: 0.9921
	32768/60000 [===============>..............] - ETA: 2s - loss: 0.0269 - accuracy: 0.9920
	36864/60000 [=================>............] - ETA: 1s - loss: 0.0270 - accuracy: 0.9920
	40960/60000 [===================>..........] - ETA: 1s - loss: 0.0278 - accuracy: 0.9916
	45056/60000 [=====================>........] - ETA: 1s - loss: 0.0276 - accuracy: 0.9917
	49152/60000 [=======================>......] - ETA: 0s - loss: 0.0275 - accuracy: 0.9917
	53248/60000 [=========================>....] - ETA: 0s - loss: 0.0274 - accuracy: 0.9918
	57344/60000 [===========================>..] - ETA: 0s - loss: 0.0274 - accuracy: 0.9918
	60000/60000 [==============================] - 6s 92us/sample - loss: 0.0274 - accuracy: 0.9918 - val_loss: 0.0684 - val_accuracy: 0.9798
	Epoch 24/30

	4096/60000 [=>............................] - ETA: 4s - loss: 0.0309 - accuracy: 0.9917
	8192/60000 [===>..........................] - ETA: 4s - loss: 0.0273 - accuracy: 0.9923
	12288/60000 [=====>........................] - ETA: 3s - loss: 0.0279 - accuracy: 0.9923
	16384/60000 [=======>......................] - ETA: 3s - loss: 0.0280 - accuracy: 0.9924
	20480/60000 [=========>....................] - ETA: 3s - loss: 0.0277 - accuracy: 0.9921
	24576/60000 [===========>..................] - ETA: 2s - loss: 0.0266 - accuracy: 0.9925
	28672/60000 [=============>................] - ETA: 2s - loss: 0.0261 - accuracy: 0.9926
	32768/60000 [===============>..............] - ETA: 2s - loss: 0.0258 - accuracy: 0.9925
	36864/60000 [=================>............] - ETA: 1s - loss: 0.0252 - accuracy: 0.9928
	40960/60000 [===================>..........] - ETA: 1s - loss: 0.0254 - accuracy: 0.9929
	45056/60000 [=====================>........] - ETA: 1s - loss: 0.0250 - accuracy: 0.9930
	49152/60000 [=======================>......] - ETA: 0s - loss: 0.0251 - accuracy: 0.9928
	53248/60000 [=========================>....] - ETA: 0s - loss: 0.0250 - accuracy: 0.9927
	57344/60000 [===========================>..] - ETA: 0s - loss: 0.0250 - accuracy: 0.9926
	60000/60000 [==============================] - 5s 92us/sample - loss: 0.0251 - accuracy: 0.9926 - val_loss: 0.0810 - val_accuracy: 0.9771
	Epoch 25/30

	4096/60000 [=>............................] - ETA: 4s - loss: 0.0230 - accuracy: 0.9927
	8192/60000 [===>..........................] - ETA: 4s - loss: 0.0233 - accuracy: 0.9924
	12288/60000 [=====>........................] - ETA: 4s - loss: 0.0217 - accuracy: 0.9932
	16384/60000 [=======>......................] - ETA: 3s - loss: 0.0242 - accuracy: 0.9921
	20480/60000 [=========>....................] - ETA: 3s - loss: 0.0247 - accuracy: 0.9921
	24576/60000 [===========>..................] - ETA: 3s - loss: 0.0248 - accuracy: 0.9923
	28672/60000 [=============>................] - ETA: 2s - loss: 0.0243 - accuracy: 0.9925
	32768/60000 [===============>..............] - ETA: 2s - loss: 0.0239 - accuracy: 0.9927
	36864/60000 [=================>............] - ETA: 2s - loss: 0.0242 - accuracy: 0.9926
	40960/60000 [===================>..........] - ETA: 1s - loss: 0.0247 - accuracy: 0.9925
	45056/60000 [=====================>........] - ETA: 1s - loss: 0.0248 - accuracy: 0.9924
	49152/60000 [=======================>......] - ETA: 0s - loss: 0.0247 - accuracy: 0.9924
	53248/60000 [=========================>....] - ETA: 0s - loss: 0.0247 - accuracy: 0.9924
	57344/60000 [===========================>..] - ETA: 0s - loss: 0.0248 - accuracy: 0.9924
	60000/60000 [==============================] - 6s 93us/sample - loss: 0.0251 - accuracy: 0.9923 - val_loss: 0.0722 - val_accuracy: 0.9796
	Epoch 26/30

	4096/60000 [=>............................] - ETA: 4s - loss: 0.0267 - accuracy: 0.9905
	8192/60000 [===>..........................] - ETA: 4s - loss: 0.0230 - accuracy: 0.9928
	12288/60000 [=====>........................] - ETA: 4s - loss: 0.0225 - accuracy: 0.9932
	16384/60000 [=======>......................] - ETA: 3s - loss: 0.0218 - accuracy: 0.9937
	20480/60000 [=========>....................] - ETA: 3s - loss: 0.0227 - accuracy: 0.9931
	24576/60000 [===========>..................] - ETA: 3s - loss: 0.0221 - accuracy: 0.9932
	28672/60000 [=============>................] - ETA: 2s - loss: 0.0222 - accuracy: 0.9932
	32768/60000 [===============>..............] - ETA: 2s - loss: 0.0224 - accuracy: 0.9932
	36864/60000 [=================>............] - ETA: 2s - loss: 0.0221 - accuracy: 0.9934
	40960/60000 [===================>..........] - ETA: 1s - loss: 0.0219 - accuracy: 0.9934
	45056/60000 [=====================>........] - ETA: 1s - loss: 0.0215 - accuracy: 0.9936
	49152/60000 [=======================>......] - ETA: 0s - loss: 0.0215 - accuracy: 0.9936
	53248/60000 [=========================>....] - ETA: 0s - loss: 0.0218 - accuracy: 0.9935
	57344/60000 [===========================>..] - ETA: 0s - loss: 0.0216 - accuracy: 0.9935
	60000/60000 [==============================] - 6s 94us/sample - loss: 0.0216 - accuracy: 0.9935 - val_loss: 0.0702 - val_accuracy: 0.9805
	Epoch 27/30

	4096/60000 [=>............................] - ETA: 4s - loss: 0.0156 - accuracy: 0.9963
	8192/60000 [===>..........................] - ETA: 4s - loss: 0.0158 - accuracy: 0.9957
	12288/60000 [=====>........................] - ETA: 4s - loss: 0.0181 - accuracy: 0.9950
	16384/60000 [=======>......................] - ETA: 3s - loss: 0.0189 - accuracy: 0.9948
	20480/60000 [=========>....................] - ETA: 3s - loss: 0.0191 - accuracy: 0.9952
	24576/60000 [===========>..................] - ETA: 3s - loss: 0.0198 - accuracy: 0.9946
	28672/60000 [=============>................] - ETA: 2s - loss: 0.0194 - accuracy: 0.9947
	32768/60000 [===============>..............] - ETA: 2s - loss: 0.0190 - accuracy: 0.9948
	36864/60000 [=================>............] - ETA: 1s - loss: 0.0192 - accuracy: 0.9946
	40960/60000 [===================>..........] - ETA: 1s - loss: 0.0189 - accuracy: 0.9946
	45056/60000 [=====================>........] - ETA: 1s - loss: 0.0194 - accuracy: 0.9945
	49152/60000 [=======================>......] - ETA: 0s - loss: 0.0193 - accuracy: 0.9946
	53248/60000 [=========================>....] - ETA: 0s - loss: 0.0195 - accuracy: 0.9945
	57344/60000 [===========================>..] - ETA: 0s - loss: 0.0192 - accuracy: 0.9947
	60000/60000 [==============================] - 5s 91us/sample - loss: 0.0194 - accuracy: 0.9946 - val_loss: 0.0734 - val_accuracy: 0.9796
	Epoch 28/30

	4096/60000 [=>............................] - ETA: 4s - loss: 0.0164 - accuracy: 0.9963
	8192/60000 [===>..........................] - ETA: 4s - loss: 0.0172 - accuracy: 0.9956
	12288/60000 [=====>........................] - ETA: 4s - loss: 0.0170 - accuracy: 0.9960
	16384/60000 [=======>......................] - ETA: 3s - loss: 0.0183 - accuracy: 0.9957
	20480/60000 [=========>....................] - ETA: 3s - loss: 0.0182 - accuracy: 0.9957
	24576/60000 [===========>..................] - ETA: 3s - loss: 0.0179 - accuracy: 0.9956
	28672/60000 [=============>................] - ETA: 2s - loss: 0.0184 - accuracy: 0.9952
	32768/60000 [===============>..............] - ETA: 2s - loss: 0.0181 - accuracy: 0.9953
	36864/60000 [=================>............] - ETA: 1s - loss: 0.0183 - accuracy: 0.9953
	40960/60000 [===================>..........] - ETA: 1s - loss: 0.0185 - accuracy: 0.9951
	45056/60000 [=====================>........] - ETA: 1s - loss: 0.0187 - accuracy: 0.9949
	49152/60000 [=======================>......] - ETA: 0s - loss: 0.0188 - accuracy: 0.9948
	53248/60000 [=========================>....] - ETA: 0s - loss: 0.0188 - accuracy: 0.9947
	57344/60000 [===========================>..] - ETA: 0s - loss: 0.0189 - accuracy: 0.9946
	60000/60000 [==============================] - 6s 93us/sample - loss: 0.0187 - accuracy: 0.9947 - val_loss: 0.0779 - val_accuracy: 0.9781
	Epoch 29/30

	4096/60000 [=>............................] - ETA: 5s - loss: 0.0223 - accuracy: 0.9939
	8192/60000 [===>..........................] - ETA: 4s - loss: 0.0199 - accuracy: 0.9948
	12288/60000 [=====>........................] - ETA: 4s - loss: 0.0203 - accuracy: 0.9944
	16384/60000 [=======>......................] - ETA: 3s - loss: 0.0192 - accuracy: 0.9944
	20480/60000 [=========>....................] - ETA: 3s - loss: 0.0178 - accuracy: 0.9950
	24576/60000 [===========>..................] - ETA: 3s - loss: 0.0178 - accuracy: 0.9948
	28672/60000 [=============>................] - ETA: 2s - loss: 0.0180 - accuracy: 0.9948
	32768/60000 [===============>..............] - ETA: 2s - loss: 0.0175 - accuracy: 0.9949
	36864/60000 [=================>............] - ETA: 2s - loss: 0.0174 - accuracy: 0.9950
	40960/60000 [===================>..........] - ETA: 1s - loss: 0.0169 - accuracy: 0.9952
	45056/60000 [=====================>........] - ETA: 1s - loss: 0.0173 - accuracy: 0.9949
	49152/60000 [=======================>......] - ETA: 0s - loss: 0.0172 - accuracy: 0.9949
	53248/60000 [=========================>....] - ETA: 0s - loss: 0.0170 - accuracy: 0.9950
	57344/60000 [===========================>..] - ETA: 0s - loss: 0.0176 - accuracy: 0.9948
	60000/60000 [==============================] - 6s 92us/sample - loss: 0.0178 - accuracy: 0.9948 - val_loss: 0.0757 - val_accuracy: 0.9790
	Epoch 30/30

	4096/60000 [=>............................] - ETA: 4s - loss: 0.0154 - accuracy: 0.9949
	8192/60000 [===>..........................] - ETA: 4s - loss: 0.0166 - accuracy: 0.9944
	12288/60000 [=====>........................] - ETA: 3s - loss: 0.0163 - accuracy: 0.9948
	16384/60000 [=======>......................] - ETA: 3s - loss: 0.0162 - accuracy: 0.9951
	20480/60000 [=========>....................] - ETA: 3s - loss: 0.0166 - accuracy: 0.9948
	24576/60000 [===========>..................] - ETA: 2s - loss: 0.0165 - accuracy: 0.9948
	28672/60000 [=============>................] - ETA: 2s - loss: 0.0161 - accuracy: 0.9950
	32768/60000 [===============>..............] - ETA: 2s - loss: 0.0159 - accuracy: 0.9952
	36864/60000 [=================>............] - ETA: 1s - loss: 0.0166 - accuracy: 0.9950
	40960/60000 [===================>..........] - ETA: 1s - loss: 0.0170 - accuracy: 0.9949
	45056/60000 [=====================>........] - ETA: 1s - loss: 0.0170 - accuracy: 0.9949
	49152/60000 [=======================>......] - ETA: 0s - loss: 0.0171 - accuracy: 0.9949
	53248/60000 [=========================>....] - ETA: 0s - loss: 0.0169 - accuracy: 0.9949
	57344/60000 [===========================>..] - ETA: 0s - loss: 0.0175 - accuracy: 0.9948
	60000/60000 [==============================] - 5s 92us/sample - loss: 0.0177 - accuracy: 0.9948 - val_loss: 0.0777 - val_accuracy: 0.9792
	loss accuracy val_loss val_accuracy
	27 0.018740 0.994683 0.077904 0.9781
	28 0.017760 0.994750 0.075688 0.9790
	29 0.017704 0.994767 0.077669 0.9792
	(1, 28, 28, 1)
	[<tensorflow.python.keras.layers.convolutional.Conv2D object at 0x000002A710972C08>, <tensorflow.python.keras.layers.pooling.MaxPooling2D object at 0x000002A719A337C8>, <tensorflow.python.keras.layers.core.Flatten object at 0x000002A719A32AC8>, <tensorflow.python.keras.layers.core.Dense object at 0x000002A719A49688>, <tensorflow.python.keras.layers.core.Dense object at 0x000002A719A2F248>]
	[array([[[[-0.82907677, -0.7962916 , 0.38359177]],

	[[ 0.3429307 , -0.27833146, 0.59408164]],

	[[ 0.5247829 , 0.5927083 , 0.2917029 ]]],


	[[[-0.8684553 , -0.32124314, 0.37797683]],

	[[-0.1796955 , 0.10554386, 0.27022994]],

	[[ 0.6602316 , 0.61931485, 0.3056805 ]]],


	[[[-0.12969881, -0.27535865, -0.30478483]],

	[[-0.8268494 , 0.5394653 , 0.34637433]],

	[[ 0.40404692, 0.09111527, 0.43088043]]]], dtype=float32), array([ 0.07662638, 0.02554144, -0.0072965 ], dtype=float32)]
	(26, 26, 3)
	(13, 13, 3)
	(507,)

view raw image_recognition_2.log hosted with ❤ by GitHub

rteak

return to experience and knowledge

Menu

キカガク（KIKAGAKU）で「畳み込みニューラルネットワークの実装」を学ぶ

学習内容

データセットの準備

CNN モデルの定義

CNN モデルの順伝播の流れ

ソースコード

出力結果

実行ログ

Welcome

Popular Posts

Links

OTHERS

note

VIEWS

学習内容

データセットの準備

CNN モデルの定義

CNN モデルの順伝播の流れ

ソースコード

出力結果

実行ログ

関連記事

Welcome

Popular Posts

Links

OTHERS

note

VIEWS