keras-6 cnn

1. python

from keras import models
from keras import layers

model = models.Sequential()
model.add(layers.Conv2D(32,(3,3), activation='relu', input_shape=(28,28,1)))
model.add(layers.MaxPooling2D((2,2)))
model.add(layers.Conv2D(32,(3,3), activation='relu'))
model.add(layers.MaxPooling2D((2,2)))
model.add(layers.Conv2D(32,(3,3), activation='relu'))
model.add(layers.MaxPooling2D((2,2)))

sequential 클래스를 사용하여 신경망을 쌓는다. conv2d층과 maxpooling 층을 차례로 쌓는다.

model.add(layers.Flatten())
model.add(layers.Dense(64, activation='relu'))
model.add(layers.Dense(10, activation='softmax'))

conv2d 층은 3d텐서를 출력하기때문에 flatten을 통해서 1차원으로 펼쳐서 maxpooling으로 이동한다. 

from keras.datasets import mnist
from tensorflow.keras.utils import to_categorical

(train_images, train_labels),(test_images, test_labels) = mnist.load_data()

# scaling
train_data = train_images.reshape((60000,28,28,1))
train_data = train_data.astype('float32') / 255

test_data = test_images.reshape((10000,28,28,1))
test_data = test_data.astype('float32') / 255

train_target = to_categorical(train_labels)
test_target = to_categorical(test_labels)

mnist 데이터를 불러오고 데이터를 분류한다.

model.compile(optimizer='rmsprop',
              loss='categorical_crossentropy',
              metrics=['accuracy'])
model.fit(train_data, train_target, epochs=5, batch_size=64)

Epoch 1/5
938/938 [==============================] - 16s 5ms/step - loss: 0.3730 - accuracy: 0.8853
Epoch 2/5
938/938 [==============================] - 4s 4ms/step - loss: 0.1240 - accuracy: 0.9618
Epoch 3/5
938/938 [==============================] - 4s 4ms/step - loss: 0.0861 - accuracy: 0.9733
Epoch 4/5
938/938 [==============================] - 4s 4ms/step - loss: 0.0688 - accuracy: 0.9784
Epoch 5/5
938/938 [==============================] - 4s 4ms/step - loss: 0.0548 - accuracy: 0.9827
<keras.callbacks.History at 0x7fef8d2393c0>

학습을 진행한다.

test_loss, test_acc = model.evaluate(test_data, test_target)

313/313 [==============================] - 1s 3ms/step - loss: 0.0693 - accuracy: 0.9801

모델을 평가한다.

test_acc

0.9800999760627747

정확도는 98프로이다.