workSpace/PYTHON
[DL] MNIST analysis with CNN model using keras ( Accuracy 98.94 % )
J o e
2021. 2. 3. 10:23
In [1]:
import numpy as np
from keras import models, layers, datasets, backend
from keras.utils import np_utils
import matplotlib.pyplot as plt
import numpy as np
import keras
In [2]:
num_classes = 10
(X_train, y_train),(X_test, y_test) = datasets.mnist.load_data()
In [3]:
'''
cnn을 다룰때는 1차원으로 값을 펼치지 않고, 이미지를 통으로 넣기 때문에
채널이 중요하다
흑백이미지라도 1이라는 값을 넣어줘야한다.
채널을 표현하는 방법
1) channel First 1,28,28
2) channel Last 28,28,1
여기서는 channel Last로 되어있다... ToTensor를 안 써주므로, 기존 값인 channel로 되어있고,
이것(채널)을 알아내는 방법은 backend로 알아 볼 수 있다.
'''
print(X_train.shape)
print(X_train.shape[1:])
img_rows, img_cols = X_train.shape[1:]
(60000, 28, 28)
(28, 28)
In [4]:
channel = backend.image_data_format()
channel # 이를 통해서 backend로 알아 볼 수 있다.
Out[4]:
'channels_last'전처리¶
CNN에서는 채널을 꼭 넣어줘야 한다. 그리고 채널을
In [5]:
if backend.image_data_format == 'chennel_first':
X_train = X_train.reshape(X_train.shape[0], 1, img_rows, img_cols)
X_test = X_test.reshape(X_test.shape[0], 1, img_rows, img_cols)
input_shape = (1, img_rows, img_cols)
else:
X_train = X_train.reshape(X_train.shape[0], img_rows, img_cols, 1)
X_test = X_test.reshape(X_test.shape[0], img_rows, img_cols, 1)
input_shape = (img_rows, img_cols, 1)
스케일링¶
In [6]:
X_train = X_train/255
X_test = X_test/255
# 아래와 같은 코드인지 확인해보자.
In [7]:
# X_train = X_train.astype('float32')
# X_test = X_test.astype('float32')
# X_train /=255
# X_test /=255
In [8]:
Y_train = keras.utils.to_categorical(y_train, num_classes)
Y_test = keras.utils.to_categorical(y_test, num_classes)
In [20]:
model = keras.Sequential()
model.add(layers.Conv2D(32,
kernel_size=(3,3),
activation = 'relu',
input_shape = input_shape))
model.add(layers.Conv2D(64,
kernel_size=(3,3),
activation = 'relu'))
model.add(layers.MaxPool2D(pool_size=(2,2)))
model.add(layers.Dropout(0.25))
model.add(layers.Flatten()) # 1차원으로 나열해준다.
model.add(layers.Dense(128, activation='relu'))
model.add(layers.Dropout(0.25))
model.add(layers.Dense(num_classes, activation='softmax'))
model.summary() # 표로 전체적인 결과값을 볼수 있다.
Model: "sequential_1"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv2d_2 (Conv2D) (None, 26, 26, 32) 320
_________________________________________________________________
conv2d_3 (Conv2D) (None, 24, 24, 64) 18496
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 12, 12, 64) 0
_________________________________________________________________
dropout_2 (Dropout) (None, 12, 12, 64) 0
_________________________________________________________________
flatten_1 (Flatten) (None, 9216) 0
_________________________________________________________________
dense_2 (Dense) (None, 128) 1179776
_________________________________________________________________
dropout_3 (Dropout) (None, 128) 0
_________________________________________________________________
dense_3 (Dense) (None, 10) 1290
=================================================================
Total params: 1,199,882
Trainable params: 1,199,882
Non-trainable params: 0
_________________________________________________________________
padding의 디폴트 값은 0 이다. padding을 안 써줘서28에서 26으로 줄어든다.
In [10]:
model.compile(loss=keras.losses.categorical_crossentropy, optimizer='rmsprop', metrics=['accuracy']) #rmsprop으로 성능 평가를 해보자
In [11]:
num_batch_size = 100
num_epochs = 50
validation_split = 0.2
In [12]:
history = model.fit(X_train, Y_train, batch_size=num_batch_size, epochs=num_epochs,validation_split=validation_split)
Epoch 1/50
480/480 [==============================] - 6s 8ms/step - loss: 0.4061 - accuracy: 0.8710 - val_loss: 0.0723 - val_accuracy: 0.9787
Epoch 2/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0656 - accuracy: 0.9802 - val_loss: 0.0509 - val_accuracy: 0.9844
Epoch 3/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0437 - accuracy: 0.9872 - val_loss: 0.0451 - val_accuracy: 0.9875
Epoch 4/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0351 - accuracy: 0.9891 - val_loss: 0.0419 - val_accuracy: 0.9883
Epoch 5/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0315 - accuracy: 0.9901 - val_loss: 0.0432 - val_accuracy: 0.9893
Epoch 6/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0225 - accuracy: 0.9924 - val_loss: 0.0429 - val_accuracy: 0.9885
Epoch 7/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0218 - accuracy: 0.9931 - val_loss: 0.0393 - val_accuracy: 0.9892
Epoch 8/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0201 - accuracy: 0.9936 - val_loss: 0.0366 - val_accuracy: 0.9900
Epoch 9/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0191 - accuracy: 0.9936 - val_loss: 0.0402 - val_accuracy: 0.9896
Epoch 10/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0188 - accuracy: 0.9940 - val_loss: 0.0381 - val_accuracy: 0.9906
Epoch 11/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0176 - accuracy: 0.9946 - val_loss: 0.0605 - val_accuracy: 0.9885
Epoch 12/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0190 - accuracy: 0.9944 - val_loss: 0.0439 - val_accuracy: 0.9901
Epoch 13/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0191 - accuracy: 0.9942 - val_loss: 0.0753 - val_accuracy: 0.9873
Epoch 14/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0173 - accuracy: 0.9946 - val_loss: 0.0611 - val_accuracy: 0.9886
Epoch 15/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0155 - accuracy: 0.9954 - val_loss: 0.0387 - val_accuracy: 0.9905
Epoch 16/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0212 - accuracy: 0.9938 - val_loss: 0.0398 - val_accuracy: 0.9892
Epoch 17/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0207 - accuracy: 0.9941 - val_loss: 0.0464 - val_accuracy: 0.9916
Epoch 18/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0201 - accuracy: 0.9944 - val_loss: 0.0712 - val_accuracy: 0.9881
Epoch 19/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0228 - accuracy: 0.9933 - val_loss: 0.0373 - val_accuracy: 0.9912
Epoch 20/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0203 - accuracy: 0.9940 - val_loss: 0.0604 - val_accuracy: 0.9887
Epoch 21/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0202 - accuracy: 0.9939 - val_loss: 0.0573 - val_accuracy: 0.9902
Epoch 22/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0234 - accuracy: 0.9934 - val_loss: 0.0429 - val_accuracy: 0.9919
Epoch 23/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0251 - accuracy: 0.9932 - val_loss: 0.0459 - val_accuracy: 0.9895
Epoch 24/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0235 - accuracy: 0.9937 - val_loss: 0.0669 - val_accuracy: 0.9891
Epoch 25/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0269 - accuracy: 0.9929 - val_loss: 0.0445 - val_accuracy: 0.9909
Epoch 26/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0224 - accuracy: 0.9937 - val_loss: 0.0545 - val_accuracy: 0.9900
Epoch 27/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0236 - accuracy: 0.9940 - val_loss: 0.0555 - val_accuracy: 0.9903
Epoch 28/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0238 - accuracy: 0.9934 - val_loss: 0.0587 - val_accuracy: 0.9908
Epoch 29/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0246 - accuracy: 0.9934 - val_loss: 0.0540 - val_accuracy: 0.9891
Epoch 30/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0257 - accuracy: 0.9935 - val_loss: 0.0454 - val_accuracy: 0.9889
Epoch 31/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0227 - accuracy: 0.9941 - val_loss: 0.0581 - val_accuracy: 0.9885
Epoch 32/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0244 - accuracy: 0.9928 - val_loss: 0.0433 - val_accuracy: 0.9902
Epoch 33/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0223 - accuracy: 0.9937 - val_loss: 0.0429 - val_accuracy: 0.9893
Epoch 34/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0252 - accuracy: 0.9935 - val_loss: 0.0474 - val_accuracy: 0.9901
Epoch 35/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0267 - accuracy: 0.9928 - val_loss: 0.0514 - val_accuracy: 0.9910
Epoch 36/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0295 - accuracy: 0.9933 - val_loss: 0.0665 - val_accuracy: 0.9890
Epoch 37/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0277 - accuracy: 0.9927 - val_loss: 0.0615 - val_accuracy: 0.9890
Epoch 38/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0255 - accuracy: 0.9935 - val_loss: 0.0662 - val_accuracy: 0.9895
Epoch 39/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0293 - accuracy: 0.9919 - val_loss: 0.0583 - val_accuracy: 0.9889
Epoch 40/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0277 - accuracy: 0.9928 - val_loss: 0.0465 - val_accuracy: 0.9887
Epoch 41/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0266 - accuracy: 0.9925 - val_loss: 0.0694 - val_accuracy: 0.9901
Epoch 42/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0252 - accuracy: 0.9929 - val_loss: 0.0440 - val_accuracy: 0.9889
Epoch 43/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0285 - accuracy: 0.9922 - val_loss: 0.0604 - val_accuracy: 0.9902
Epoch 44/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0256 - accuracy: 0.9930 - val_loss: 0.0827 - val_accuracy: 0.9895
Epoch 45/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0270 - accuracy: 0.9928 - val_loss: 0.0618 - val_accuracy: 0.9886
Epoch 46/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0274 - accuracy: 0.9936 - val_loss: 0.0584 - val_accuracy: 0.9902
Epoch 47/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0320 - accuracy: 0.9916 - val_loss: 0.0571 - val_accuracy: 0.9886
Epoch 48/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0292 - accuracy: 0.9927 - val_loss: 0.0999 - val_accuracy: 0.9898
Epoch 49/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0324 - accuracy: 0.9923 - val_loss: 0.0618 - val_accuracy: 0.9905
Epoch 50/50
480/480 [==============================] - 3s 7ms/step - loss: 0.0309 - accuracy: 0.9919 - val_loss: 0.0621 - val_accuracy: 0.9907
In [25]:
# loss 관련된 plot 함수
def plot_loss(history):
plt.plot(history.history['loss'])
plt.plot(history.history['val_loss'])
plt.title('model Loss')
plt.ylabel('Loss')
plt.xlabel('Epoch') # X 측이 늘어남에 따라서 Y축이 얼마나 떨어지는가가 궁금한것이다.
plt.legend(['Train','Validation'])
plt.savefig('Loss_Plot.png', dpi=300) # dpi해상도를 조절 할 수 있다.
plt.show()
def plot_acc(history):
plt.plot(history.history['accuracy'])
plt.plot(history.history['val_accuracy'])
plt.title('Model Accuracy')
plt.ylabel('Accuracy')
plt.xlabel('Epoch')
plt.legend(['Train','Validation'])
plt.savefig('Acc_Plot.png', dpi=300) # dpi해상도를 조절 할 수 있다.
plt.show()
In [26]:
plot_loss(history)
In [27]:
plot_acc(history)
Model Evaluation¶
In [19]:
model_performance = model.evaluate(X_test,Y_test,batch_size=num_batch_size)
print(f'loss:{model_performance[0]}, accuracy:{model_performance[1]}')
print(*model_performance)
100/100 [==============================] - 0s 3ms/step - loss: 0.0517 - accuracy: 0.9895
loss:0.05172428488731384, accuracy:0.9894999861717224
0.05172428488731384 0.9894999861717224
In [ ]: