CIFAR_10

0. Cifar10 Dataset¶

CIFAR-10 데이터 세트는 10개 클래스에 60000 32x32 컬러 이미지로 구성되며 클래스당 6000개의 이미지가 있다. 50000개의 교육 이미지와 10000개의 테스트 이미지가 있습니다.

데이터 세트는 각각 10000개의 이미지가 있는 5개의 교육 배치와 1개의 테스트 배치로 나뉜다. 테스트 배치에는 각 클래스에서 무작위로 선택한 1000개의 이미지가 들어 있습니다. 교육 배치에는 나머지 이미지가 임의의 순서로 포함되지만 일부 교육 배치에는 한 클래스의 이미지가 다른 클래스보다 더 많이 포함될 수 있습니다. 이 중 교육 배치에는 각 클래스의 5,000개 이미지가 정확히 포함됩니다.

1. 데이터 시각화¶

In [1]:

import numpy as np

In [2]:

from tensorflow.keras.datasets import cifar10

In [3]:

(x_train,y_train),(x_test,y_test) = cifar10.load_data()
print(x_train.shape,x_test.shape)
print(y_train.shape,y_test.shape)

(50000, 32, 32, 3) (10000, 32, 32, 3)
(50000, 1) (10000, 1)

In [4]:

import  matplotlib.pyplot as plt

In [23]:

plt.figure(figsize=(5,5))
for i in range(1,16):
    plt.subplot(3,5,i)
    plt.imshow(x_train[i])

In [6]:

x_train = x_train/255.
x_test = x_test/255.

In [7]:

from tensorflow.keras.utils import to_categorical

In [8]:

y_train = to_categorical(y_train)
y_test = to_categorical(y_test)

2. model 구성¶

In [9]:

from tensorflow.keras.layers import Conv2D,BatchNormalization,Flatten,Dense
from tensorflow.keras.models import Sequential

In [10]:

model = Sequential()
model.add(Conv2D(32,(3,3),activation='relu',input_shape=(32,32,3)))
model.add(BatchNormalization())
model.add(Conv2D(64,(3,3),activation='relu'))
model.add(BatchNormalization())
model.add(Conv2D(64,(3,3),activation='relu',strides=2))
model.add(BatchNormalization())

model.add(Conv2D(32,(3,3),activation='relu',))
model.add(BatchNormalization())
model.add(Conv2D(64,(3,3),activation='relu'))
model.add(BatchNormalization())
model.add(Conv2D(64,(3,3),activation='relu',strides=2))
model.add(BatchNormalization())
model.add(Flatten())
model.add(Dense(32,activation='relu'))
model.add(Dense(10,activation='softmax'))

In [11]:

from tensorflow.keras.callbacks import EarlyStopping,ModelCheckpoint,ReduceLROnPlateau

In [12]:

early_stopping = EarlyStopping(monitor="val_loss",patience=16)
reducelr = ReduceLROnPlateau(monitor = "val_loss",patience = 5)

In [13]:

model.compile(loss="categorical_crossentropy",optimizer="adam",metrics=["accuracy"])
history = model.fit(x_train,y_train,epochs =100 , validation_data=(x_test,y_test),batch_size = 32 , callbacks = [early_stopping,reducelr])

Epoch 1/100
1563/1563 [==============================] - 13s 6ms/step - loss: 1.4881 - accuracy: 0.4714 - val_loss: 1.2884 - val_accuracy: 0.5460
Epoch 2/100
1563/1563 [==============================] - 9s 6ms/step - loss: 1.0333 - accuracy: 0.6336 - val_loss: 1.1228 - val_accuracy: 0.5991
Epoch 3/100
1563/1563 [==============================] - 10s 6ms/step - loss: 0.8313 - accuracy: 0.7083 - val_loss: 0.9835 - val_accuracy: 0.6657
Epoch 4/100
1563/1563 [==============================] - 10s 6ms/step - loss: 0.7144 - accuracy: 0.7478 - val_loss: 0.8142 - val_accuracy: 0.7186
Epoch 5/100
1563/1563 [==============================] - 10s 6ms/step - loss: 0.6306 - accuracy: 0.7800 - val_loss: 0.7757 - val_accuracy: 0.7348
Epoch 6/100
1563/1563 [==============================] - 10s 6ms/step - loss: 0.5576 - accuracy: 0.8041 - val_loss: 0.7951 - val_accuracy: 0.7365
Epoch 7/100
1563/1563 [==============================] - 10s 6ms/step - loss: 0.5014 - accuracy: 0.8249 - val_loss: 0.7485 - val_accuracy: 0.7559
Epoch 8/100
1563/1563 [==============================] - 10s 6ms/step - loss: 0.4508 - accuracy: 0.8416 - val_loss: 0.8120 - val_accuracy: 0.7379
Epoch 9/100
1563/1563 [==============================] - 10s 6ms/step - loss: 0.3957 - accuracy: 0.8613 - val_loss: 0.7701 - val_accuracy: 0.7571
Epoch 10/100
1563/1563 [==============================] - 10s 6ms/step - loss: 0.3539 - accuracy: 0.8750 - val_loss: 0.9725 - val_accuracy: 0.7157
Epoch 11/100
1563/1563 [==============================] - 10s 6ms/step - loss: 0.3165 - accuracy: 0.8899 - val_loss: 1.0108 - val_accuracy: 0.7235
Epoch 12/100
1563/1563 [==============================] - 9s 6ms/step - loss: 0.2892 - accuracy: 0.8970 - val_loss: 0.8585 - val_accuracy: 0.7571
Epoch 13/100
1563/1563 [==============================] - 10s 6ms/step - loss: 0.1628 - accuracy: 0.9464 - val_loss: 0.7686 - val_accuracy: 0.7864
Epoch 14/100
1563/1563 [==============================] - 10s 6ms/step - loss: 0.1214 - accuracy: 0.9619 - val_loss: 0.7907 - val_accuracy: 0.7857
Epoch 15/100
1563/1563 [==============================] - 10s 6ms/step - loss: 0.1026 - accuracy: 0.9689 - val_loss: 0.8164 - val_accuracy: 0.7832
Epoch 16/100
1563/1563 [==============================] - 10s 6ms/step - loss: 0.0869 - accuracy: 0.9751 - val_loss: 0.8483 - val_accuracy: 0.7852
Epoch 17/100
1563/1563 [==============================] - 10s 6ms/step - loss: 0.0775 - accuracy: 0.9783 - val_loss: 0.8746 - val_accuracy: 0.7825
Epoch 18/100
1563/1563 [==============================] - 10s 6ms/step - loss: 0.0642 - accuracy: 0.9838 - val_loss: 0.8789 - val_accuracy: 0.7821
Epoch 19/100
1563/1563 [==============================] - 10s 6ms/step - loss: 0.0599 - accuracy: 0.9849 - val_loss: 0.8827 - val_accuracy: 0.7835
Epoch 20/100
1563/1563 [==============================] - 10s 6ms/step - loss: 0.0592 - accuracy: 0.9853 - val_loss: 0.8845 - val_accuracy: 0.7846
Epoch 21/100
1563/1563 [==============================] - 10s 6ms/step - loss: 0.0577 - accuracy: 0.9863 - val_loss: 0.8875 - val_accuracy: 0.7842
Epoch 22/100
1563/1563 [==============================] - 10s 6ms/step - loss: 0.0560 - accuracy: 0.9863 - val_loss: 0.8877 - val_accuracy: 0.7842
Epoch 23/100
1563/1563 [==============================] - 10s 6ms/step - loss: 0.0552 - accuracy: 0.9868 - val_loss: 0.8897 - val_accuracy: 0.7843

In [14]:

loss = history.history["loss"]
acc = history.history["accuracy"]
val_loss = history.history["val_loss"]
val_acc = history.history["val_accuracy"]

plt.plot(range(len(loss)),loss,label = "Loss")
plt.plot(range(len(val_loss)),val_loss,label = "val_Loss")
plt.grid()
plt.legend()
plt.show()

In [15]:

plt.plot(range(len(acc)),acc,label = "acc")
plt.plot(range(len(val_acc)),val_acc,label = "val_acc")
plt.grid()
plt.legend()
plt.show()

In [16]:

y_predict = model.predict(x_test)

In [17]:

y_predict = np.argmax(y_predict,axis=-1)

In [18]:

y_test = np.argmax(y_test,axis=-1)

In [19]:

for i in range(0,10000,200):
    plt.imshow(x_test[i])
    plt.show()
    print("y_predict : {} \t y_true : {}".format(y_predict[i],y_test[i]))

y_predict : 3 	 y_true : 3

y_predict : 5 	 y_true : 5

y_predict : 9 	 y_true : 9

y_predict : 8 	 y_true : 8

y_predict : 7 	 y_true : 7

y_predict : 5 	 y_true : 5

y_predict : 8 	 y_true : 8

y_predict : 5 	 y_true : 5

y_predict : 8 	 y_true : 8

y_predict : 0 	 y_true : 4

y_predict : 6 	 y_true : 1

y_predict : 0 	 y_true : 0

y_predict : 0 	 y_true : 0

y_predict : 8 	 y_true : 8

y_predict : 4 	 y_true : 4

y_predict : 3 	 y_true : 5

y_predict : 5 	 y_true : 5

y_predict : 1 	 y_true : 6

y_predict : 3 	 y_true : 4

y_predict : 9 	 y_true : 9

y_predict : 0 	 y_true : 8

y_predict : 4 	 y_true : 4

y_predict : 3 	 y_true : 3

y_predict : 4 	 y_true : 4

y_predict : 9 	 y_true : 9

y_predict : 7 	 y_true : 7

y_predict : 3 	 y_true : 3

y_predict : 9 	 y_true : 9

y_predict : 6 	 y_true : 6

y_predict : 4 	 y_true : 2

y_predict : 8 	 y_true : 8

y_predict : 8 	 y_true : 3

y_predict : 0 	 y_true : 0

y_predict : 7 	 y_true : 7

y_predict : 6 	 y_true : 6

y_predict : 0 	 y_true : 2

y_predict : 4 	 y_true : 4

y_predict : 3 	 y_true : 3

y_predict : 1 	 y_true : 8

y_predict : 0 	 y_true : 0

y_predict : 9 	 y_true : 9

y_predict : 3 	 y_true : 3

y_predict : 0 	 y_true : 0

y_predict : 3 	 y_true : 3

y_predict : 0 	 y_true : 0

y_predict : 8 	 y_true : 8

y_predict : 8 	 y_true : 8

y_predict : 6 	 y_true : 6

y_predict : 8 	 y_true : 8

y_predict : 1 	 y_true : 1

In [ ]: