[TIL]21.07.26 CNN

cnn 첫날 이었다 솔직히 정말 오랜만에 이해하기 어려웠다. 보통 이해는 하고 코드짜는게 바로 아이디어로 떠오르지 않는다 였는데 이번꺼는 개념부터가 어렵게 다가왔다.

 

from tensorflow.keras.datasets import fashion_mnist

from tensorflow.keras.models import Sequential

from tensorflow.keras.layers import Dense

from tensorflow.keras.callbacks import ModelCheckpoint , EarlyStopping

import sys

import matplotlib.pyplot as plt

import numpy as np

import tensorflow as tf

(X_train, Y_class_train), (X_test, Y_class_test) = fashion_mnist.load_data()

#seed 값 설정

seed=0

np.random.seed(seed)

tf.random.set_seed(seed)

print('학습셋 이미지 수: %d 개'%(X_train.shape[0]))

print('테스트셋 이미지 수: %d 개'%(X_test.shape[0]))

 

# #그래프로 확인하기

# import matplotlib.pyplot as plt

# plt.imshow(X_train[0],cmap='Greys')

# plt.show()

 

# #코드로 확인

# for x in X_train[0]:

#   for i in x:

#     # sys.stdout.write('%d\t'%i)

#     sys.stdout.write('%3d'%i)

#   sys.stdout.write('\n')

#차원 변환 과정

X_train = X_train.reshape(X_train.shape[0],784).astype('float64')/255

 

X_test=X_test.reshape(X_test.shape[0],784).astype('float64')/255

 

#class 5 를 [0,0,0,0,1,0,0,0,0,0]로 바꿔야함 바이너리화 과정

import tensorflow as tf

Y_train=tf.keras.utils.to_categorical(Y_class_train,10)

Y_test=tf.keras.utils.to_categorical(Y_class_test,10)

model= Sequential()

model.add(Dense(512, input_dim=784,activation='relu'))

model.add(Dense(10,activation = 'softmax'))

model.compile(loss='categorical_crossentropy',optimizer='adam',metrics=['accuracy'])

 

import os

MODEL_DIR='./model/'

if not os.path.exists(MODEL_DIR):

  os.mkdir(MODEL_DIR)

 

modelpath='./model/{epoch:02d}-{val_loss:.4f}.hdf5'

Checkpointer= ModelCheckpoint(filepath=modelpath, monitor='val_loss', verbose=0,save_best_only=True)

#자동 중단 설정

early_stopping_callback=EarlyStopping(monitor='val_loss',patience=10)

history=model.fit(X_train, Y_train , validation_data=(X_test,Y_test), epochs=30,

                  batch_size=200,verbose=0, callbacks=[early_stopping_callback,Checkpointer])

 

print('\n Test Accuracy : %.4f'%(model.evaluate(X_test, Y_test)[1]))

 

y_vloss=history.history['val_loss']

y_loss=history.history['loss']

 

 

 

#x값을 지정하고 정확도는 파랑 오차를 빨강 으로 표시

 

x_len=np.arange(len(y_loss))

plt.plot(x_len, y_vloss,marker='.',c='red',label='Testset_loss')

plt.plot(x_len,y_loss,'o',c='blue',label='Trainset_loss')

 

plt.legend()

plt.grid()

plt.xlabel('epoch')

plt.ylabel('loss')

plt.show()

import tensorflow as tf

from tensorflow import keras

import numpy as np

import matplotlib.pyplot as plt

 

#image의 전처리가 완료되어있음

fashion_mnist = keras.datasets.fashion_mnist

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

 

class_names = ['T-shirts','Trouser','Pullover','Dress','Coat','Sandal','Shirt','Sneaker','Bag','Ankle boot']

 

print(train_images.shape)

print(train_labels)

print(test_images.shape)

print(len(test_labels))

#훈련 셋트에 있는 첫번째 이미지- 픽셀 값의 범위가 0~255 사이 확인

plt.figure()

plt.imshow(train_images[0])

plt.colorbar()

plt.grid(False)

plt.show()

 

#0~1사이로 만들어줌

train_images=train_images /255.0

test_images=test_images /255.0

 

plt.figure(figsize=(10,10))

for i in range(25):

  plt.subplot(5,5,i+1)

  plt.xticks([])

  plt.yticks([])

  plt.grid(False)

  plt.imshow(train_images[i],cmap=plt.cm.binary)

  plt.xlabel(class_names[train_labels[i]])

plt.show()

 

from tensorflow.keras.models import Sequential

from tensorflow.keras.layers import Dense

from tensorflow.keras.layers import Flatten

from tensorflow.keras.callbacks import ModelCheckpoint, EarlyStopping

 

model= Sequential()

#2차원 배열(28x28)의 이미지를 28*28= 784 픽셀의 1차원 배열로 변환

model.add(Flatten(input_shape=(28,28)))

model.add(Dense(128,activation='relu'))

model.add(Dense(10,activation='softmax'))

 

#sparse_categorical_crossentropy 총합이 1이 되도록 분류

model.compile(optimizer='adam',loss='sparse_categorical_crossentropy',metrics=['accuracy'])

 

model.fit(train_images,train_labels,epochs=5)

test_loss,test_acc=model.evaluate(test_images, test_labels,verbose=2)

 

print('\n 테스트 정확도 :',test_acc)

#훈련된 모델을 사용하여 이미지에 대한예측을 생성

 

predictions=model.predict(test_images)

#예측은 10개의 숫자 배열로 푯됨 10개의 옷 품목에 상응하는 모델의신뢰도

print(predictions[0])

np.argmax(predictions[0])

print(test_labels[0])

#10개 클래스에 대한 예측을 모두 그래프로 표현

def plot_image(i,predictions_array,true_label,img):

  predictions_array,true_label,img= predictions_array[i], true_label[i],img[i]

  plt.grid(False)

  plt.xticks([])

  plt.yticks([])

  plt.imshow(img,cmap=plt.cm.binary)

 

  predicted_label=np.argmax(predictions_array)

  if predicted_label==true_label:

    color='blue'

  else:

    color='red'

 

  plt.xlabel('{}{:2.0f}%({})'.format(class_names[predicted_label],100*np.max(predictions_array),

                                     class_names[true_label]),color=color)

  

  #predictions_array 10개의 실수 결과 값/ true_label 분류번호

def plot_value_array(i,predictions_array,true_label):

  predictions_array,true_label=predictions_array[i],true_label[i]

  plt.grid(False)

  plt.xticks([])

  plt.yticks([])

  thisplot=plt.bar(range(10),predictions_array,color="#777777")

  plt.ylim([0,1])

  predicted_label=np.argmax(predictions_array)

 

  thisplot[predicted_label].set_color('red')

  thisplot[true_label].set_color('blue')

 

i=0

plt.figure(figsize=(6,3))

plt.subplot(1,2,1)

#i,predictions,test_labels,test_images)의 0번쨰 index값을 넘겨줌(0해당 이미지로 표시)

plot_image(i,predictions,test_labels,test_images)

plt.subplot(1,2,2)

#값을 bar로 표시

plot_value_array(i,predictions,test_labels)

plt.show()

 

i=12

plt.figure(figsize=(6,3))

plt.subplot(1,2,1)

plot_image(i,predictions,test_labels,test_images)

plt.subplot(1,2,2)

plot_value_array(i,predictions,test_labels)

plt.show()

 

#처음 X개의 테스트 이미지와 예측 레이블,진짜 레이블을 출력합니다

#올바른 예측은 파랑색으로 잘못된 예측은 빨강색으로 나타냅니다.

num_rows=5

num_cols=3

num_images=num_rows*num_cols

plt.figure(figsize=(2*2*num_cols,2*num_rows))

for i in range(num_images):

  plt.subplot(num_rows,2*num_cols,2*i+1)

  plot_image(i,predictions,test_labels,test_images)

  plt.subplot(num_rows,2*num_cols,2*i+2)

  plot_value_array(i,predictions,test_labels)

plt.show()

    

 

 

 

#테스트 세트에서 이미지 하나를 선택합니다

img=test_images[0]

print(img.shape) #(28,28)

#이미지 하나만사용할 때도 배치에 추가- 학습이 3차원 이었으므로 하나만 predict할떄도 3차원으로

img=(np.expand_dims(img,0)) #np.expand_dims(배열,축)을 통해 지정된 축의 차원을 확장

print(img.shape) # (1,28,28)

 

predictions_single=model.predict(img)

 

print(predictions_single)

 

#barplot으로 표시하고 아래 (x축)에 class_name 표시

plot_value_array(0,predictions_single,test_labels)

_=plt.xticks(range(10),class_names,rotation=45)

 

#모델의 예측값 확인

np.argmax(predictions_single[0])

 

 

 

from tensorflow.keras.datasets import mnist

from tensorflow.keras.models import Sequential

from tensorflow.keras.layers import Dense, Dropout, Flatten, Conv2D, MaxPooling2D

from tensorflow.keras.callbacks import ModelCheckpoint , EarlyStopping

 

import os

import matplotlib.pyplot as plt

import numpy as np

import tensorflow as tf

seed= 0

np.random.seed(seed)

tf.random.set_seed(seed)

 

(X_train, Y_train), (X_test, Y_test) = mnist.load_data()

X_train=X_train.reshape(X_train.shape[0],28,28,1).astype('float32')/255

X_test=X_test.reshape(X_test.shape[0],28,28,1).astype('float32')/255

Y_train=tf.keras.utils.to_categorical(Y_train)

Y_test=tf.keras.utils.to_categorical(Y_test)

model=Sequential()

model.add(Conv2D(32,kernel_size=(3,3),input_shape=(28,28,1),activation='relu'))

model.add(Conv2D(64,(3,3),activation='relu'))

model.add(MaxPooling2D(pool_size=2))

model.add(Dropout(0.25))

model.add(Flatten())

model.add(Dense(128,activation='relu'))

model.add(Dropout(0.5))

model.add(Dense(10,activation='softmax'))

model.compile(loss='categorical_crossentropy',optimizer='adam',metrics=['accuracy'])

 

MODEL_DIR='.model/'

if not os.path.exists(MODEL_DIR):

  os.mkdir(MODEL_DIR)

 

modelpath="./model/{epoch:02d}{val_loss:.4f}.hdf5"

 

Checkpointer= ModelCheckpoint(filepath=modelpath, monitor='val_loss', verbose=1,save_best_only=True)

 

early_stopping_callback=EarlyStopping(monitor='val_loss',patience=10)

 

history=model.fit(X_train, Y_train , validation_data=(X_test,Y_test), epochs=30,

                  batch_size=200,verbose=0, callbacks=[early_stopping_callback,Checkpointer])

 

print('\n Test Accuracy : %.4f'%(model.evaluate(X_test, Y_test)[1]))

 

y_vloss=history.history['val_loss']

y_loss=history.history['loss']

 

x_len=np.arange(len(y_loss))

plt.plot(x_len, y_vloss,marker='.',c='red',label='Testset_loss')

plt.plot(x_len,y_loss,'o',c='blue',label='Trainset_loss')

 

plt.legend(loc='upper right')

plt.grid()

plt.xlabel('epoch')

plt.ylabel('loss')

plt.show()

 

Epoch 00020: val_loss did not improve from 0.02511 313/313 [==============================] - 6s 18ms/step - loss: 0.0353 - accuracy: 0.9915 Test Accuracy : 0.9915

 

 

import tensorflow as tf

import sys

import os

import matplotlib.pyplot as plt

import numpy as np

import cv2

from tensorflow.keras.models import load_model

 

model=load_model('/content/model/100.0251.hdf5')

 

model.summary()

 

n=5

 

test_num=[[0]*n for _ in range(n)]

test_num[0]=plt.imread('/content/drive/MyDrive/Colab Notebooks/dataset/0.jpg')

test_num[1]=plt.imread('/content/drive/MyDrive/Colab Notebooks/dataset/2.jpg')

test_num[2]=plt.imread('/content/drive/MyDrive/Colab Notebooks/dataset/3.jpg')

test_num[3]=plt.imread('/content/drive/MyDrive/Colab Notebooks/dataset/6.jpg')

for i in range(4):

  test_num[i]=cv2.cvtColor(test_num[i],cv2.COLOR_BGR2GRAY)

  test_num[i]=cv2.bitwise_not(test_num[i])

  plt.imshow(test_num[i],cmap='Greys')

  plt.show()

  test_num[i]=test_num[i].reshape(1,28,28,1).astype('float64') / 255

  print('The Answer',i,'is',model.predict_classes(test_num[i]))

 

 

 

import tensorflow as tf

import sys

import os

import matplotlib.pyplot as plt

import numpy as np

import cv2

from tensorflow.keras.models import load_model

from tensorflow.keras.datasets import mnist

 

(X_train, Y_train), (X_test, Y_test) = mnist.load_data()

X_train=X_train.reshape(X_train.shape[0],28,28,1).astype('float32')/255

X_test=X_test.reshape(X_test.shape[0],28,28,1).astype('float32')/255

 

model=load_model('/content/model/100.0251.hdf5')

model.summary()

wrong_result=[]

predicted_result= model.predict(X_test)

predicted_labels =np.argmax(predicted_result, axis=1)

 

for n in range(0,len(Y_test)):

  if predicted_labels[n] !=Y_test[n]:

    wrong_result.append(n)

len(wrong_result)

 

 

plt.figure(figsize=(14,12))

 

for idx, n in enumerate(wrong_result):

  plt.subplot(len(wrong_result)/4+1,4,idx+1)

  plt.subplots_adjust(left=0.0,bottom=0.0,right=1.5,top=1.52)

  plt.imshow(X_test[n].reshape(28,28),cmap='Greys',interpolation='nearest')

  plt.title('label:'+str(Y_test[n])+'predict :'+str(predicted_labels[n]))

  plt.axis('off')

plt.show()

from tensorflow.keras.datasets import mnist

from tensorflow.keras.models import Sequential

from tensorflow.keras.layers import Dense, Dropout, Flatten, Conv2D, MaxPooling2D

from tensorflow.keras.callbacks import ModelCheckpoint , EarlyStopping

from tensorflow import keras

import os

import matplotlib.pyplot as plt

import numpy as np

import tensorflow as tf

seed= 0

np.random.seed(seed)

tf.random.set_seed(seed)

fashion_mnist = keras.datasets.fashion_mnist

(X_train,Y_train),(X_test,Y_test) = fashion_mnist.load_data()

 

class_names = ['T-shirts','Trouser','Pullover','Dress','Coat','Sandal','Shirt','Sneaker','Bag','Ankle boot']

X_train=X_train.reshape(X_train.shape[0],28,28,1).astype('float32')/255

X_test=X_test.reshape(X_test.shape[0],28,28,1).astype('float32')/255

Y_train=tf.keras.utils.to_categorical(Y_train)

Y_test=tf.keras.utils.to_categorical(Y_test)

model=Sequential()

model.add(Conv2D(32,kernel_size=(3,3),input_shape=(28,28,1),activation='relu'))

model.add(Conv2D(64,(3,3),activation='relu'))

model.add(MaxPooling2D(pool_size=2))

model.add(Dropout(0.25))

model.add(Flatten())

model.add(Dense(128,activation='relu'))

model.add(Dropout(0.5))

model.add(Dense(10,activation='softmax'))

model.compile(loss='categorical_crossentropy',optimizer='adam',metrics=['accuracy'])

 

MODEL_DIR='.model/'

if not os.path.exists(MODEL_DIR):

  os.mkdir(MODEL_DIR)

 

modelpath="./model/{epoch:02d}{val_loss:.4f}.hdf5"

 

Checkpointer= ModelCheckpoint(filepath=modelpath, monitor='val_loss', verbose=1,save_best_only=True)

 

early_stopping_callback=EarlyStopping(monitor='val_loss',patience=10)

 

history=model.fit(X_train, Y_train , validation_data=(X_test,Y_test), epochs=30,

                  batch_size=200,verbose=0, callbacks=[early_stopping_callback,Checkpointer])

 

print('\n Test Accuracy : %.4f'%(model.evaluate(X_test, Y_test)[1]))

 

y_vloss=history.history['val_loss']

y_loss=history.history['loss']

 

x_len=np.arange(len(y_loss))

plt.plot(x_len, y_vloss,marker='.',c='red',label='Testset_loss')

plt.plot(x_len,y_loss,'o',c='blue',label='Trainset_loss')

 

plt.legend(loc='upper right')

plt.grid()

plt.xlabel('epoch')

plt.ylabel('loss')

plt.show()

Epoch 00023: val_loss did not improve from 0.21461 313/313 [==============================] - 7s 22ms/step - loss: 0.2480 - accuracy: 0.9322 Test Accuracy : 0.9322

import tensorflow as tf

from tensorflow import keras

import numpy as np

import matplotlib.pyplot as plt

 

#image의 전처리가 완료되어있음

fashion_mnist = keras.datasets.fashion_mnist

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

 

class_names = ['T-shirts','Trouser','Pullover','Dress','Coat','Sandal','Shirt','Sneaker','Bag','Ankle boot']

 

print(train_images.shape)

print(train_labels)

print(test_images.shape)

print(len(test_labels))

#훈련 셋트에 있는 첫번째 이미지- 픽셀 값의 범위가 0~255 사이 확인

plt.figure()

plt.imshow(train_images[0])

plt.colorbar()

plt.grid(False)

plt.show()

 

#0~1사이로 만들어줌

train_images=train_images /255.0

test_images=test_images /255.0

 

plt.figure(figsize=(10,10))

for i in range(25):

  plt.subplot(5,5,i+1)

  plt.xticks([])

  plt.yticks([])

  plt.grid(False)

  plt.imshow(train_images[i],cmap=plt.cm.binary)

  plt.xlabel(class_names[train_labels[i]])

plt.show()

 

X_train=X_train.reshape(X_train.shape[0],28,28,1).astype('float32')/255

X_test=X_test.reshape(X_test.shape[0],28,28,1).astype('float32')/255

 

from tensorflow.keras.models import Sequential

from tensorflow.keras.layers import Dense

from tensorflow.keras.layers import Flatten

from tensorflow.keras.callbacks import ModelCheckpoint, EarlyStopping

 

model=Sequential()

model.add(Conv2D(32,kernel_size=(3,3),input_shape=(28,28,1),activation='relu'))

model.add(Conv2D(64,(3,3),activation='relu'))

model.add(MaxPooling2D(pool_size=2))

model.add(Dropout(0.25))

model.add(Flatten())

model.add(Dense(128,activation='relu'))

model.add(Dropout(0.5))

model.add(Dense(10,activation='softmax'))

 

#sparse_categorical_crossentropy 총합이 1이 되도록 분류

model.compile(optimizer='adam',loss='sparse_categorical_crossentropy',metrics=['accuracy'])

 

model.fit(train_images,train_labels,epochs=5)

test_loss,test_acc=model.evaluate(test_images, test_labels,verbose=2)

 

print('\n 테스트 정확도 :',test_acc)

#훈련된 모델을 사용하여 이미지에 대한예측을 생성

 

predictions=model.predict(test_images)

#예측은 10개의 숫자 배열로 푯됨 10개의 옷 품목에 상응하는 모델의신뢰도

print(predictions[0])

np.argmax(predictions[0])

print(test_labels[0])