15.2 가장 인기있는 머신러닝과 딥러닝 플랫폼 : 텐서플로우¶
In [1]:
!pip install tensorflow
In [2]:
import tensorflow as tf
print(tf.__version__)
15.3 구글 코래버러토리를Colaboratory를 이용한 텐서플로우 사용¶
In [3]:
# tensorflow¿Í tf.keras¸¦ ÀÓÆ÷Æ®ÇÑ´Ù
import tensorflow as tf
from tensorflow import keras
import numpy as np
import matplotlib.pyplot as plt
# ÆÐ¼Ç MNIST µ¥ÀÌÅÍ´Â kerasÀÇ µ¥ÀÌÅͼ¿¡ Àִµ¥ À̸¦ Àоî¿Í¼ ÇнÀ¿ë, Å×½ºÆ® µ¥ÀÌÅÍ·Î ±¸ºÐ
fashion_mnist = keras.datasets.fashion_mnist
(train_images, train_labels), (test_images, test_labels) = fashion_mnist.load_data()
In [4]:
print(train_images.shape) # ÇнÀ À̹ÌÁöÀÇ ÇüÅÂ¿Í ·¹À̺íÀ» Ãâ·ÂÇÑ´Ù
print(train_labels)
print(test_images.shape)
15.5 패션 MNIST는 레이블이 있는 이미지 데이터이다¶
In [5]:
fig = plt.figure()
ax1 = fig.add_subplot(1, 3, 1)
ax2 = fig.add_subplot(1, 3, 2)
ax3 = fig.add_subplot(1, 3, 3)
ax1.imshow(train_images[0]) # ù ¹ø° ÈÆ·Ã¿ë µ¥ÀÌÅÍ
ax2.imshow(train_images[1]) # µÎ ¹ø° ÈÆ·Ã¿ë µ¥ÀÌÅÍ
ax3.imshow(train_images[2]) # ¼¼ ¹ø° ÈÆ·Ã¿ë µ¥ÀÌÅÍ
plt.show()
In [6]:
print(train_labels[:3])
15.6 딥러닝? 인공 신경망을 구축해 보자¶
In [7]:
model = keras.Sequential([
keras.layers.Flatten(input_shape=(28, 28)),
keras.layers.Dense(128, activation='relu'),
keras.layers.Dense(10, activation='softmax')
])
15.7 인공 신경망을 학습시켜 보자 : 최적화와 에폭¶
In [8]:
model.compile(optimizer='adam',
loss='sparse_categorical_crossentropy',
metrics=['accuracy'])
In [9]:
model.fit(train_images, train_labels, epochs=5)
Out[9]:
In [10]:
test_loss, test_acc = model.evaluate(test_images, test_labels, verbose=2)
print('\nÅ×½ºÆ® Á¤È®µµ:', test_acc)
15.8 학습된 신경망을 새 이미지에 적용해 보자¶
In [11]:
test_images.shape
Out[11]:
In [12]:
import numpy as np
randIdx = np.random.randint(0, 1000)
plt.imshow(test_images[randIdx])
Out[12]:
In [13]:
yhat = model.predict( test_images[randIdx][np.newaxis, :, :])
yhat
Out[13]:
15.9 클래스를 찾아 출력하게 만들기¶
In [14]:
yhat = np.argmax( model.predict( test_images[randIdx][np.newaxis, :, :]) )
yhat
Out[14]:
In [15]:
class_names = ['T-shirt/top', 'Trouser', 'Pullover', 'Dress', 'Coat',
'Sandal', 'Shirt', 'Sneaker', 'Bag', 'Ankle boot']
In [16]:
yhat = np.argmax( model.predict( test_images[randIdx][np.newaxis, :, :]) )
print(class_names[yhat])
15.10 나만의 데이터 이용법 : 드라이브에 올린 파일 접근하기¶
In [17]:
from google.colab import drive
drive.mount('/content/drive')
In [18]:
!pwd
15.11 클라우드에 자료를 코랩에서 사용하기¶
In [19]:
!ls ./drive/'My Drive' -la
15.12 드라이브의 이미지 파일을 읽어 화면에 표시해 보기¶
In [20]:
import matplotlib.image as mpimg
import matplotlib.pyplot as plt
img = mpimg.imread('./drive/My Drive/myData.png')
plt.imshow(img)
Out[20]:
In [21]:
import cv2
img = cv2.imread('./drive/My Drive/myData.png', cv2.IMREAD_GRAYSCALE)
img = cv2.resize(img, (28, 28) )
plt.imshow(img)
Out[21]:
15.13 이미지를 모델에 입력으로 제공해 인식을 시도하자¶
In [22]:
input_data = img[np.newaxis, :, :]
input_data.shape
Out[22]:
In [23]:
class_names = ['T-shirt/top', 'Trouser', 'Pullover', 'Dress', 'Coat',
'Sandal', 'Shirt', 'Sneaker', 'Bag', 'Ankle boot']
yhat = np.argmax( model.predict( input_data ) )
print(class_names[yhat])
In [24]:
input_mirror = input_data[:, :, ::-1]
plt.imshow(input_mirror[0])
Out[24]:
In [25]:
class_names = ['T-shirt/top', 'Trouser', 'Pullover', 'Dress', 'Coat',
'Sandal', 'Shirt', 'Sneaker', 'Bag', 'Ankle boot']
yhat = np.argmax( model.predict( input_mirror ) )
print(class_names[yhat])
15.14 신경망 학습과 편향 : 학습의 한계를 인식하자¶
In [26]:
model2 = keras.Sequential([
keras.layers.Flatten(input_shape=(28, 28)),
keras.layers.Dense(128, activation='relu'),
keras.layers.Dense(64, activation='relu'),
keras.layers.Dense(32, activation='relu'),
keras.layers.Dense(10, activation='softmax')
])
model2.compile(optimizer='adam',
loss='sparse_categorical_crossentropy',
metrics=['accuracy'])
model2.fit(train_images, train_labels, epochs=5)
Out[26]:
15.16 학습된 모델 저장하고 불러와서 사용하기¶
In [27]:
model2.save('./drive/My Drive/myFirstModel.h5')
!ls ./drive/'My Drive' -la
In [28]:
model_imported = keras.models.load_model('./drive/My Drive/myFirstModel.h5')
model_imported.summary()
In [29]:
!ls -al ./drive/'My Drive'
In [30]:
import cv2
img = cv2.imread('./drive/My Drive/bag_cartoon.png', cv2.IMREAD_GRAYSCALE)
plt.imshow(img)
Out[30]:
In [31]:
img = cv2.resize(img, (28, 28) )
plt.imshow(img)
Out[31]:
In [32]:
input_data = img[np.newaxis, :, :]
yhat = np.argmax( model_imported.predict( input_data ) )
print(class_names[yhat])