[핸즈온 머신러닝 2판] Chapter 3장 연습문제 풀이

1. MNIST 데이터셋으로 분류기를 만들어 테스트 세트에서 97% 정확도를 달성해보세요.

from sklearn.datasets import fetch_openml
import numpy as np

mnist = fetch_openml('mnist_784', version = 1)

X, y = mnist["data"], mnist["target"]
y = y.astype(np.int)

#Train, Test set 나누기
X_train, X_test, y_train, y_test = X[:60000], X[60000:], y[:60000], y[60000:]

from sklearn.neighbors import KNeighborsClassifier

knn_clf = KNeighborsClassifier(n_neighbors = 3, weights = "distance")
knn_clf.fit(X_train, y_train)

from sklearn.metrics import accuracy_score
y_pred = knn_clf.predict(X_test)
print(accuracy_score(y_test,y_pred))

KNeiborsClassifier로 선정하였다. 

n_neighbors = 2로 했을 때, 96.9정도의 성능이 나와 3으로 변경하니 0.9717에 도달하였다.

---

2. MNIST 이미지를 (왼, 오른, 위, 아래) 어느 방향으로든 한 픽셀 이동시킬 수 있는 함수를 만들어보세요. 그런 다음 훈련 세트에 있는 각 이미지에 대해 네 개의 이동된 복사본 (방향마다 한 개씩)을 만들어 훈련 세트에 추가하세요. 마지막으로 이 확장된 데이터셋에서 앞에서 찾은 최선의 모델을 훈련시키고 테스트 세트에서 정확도를 측정해보세요. 모델 성능이 더 높아졌는지 확인해보세요!

 

from sklearn.datasets import fetch_openml
import numpy as np

mnist = fetch_openml('mnist_784', version = 1)
X, y = mnist["data"], mnist["target"]

y = y.astype(np.int)

#Train, Test set 나누기
X_train, X_test, y_train, y_test = X[:60000], X[60000:], y[:60000], y[60000:]

여기까지는 위와 동일하다.

from scipy.ndimage.interpolation import shift

#이미지 픽셀 변경 함수
def shift_image(image, dx, dy):
  image = image.reshape((28, 28))
  shifted_image = shift(image, [dy, dx], cval = 0 , mode = "constant")
  return shifted_image.reshape([-1])
  
#증강할 이미지 시각화
image = X_train.to_numpy()[1000]
shifted_image_down = shift_image(image, 0, 5)
shifted_image_left = shift_image(image, -5, 0)
shifted_image_right = shift_image(image, 5, 0)
shifted_image_up = shift_image(image, 0, -5)

plt.figure(figsize=(12,5))
plt.subplot(1,5,1)
plt.title("Original", fontsize=14)
plt.imshow(image.reshape(28, 28), interpolation="nearest", cmap="Greys")
plt.subplot(1,5,2)
plt.title("Shifted down", fontsize=14)
plt.imshow(shifted_image_down.reshape(28, 28), interpolation="nearest", cmap="Greys")
plt.subplot(1,5,3)
plt.title("Shifted left", fontsize=14)
plt.imshow(shifted_image_left.reshape(28, 28), interpolation="nearest", cmap="Greys")
plt.subplot(1,5,4)
plt.title("Shifted right", fontsize=14)
plt.imshow(shifted_image_right.reshape(28, 28), interpolation="nearest", cmap="Greys")
plt.subplot(1,5,5)
plt.title("Shifted up", fontsize=14)
plt.imshow(shifted_image_up.reshape(28, 28), interpolation="nearest", cmap="Greys")

plt.show()

#증강기법을 적용
X_train_augmented = [X_train.to_numpy()[i] for i in range(len(X_train))]
y_train_augmented = [label for label in y_train]

for dx, dy in ((1, 0), (-1, 0), (0, 1), (0, -1)):
    for image, label in zip(X_train.values, y_train):
        X_train_augmented.append(shift_image(image, dx, dy))
        y_train_augmented.append(label)

X_train_augmented = np.array(X_train_augmented)
y_train_augmented = np.array(y_train_augmented)

위 코드는 증강기법을 적용하여 데이터의 수를 불리는 과정이다.

from sklearn.neighbors import KNeighborsClassifier

knn_clf = KNeighborsClassifier(n_neighbors = 3, weights = "distance")
knn_clf.fit(X_train_augmented, y_train_augmented)

from sklearn.metrics import accuracy_score
y_pred = knn_clf.predict(X_test)
print(accuracy_score(y_test,y_pred))

동일하게 KNeiborsClassifier로 진행하니 0.9717 -> 0.9763 으로 유의미한 성능 차이를 거두었다.