DataManyo

3.4 UCI 데이터 세트에서 두 개의 데이터를 골라 10-fold 교차 검증법과 Leav-one-out이 측정한 로지스틱 회귀의 오차율을 비교하라.

참고 답안 코드 (1):

import numpy as np
from sklearn import linear_model
from sklearn.model_selection import LeaveOneOut
from sklearn.model_selection import cross_val_score

#data 경로는 사용자가 지정해 놓은 위치로 변경합니다.
data_path = r'C:\Users\hanmi\Documents\xiguabook\Transfusion.txt'

data = np.loadtxt(data_path, delimiter=',').astype(int)

X = data[:, :4]
y = data[:, 4]

m, n = X.shape

# normalization
X = (X - X.mean(0)) / X.std(0)

# shuffle
index = np.arange(m)
np.random.shuffle(index)

X = X[index]
y = y[index]

# sklearn 라이브러리를 사용
# k-10 cross validation
lr = linear_model.LogisticRegression(C=2)

score = cross_val_score(lr, X, y, cv=10)

print(score.mean())

# LOO
loo = LeaveOneOut()

accuracy = 0
for train, test in loo.split(X, y):
    lr_ = linear_model.LogisticRegression(C=2)
    X_train = X[train]
    X_test = X[test]
    y_train = y[train]
    y_test = y[test]
    lr_.fit(X_train, y_train)

    accuracy += lr_.score(X_test, y_test)

print(accuracy / m)

# 결과는 비슷합니다.

# k-10

num_split = int(m / 10)
score_my = []
for i in range(10):
    lr_ = linear_model.LogisticRegression(C=2)
    test_index = range(i * num_split, (i + 1) * num_split)
    X_test_ = X[test_index]
    y_test_ = y[test_index]

    X_train_ = np.delete(X, test_index, axis=0)
    y_train_ = np.delete(y, test_index, axis=0)

    lr_.fit(X_train_, y_train_)

    score_my.append(lr_.score(X_test_, y_test_))

print(np.mean(score_my))

# LOO
score_my_loo = []
for i in range(m):
    lr_ = linear_model.LogisticRegression(C=2)
    X_test_ = X[i, :]
    y_test_ = y[i]

    X_train_ = np.delete(X, i, axis=0)
    y_train_ = np.delete(y, i, axis=0)

    lr_.fit(X_train_, y_train_)

    score_my_loo.append(int(lr_.predict(X_test_.reshape(1, -1)) == y_test_))

print(np.mean(score_my_loo))

# 결과는 모두 비슷합니다.

참고 답안 코드 (2):

```python
import numpy as np
from sklearn import linear_model
from sklearn.model_selection import LeaveOneOut
from sklearn.model_selection import cross_val_score

#data 경로는 사용자가 지정해 놓은 위치로 변경합니다.
data_path = r'C:\Users\hanmi\Documents\xiguabook\Transfusion.txt'

data = np.loadtxt(data_path, delimiter=',').astype(int)

X = data[:, :4]
y = data[:, 4]

m, n = X.shape

# normalization
X = (X - X.mean(0)) / X.std(0)

# shuffle
index = np.arange(m)
np.random.shuffle(index)

X = X[index]
y = y[index]

# sklearn 라이브러리를 사용
# k-10 cross validation
lr = linear_model.LogisticRegression(C=2)

score = cross_val_score(lr, X, y, cv=10)

print(score.mean())

# LOO
loo = LeaveOneOut()

accuracy = 0
for train, test in loo.split(X, y):
    lr_ = linear_model.LogisticRegression(C=2)
    X_train = X[train]
    X_test = X[test]
    y_train = y[train]
    y_test = y[test]
    lr_.fit(X_train, y_train)

    accuracy += lr_.score(X_test, y_test)

print(accuracy / m)

# 결과는 비슷합니다.

# k-10

num_split = int(m / 10)
score_my = []
for i in range(10):
    lr_ = linear_model.LogisticRegression(C=2)
    test_index = range(i * num_split, (i + 1) * num_split)
    X_test_ = X[test_index]
    y_test_ = y[test_index]

    X_train_ = np.delete(X, test_index, axis=0)
    y_train_ = np.delete(y, test_index, axis=0)

    lr_.fit(X_train_, y_train_)

    score_my.append(lr_.score(X_test_, y_test_))

print(np.mean(score_my))

# LOO
score_my_loo = []
for i in range(m):
    lr_ = linear_model.LogisticRegression(C=2)
    X_test_ = X[i, :]
    y_test_ = y[i]

    X_train_ = np.delete(X, i, axis=0)
    y_train_ = np.delete(y, i, axis=0)

    lr_.fit(X_train_, y_train_)

    score_my_loo.append(int(lr_.predict(X_test_.reshape(1, -1)) == y_test_))

print(np.mean(score_my_loo))

# 결과는 모두 비슷합니다.
```
source: https://blog.csdn.net/weixin_43518584/article/details/105588310
'''