Linear Regression : Logistic Regression [Wine] (4)

0. Dataset

# 데이터 불러오기. y값은 이미 범주형으로 되어있음.
import pandas as pd
import numpy as np

dat_wine=pd.read_csv('http://archive.ics.uci.edu/ml/machine-learning-databases/'
                     'wine/wine.data',header=None)
dat_wine.head()
dat_wine.columns = ['class label', 'alchohol', 'malic acid', 'ash', 
                    'alcalinity of ash', 'magnesium', 'total phenols', 
                    'flavanoids', 'nonflavanoid phenols', 
                    'proanthocyanins', 'color intensity', 'hue', 
                    'OD208', 'proline']  # Column names
print('class label:', np.unique(dat_wine['class label']))  # Class 출력
dat_wine.head()

# 전체 data를 training set과 test set으로 split
from sklearn.model_selection import train_test_split
X, y = dat_wine.iloc[:,1:].values, dat_wine.iloc[:,0].values
X_train, X_test, y_train,y_test = train_test_split(X, y, test_size=0.3, random_state=1, stratify=y)

 

1. Data Normalization

# 표준화
from sklearn.preprocessing import StandardScaler
std = StandardScaler()
X_train_std = std.fit_transform(X_train)
X_test_std = std.transform(X_test)

 

2. Inference

# Logistic Regression with L2 or L1 Regularization
from sklearn.linear_model import LogisticRegression
lr2_10 = LogisticRegression(penalty='l2', C=10.0, solver='saga')  # L2 with C(=1/λ)=10
lr2_1 = LogisticRegression(penalty='l2', C=1.0, solver='saga')    # L2 with C(=1/λ)=1
lr2_0_1 = LogisticRegression(penalty='l2', C=0.1, solver='saga')  # L2 with C(=1/λ)=0.1

lr1_10 = LogisticRegression(penalty='l1', C=10.0, solver='saga')  # L1 with C(=1/λ)=10
lr1_1 = LogisticRegression(penalty='l1', C=1.0, solver='saga')    # L1 with C(=1/λ)=1
lr1_0_1 = LogisticRegression(penalty='l1', C=0.1, solver='saga')  # L1 with C(=1/λ)=0.1

 

3. Score

# 규제화 방법(L2 or L1)과 규제강도(λ)를 바꿔가며 accuracy score 계산
lr2_10.fit(X_train_std, y_train)
print('Training accuracy with L2 and λ=0.1:', lr2_10.score(X_train_std, y_train))
print('Test accuracy with L2 and λ=0.1:', lr2_10.score(X_test_std, y_test))

lr2_1.fit(X_train_std, y_train)  # warning..
print('Training accuracy with L2 and λ=1:', lr2_1.score(X_train_std, y_train))
print('Test accuracy with L2 and λ=1:', lr2_1.score(X_test_std, y_test))

lr2_0_1.fit(X_train_std, y_train)
print('Training accuracy with L2 and λ=10:', lr2_0_1.score(X_train_std, y_train))
print('Test accuracy with L2 and λ=10:', lr2_0_1.score(X_test_std, y_test))

lr1_10.fit(X_train_std, y_train)
print('Training accuracy with L1 and λ=0.1:', lr1_10.score(X_train_std, y_train))
print('Test accuracy with L1 and λ=0.1:', lr1_10.score(X_test_std, y_test))

lr1_1.fit(X_train_std, y_train)
print('Training accuracy with L1 and λ=1:', lr1_1.score(X_train_std, y_train))
print('Test accuracy with L1 and λ=1:', lr1_1.score(X_test_std, y_test))

lr1_0_1.fit(X_train_std, y_train)
print('Training accuracy with L1 and λ=10:', lr1_0_1.score(X_train_std, y_train))
print('Test accuracy with L1 and λ=10:', lr1_0_1.score(X_test_std, y_test))

 

# L2 규제의 규제강도(C=1/λ)를 바꿔가며 계수 추정치 관찰
print(lr2_10.intercept_)
print(lr2_1.intercept_)
print(lr2_0_1.intercept_)

print(lr2_10.coef_)
print(lr2_1.coef_)
print(lr2_0_1.coef_)

 

# L1 규제의 규제강도(C=1/λ)를 바꿔가며 계수 추정치 관찰
print(lr1_10.intercept_)
print(lr1_1.intercept_)
print(lr1_0_1.intercept_)

print(lr1_10.coef_)
print(lr1_1.coef_)
print(lr1_0_1.coef_)