sklearn.svm.SVC.decision bounds

import numpy as np
import matplotlib.pyplot as plt

 

# data set
from sklearn.datasets import load_breast_cancer
cancer=load_breast_cancer()
col1=0
col2=5
X=cancer.data[:,[col1,col2]]
y=cancer.target
from sklearn.model_selection import train_test_split
X_train,X_test,y_train,y_test=train_test_split(X,y)

X_mean=X_train.mean(axis=0)
X_std=X_train.std(axis=0)

X_train_norm=(X_train-X_mean)/X_std
X_test_norm=(X_test-X_mean)/X_std

 

# train and test
from sklearn.svm import SVC
model=SVC(C=1,gamma=1,probability=True)

model.fit(X_train_norm,y_train)

train_score=model.score(X_train_norm,y_train)
test_score=model.score(X_test_norm,y_test)
display(train_score,test_score)

 

# visualization

scale=200
xmax=X_train_norm[:,0].max()+1
xmin=X_train_norm[:,0].min()-1
ymax=X_train_norm[:,1].max()+1
ymin=X_train_norm[:,1].min()-1

xx=np.linspace(xmin,xmax,scale)
yy=np.linspace(ymin,ymax,scale)
data1,data2=np.meshgrid(xx,yy)

X_grid=np.c_[data1.ravel(),data2.ravel()]
decision_values=model.predict_proba(X_grid)[:,0]

sv=model.support_vectors_

fig=plt.figure(figsize=[14,12])

CS=plt.contour(data1,data2,decision_values.reshape(data1.shape),levels=[0.01,0.1,0.2,0.5,0.8,0.9,0.99])
plt.clabel(CS,inline=2,fontsize=10)

plt.scatter(sv[:,0],sv[:,1],marker='s',c='k',s=100)
# 서포트 벡터 머신이니까 서포트 벡터가 있다
# 서포트 벡터 : 비용함수를 적용할때 사용한 셈플들
# 나머지 점들은 전혀 고려 하지 않는다.
# 아래 그림에서 검정색으로 포장된 셈플들이 서포트 벡터이다.
# 선을 넘어간 애들은 기본적으로 서포트 벡터가 된다.
# 
plt.scatter(X_train_norm[:,0][y_train==0],X_train_norm[:,1][y_train==0],marker='o',c='r',label='malignant')
plt.scatter(X_train_norm[:,0][y_train==1],X_train_norm[:,1][y_train==1],marker='^',c='y',label='benign')

plt.legend()
plt.colorbar(CS,shrink=0.5)
plt.xlabel(cancer.feature_names[col1])
plt.ylabel(cancer.feature_names[col2])
plt.title('SVM - decision bounds',fontsize=20)