[人工智能]SVM支持向量机（二）13

1??导包

import numpy as np
from sklearn import datasets
from sklearn.svm import SVC
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score
import matplotlib.pyplot as plt

2??加载数据

X,y=datasets.load_wine(return_X_y=True)
display(X.shape)
输出：
(178, 13)

x_train,x_test,y_train,y_test=train_test_split(X,y,test_size=0.2)
display(x_train.shape,y_train.shape)
输出：
(142, 13)
(142,)

3??建模

3.1??线性：linear

svc=SVC(kernel='linear')
svc.fit(x_train,y_train)
yred=svc.predict(x_test)
score=accuracy_score(yred,y_test)
display('使用linear核函数得分：',score)
输出：
'使用linear核函数得分：'
0.9444444444444444

svc.coef_
输出：
array([[ 0.68238177,  0.59090752,  1.17127315, -0.21150042,  0.0081547 ,
        -0.01397723,  0.2235816 ,  0.22634936, -0.41167559,  0.47889006,
        -0.16003878,  0.50632237,  0.00469413],
       [ 0.08796641,  0.13272108,  0.04146849, -0.07352927,  0.00631995,
         0.28416506,  0.43334105,  0.00655127,  0.18678886, -0.26506095,
         0.03456889,  0.29852352,  0.00403859],
       [-0.6507222 , -0.33430121, -0.36136551, -0.02392738, -0.02279356,
         0.37705139,  1.19788242,  0.28111838,  0.54743024, -0.61747287,
         0.43714574,  0.94564814, -0.00857966]])

3.2??poly多项式（方程幂次方大于1的）

svc=SVC(kernel='poly',degree=2)
svc.fit(x_train,y_train)
yred=svc.predict(x_test)
score=accuracy_score(yred,y_test)
display('使用ploy核函数得分：',score)
输出：
'使用ploy核函数得分：'
0.7222222222222222

3.3??rbf高斯分布

svc=SVC(kernel='rbf')
svc.fit(x_train,y_train)
yred=svc.predict(x_test)
score=accuracy_score(yred,y_test)
display('使用rbf核函数得分：',score)
输出：
'使用rbf核函数得分：'
0.6111111111111112

3.4??sigmoid核函数

svc=SVC(kernel='sigmoid')
svc.fit(x_train,y_train)
yred=svc.predict(x_test)
score=accuracy_score(yred,y_test)
display('使用sigmoid核函数得分：',score)
输出：

'使用sigmoid核函数得分：'
0.19444444444444445

4??非线性核函数

import matplotlib.pyplot as plt
from matplotlib.colors import ListedColormap

4.1??创造数据

X,y=datasets.make_circles(n_samples=100,factor=0.7)#获取圆圈的数据，数量100个，大小0.7
display(X.shape,y.shape)
X+=np.random.randn(100,2)*0.03
输出：
(100, 2)
(100,)

plt.figure(figsize=(5,5))
camp=ListedColormap(colors=['red','blue'])
plt.scatter(X[:,0],X[:,1],c=y,cmap=camp)

?

?4.2??线性核函数

svc=SVC(kernel='linear')
svc.fit(X,y)
svc.score(X,y)
输出：
0.51

4.3??多项式poly（升维）

svc=SVC(kernel='poly',degree=2)
svc.fit(X,y)
svc.score(X,y)
输出：
1.0

4.4??rbf高斯核函数

svc=SVC(kernel='rbf',degree=2)
svc.fit(X,y)
svc.score(X,y)
输出：
1.0

5??支持向量机回归问题???????

import numpy as np
from sklearn.svm import SVR
import matplotlib.pyplot as plt

5.1??创建数据

X=np.linspace(0,2*np.pi,100).reshape(-1,1)
y=np.sin(X)
plt.scatter(X,y)

?5.2??建模

5.2.1??线性核函数

svr=SVR(kernel='linear')
svr.fit(X,y.ravel())
y_=svr.predict(X)

svr.score(X,y)
plt.scatter(X,y)
plt.scatter(X,y_)

?5.2.2??poly多项式核函数

svr=SVR(kernel='poly',degree=3)
svr.fit(X,y.ravel())
y_=svr.predict(X)

svr.score(X,y)
plt.scatter(X,y)
plt.scatter(X,y_)

?5.3??rbf径向基（高斯核函数）

?

svr=SVR(kernel='rbf',coef0=300)#cofe0为截距的参数
svr.fit(X,y.ravel())
y_=svr.predict(X)

svr.score(X,y)
plt.scatter(X,y)
plt.scatter(X,y_)

?

?6??天猫双十一销量数据预测

X=np.linspace(2009,2019,11)-2008
y = np.array([0.5,9.36,52,191,350,571,912,1207,1682,2135,2684])
plt.scatter(X,y)

x_test=np.linspace(2009,2019,100).reshape(-1, 1)-2008

6.1??linear核函数

svr=SVR(kernel='linear')
svr.fit(X.reshape(-1, 1),y)
y_=svr.predict(x_test)
plt.scatter(X,y)
plt.scatter(x_test,y_,c='green')

svr=SVR(kernel='poly',degree=3,coef0=200)#coef0截距=200
svr.fit(X.reshape(-1, 1),y)
y_=svr.predict(x_test)
plt.scatter(X,y,s=80)
plt.scatter(x_test,y_,c='red')

?

?

svr=SVR(kernel='rbf')
svr.fit(X.reshape(-1, 1),y)
y_=svr.predict(x_test)
plt.scatter(X,y)
plt.scatter(x_test,y_,c='green')

?

svr=SVR(kernel='sigmoid')#coef0截距=200
svr.fit(X.reshape(-1, 1),y)
y_=svr.predict(x_test)
plt.scatter(X,y,s=80)
plt.scatter(x_test,y_,c='red')

?