CH05_SEC07_1_SVM.ipynb

Kernel: Python 3

import numpy as np
import matplotlib.pyplot as plt
from matplotlib import rcParams
from scipy import io
import os
from sklearn import svm
from sklearn.model_selection import cross_val_score
from mpl_toolkits import mplot3d


rcParams.update({'font.size': 18})
plt.rcParams['figure.figsize'] = [12, 12]

# Random ellipse 1

n1 = 300 # Training set size

x1 = 1.5*np.random.randn(n1) - 1.5
y1 = 1.2*np.random.randn(n1) + np.power(x1+1.5,2) - 7
x2 = 1.5*np.random.randn(n1) + 1.5
y2 = 1.2*np.random.randn(n1) - np.power(x2-1.5,2) + 7

plt.plot(x1,y1,'ro',markerfacecolor=(0,1,0.2),markeredgecolor='k',ms=12)
plt.plot(x2,y2,'bo',markerfacecolor=(0.9,0,1),markeredgecolor='k',ms=12)

plt.show()

z1 = np.power(x1,2) * y1
z2 = np.power(x2,2) * y2

fig = plt.figure()
ax = plt.axes(projection='3d')

ax.plot(x1,y1,z1,'ro',markerfacecolor=(0,1,0.2),markeredgecolor='k',ms=12)
ax.plot(x2,y2,z2,'bo',markerfacecolor=(0.9,0,1),markeredgecolor='k',ms=12)

ax.view_init(20, -135)

plt.show()

r = 7 + np.random.randn(n1)
th = 2 * np.pi * np.random.randn(n1)
xr = r * np.cos(th)
yr = r * np.sin(th)
x5 = np.random.randn(n1)
y5 = np.random.randn(n1)

zr = np.power(xr,2) + np.power(yr,2)
z5 = np.power(x5,2) + np.power(y5,2)

fig = plt.figure()
ax = plt.axes(projection='3d')
ax.view_init(20, -135)

ax.plot(xr,yr,zr+40,'ro',markerfacecolor=(0,1,0.2),markeredgecolor='k',ms=12)
ax.plot(x5,y5,z5+40,'bo',markerfacecolor=(0.9,0,1),markeredgecolor='k',ms=12)


x = np.arange(-10,10.5,0.5)
y = x
X,Y = np.meshgrid(x,y)
F3 = 54 + 0*X + 0*Y

ax.plot_surface(X, Y, F3, cmap='gray',linewidth=0, antialiased=True,alpha=0.2)

ax.plot(xr,yr,np.zeros(*xr.shape),'ro',markerfacecolor=(179/255,1,179/255),markeredgecolor='k',ms=12)
ax.plot(x5,y5,np.zeros(*x5.shape),'bo',markerfacecolor=(240/255,194/255,224/255),markeredgecolor='k',ms=12)

theta = np.linspace(0,2*np.pi,100)
xrr = np.sqrt(14)*np.cos(theta)
yrr = np.sqrt(14)*np.sin(theta)

ax.plot(xrr,yrr,np.zeros(*xrr.shape),'k-',linewidth=2)

plt.show()

# Classify dogs vs. cats
dogdata_w_mat = io.loadmat(os.path.join('..','DATA','dogData_w.mat'))
catdata_w_mat = io.loadmat(os.path.join('..','DATA','catData_w.mat'))

dog_wave = dogdata_w_mat['dog_wave']
cat_wave = catdata_w_mat['cat_wave']

CD = np.concatenate((dog_wave,cat_wave),axis=1)

u,s,vT = np.linalg.svd(CD-np.mean(CD),full_matrices=0)
v = vT.T

features = np.arange(1,21)
xtrain = np.concatenate((v[:60,features],v[80:140,features]))
test = np.concatenate((v[60:80,features],v[140:160,features]))

label = np.repeat(np.array([1,-1]),60)
truth = np.repeat(np.array([1,-1]),20)


Mdl = svm.SVC(kernel='rbf',gamma='auto').fit(xtrain,label)
test_labels = Mdl.predict(test)

CMdl = cross_val_score(Mdl, xtrain, label, cv=10) #cross-validate the model
classLoss = 1-np.mean(CMdl) # average error over all cross-validation iterations