SVDD_test.py

# Author: Arman Naseri Jahfari (a.naserijahfari@tudelft.nl)

import numpy as np
from matplotlib import pyplot as plt
from SVDD import SVDD
from sklearn.metrics import confusion_matrix as conf_mat
meani = [0, 0]
covi = [[1, 0], [0, 1]]
Ni = 1000
Xi = np.random.multivariate_normal(meani, covi, Ni)
yi = np.ones(Ni).reshape(Ni, 1)
plt.scatter(Xi[:, 0], Xi[:, 1], c='b')

No = 100
Ro = np.random.uniform(2, 4, No)
theta_o = np.random.uniform(0, np.pi, No)

x_loc = (Ro * np.cos(theta_o)).reshape(No, 1)
y_loc = (Ro * np.sin(theta_o)).reshape(No, 1)

Xo = np.hstack([x_loc, y_loc])
yo = -1 * np.ones(No).reshape(No, 1)

plt.scatter(Xo[:, 0], Xo[:, 1], c='r')

X = np.vstack([Xi, Xo])
y = np.vstack([yi, yo]).ravel()



#%%

Xtrain = np.vstack([X[y == 1][0:900], X[y == -1][0:50]])
ytrain = np.hstack([y[y == 1][0:900], y[y == -1][0:50]])
Xtest = np.vstack([X[y == 1][900:1000], X[y == -1][50:100]])
ytest = np.hstack([y[y == 1][900:1000], y[y == -1][50:100]])

clf = SVDD(kernel_type='rbf', bandwidth=1, fracrej=np.array([0.1, 1]))
clf.fit(Xtrain, ytrain)
y_pred = clf.predict(Xtest)

print(conf_mat(ytest, y_pred, normalize='true'))
p = clf._plot_contour(Xtest, ytest)
p.savefig('test.pdf', bbox_inches='tight',pad_inches = 0)

#%% 3 sample example (all target) and 5 sample example (3 target, 2 outlier)
X1 = np.array([[1, 1], [1.5,2], [2,1]])
y1 = np.array([1, 1, 1])

bw = 1
clf1 = SVDD(kernel_type='rbf', bandwidth=bw)
clf1.fit(X1, y1)
y_pred = clf1.predict(X1)
print(y_pred)


p1 = clf1._plot_contour(X1, y1, [0,3,0,3])

# p.savefig('test.pdf', bbox_inches='tight',pad_inches = 0)
#%%
bw = 0.7
X2 = np.array([[1, 1], [2,2], [2,1], [1.5, 0.8], [1.5, 1] ])
y2 = np.array([1, 1, 1, -1, -1])
clf2 = SVDD(kernel_type='rbf', bandwidth=bw)
clf2.fit(X2, y2)
y_pred = clf2.predict(X2)
p2 = clf2._plot_contour(X2, y2, [0.5,3,0,3])

print(conf_mat(y2, y_pred, normalize='true'))