def test_more_complex_kkt():
X=numpy.array([
[1,0],
[2,1],
[2,0],
[4,2],
[0,1],
[0,2],
[1,2],
[2,8]])
Y=numpy.array([1,1,1,1,-1,-1,-1,-1])
C=4.
kernel=numpy.dot
Alphas,b=svm_learn_smo(X,Y,kernel,C)
def verify():
SVM=(X,Y,Alphas,b,C,kernel)
sv=numpy.array([1,1,0,0,1,0,1,0])
nsv=~sv
computed_sv = (Alphas > 0) & (Alphas < C)
computed_nsv = ~computed_sv
assert_kkt(SVM)
assert numpy.all((sv-computed_sv) >= 0) # computed_sv in sv
assert numpy.all((computed_nsv-nsv) >= 0) # nsv in computed_nsv
assert_all_correctly_classified(SVM,X,Y)
verify()
Alphas,b=svm_learn_libsvm(X,Y,kernel,C)
verify()
def test_rbf():
X=numpy.array([
[0,0,0],
[1,1,1],
])
Y=numpy.array([ 1, -1 ])
C=10
Alphas,b=svm_learn_smo(X,Y,rbf,C)
SVM=(X,Y,Alphas,b,C,rbf)
assert_all_correctly_classified(SVM,X,Y)
Alphas,b=svm_learn_libsvm(X,Y,rbf,C)
SVM=(X,Y,Alphas,b,C,rbf)
assert_all_correctly_classified(SVM,X,Y)
def test_random():
R=numpy.random.RandomState(123)
X=R.rand(10,3)
X[:5]+=1.
C=2
Y=numpy.ones(10)
Y[5:] *= -1
Alphas,b=svm_learn_smo(X,Y,rbf,C)
SVM=(X,Y,Alphas,b,C,rbf)
assert_more_than_50(SVM,X,Y)
Alphas,b=svm_learn_libsvm(X,Y,rbf,C)
SVM=(X,Y,Alphas,b,C,rbf)
assert_more_than_50(SVM,X,Y)
def test_rbf():
X = numpy.array([
[0, 0, 0],
[1, 1, 1],
])
Y = numpy.array([1, -1])
C = 10
Alphas, b = svm_learn_smo(X, Y, rbf, C)
SVM = (X, Y, Alphas, b, C, rbf)
assert_all_correctly_classified(SVM, X, Y)
Alphas, b = svm_learn_libsvm(X, Y, rbf, C)
SVM = (X, Y, Alphas, b, C, rbf)
assert_all_correctly_classified(SVM, X, Y)
def test_random():
R = numpy.random.RandomState(123)
X = R.rand(10, 3)
X[:5] += 1.
C = 2
Y = numpy.ones(10)
Y[5:] *= -1
Alphas, b = svm_learn_smo(X, Y, rbf, C)
SVM = (X, Y, Alphas, b, C, rbf)
assert_more_than_50(SVM, X, Y)
Alphas, b = svm_learn_libsvm(X, Y, rbf, C)
SVM = (X, Y, Alphas, b, C, rbf)
assert_more_than_50(SVM, X, Y)
def test_simplest():
X=numpy.array([
[1],
[2],
])
Y=numpy.array([1,-1])
C=4.
kernel=numpy.dot
Alphas,b=svm_learn_smo(X,Y,kernel,C)
SVM=(X,Y,Alphas,b,C,kernel)
Alphas_,b_=svm_learn_libsvm(X,Y,kernel,C)
assert approximate(Alphas,Alphas_)
assert approximate(b,b_)
assert approximate(Alphas,[2.,2.])
assert approximate(b,-3)
assert_kkt(SVM)
assert_all_correctly_classified(SVM,X,Y)
def test_simplest():
X = numpy.array([
[1],
[2],
])
Y = numpy.array([1, -1])
C = 4.
kernel = numpy.dot
Alphas, b = svm_learn_smo(X, Y, kernel, C)
SVM = (X, Y, Alphas, b, C, kernel)
Alphas_, b_ = svm_learn_libsvm(X, Y, kernel, C)
assert approximate(Alphas, Alphas_)
assert approximate(b, b_)
assert approximate(Alphas, [2., 2.])
assert approximate(b, -3)
assert_kkt(SVM)
assert_all_correctly_classified(SVM, X, Y)
def test_more_complex_kkt():
X = numpy.array([[1, 0], [2, 1], [2, 0], [4, 2], [0, 1], [0, 2], [1, 2],
[2, 8]])
Y = numpy.array([1, 1, 1, 1, -1, -1, -1, -1])
C = 4.
kernel = numpy.dot
Alphas, b = svm_learn_smo(X, Y, kernel, C)
def verify():
SVM = (X, Y, Alphas, b, C, kernel)
sv = numpy.array([1, 1, 0, 0, 1, 0, 1, 0])
nsv = ~sv
computed_sv = (Alphas > 0) & (Alphas < C)
computed_nsv = ~computed_sv
assert_kkt(SVM)
assert numpy.all((sv - computed_sv) >= 0) # computed_sv in sv
assert numpy.all((computed_nsv - nsv) >= 0) # nsv in computed_nsv
assert_all_correctly_classified(SVM, X, Y)
verify()
Alphas, b = svm_learn_libsvm(X, Y, kernel, C)
verify()
