def testSimpleTrain1(self):
"Two 1-dimensional input points (SIGN), 1 kernel (linear)"
kerns = np.int32([0]); # 1 linear kernel
params = np.double([0.]); # params not important
features = np.int32([-1]); # use all features, but there's only 1
Xtr = np.double([[-2.,2.]]); # support points
ytr = np.int32([-1,1]); # labels
(success, Sigma, alpha, bsvm, posw) = train_mwu_mkl(kerns, params, features, Xtr, ytr)
self.assertTrue(success)
self.assertTrue((posw == 1).all(), msg='posw={0}'.format(posw))
self.assertTrue((Sigma > 0).all(), msg='Sigma={0}'.format(Sigma))
self.assertTrue((alpha >= 0).all(), msg='alpha={0}'.format(alpha))
self.assertAlmostEqual(bsvm, 0.0, msg='bsvm={0}'.format(bsvm))
def testSimpleTrain2(self):
"Four 2-dimensional input points (XOR), 1 kernel (quad)"
kerns = np.int32([1]); # 1 polynomial kernel
params = np.double([2.]); # quadratic
features = np.int32([-1]); # use all features, but there's only 1
Xtr = np.double([[-2.,2.,-2.,2.],
[-2.,-2.,2.,2.]]); # support points
ytr = np.int32([-1,1,1,-1]); # labels
(success, Sigma, alpha, bsvm, posw) = train_mwu_mkl(kerns, params, features, Xtr, ytr)
self.assertTrue(success)
self.assertTrue((posw == 1).all(), msg='posw={0}'.format(posw))
self.assertTrue((Sigma > 0).all(), msg='Sigma={0}'.format(Sigma))
self.assertTrue((alpha >= 0).all(), msg='alpha={0}'.format(alpha))
self.assertAlmostEqual(bsvm, 0.0, msg='bsvm={0}'.format(bsvm))
