Major bugfix for nonlinear GenSVM

Nonlinear GenSVM depends on the eigendecomposition of the
kernel matrix. Mathematically, the Sigma vector in the code
should contain the square root of the eigenvalues. Taking
the square root was however neglected, which resulted in
poor performance of nonlinear GenSVM. This is now fixed,
which means that the performance of nonlinear GenSVM will
be much better.

1 files changed, 30 insertions, 27 deletions

diff --git a/tests/src/test_gensvm_train.c b/tests/src/test_gensvm_train.c
index b95de28..f7033f0 100644
--- a/tests/src/test_gensvm_train.c
+++ b/tests/src/test_gensvm_train.c
@@ -275,23 +275,24 @@ char *test_gensvm_train_seed_kernel()
 	mu_assert(model->K == data->K, "Incorrect model K");
 
 	double eps = 1e-13;
+
 	mu_assert(fabs(matrix_get(data->Sigma, 1, 0, 0) -
-				7.8302939172918506) < eps,
+				2.7982662341692670) < eps,
 			"Incorrect data->Sigma at 0, 0");
 	mu_assert(fabs(matrix_get(data->Sigma, 1, 1, 0) -
-				0.7947913383766066) < eps,
+				0.8915107056993801) < eps,
 			"Incorrect data->Sigma at 1, 0");
 	mu_assert(fabs(matrix_get(data->Sigma, 1, 2, 0) -
-				0.5288740088908547) < eps,
+				0.7272372438832145) < eps,
 			"Incorrect data->Sigma at 2, 0");
 	mu_assert(fabs(matrix_get(data->Sigma, 1, 3, 0) -
-				0.4537982052555444) < eps,
+				0.6736454596117636) < eps,
 			"Incorrect data->Sigma at 3, 0");
 	mu_assert(fabs(matrix_get(data->Sigma, 1, 4, 0) -
-				0.2226012271232192) < eps,
+				0.4718063449374322) < eps,
 			"Incorrect data->Sigma at 4, 0");
 	mu_assert(fabs(matrix_get(data->Sigma, 1, 5, 0) -
-				0.0743004417495061) < eps,
+				0.2725810737184557) < eps,
 			"Incorrect data->Sigma at 5, 0");
 
 	// we need a large eps here because there are numerical precision 
@@ -299,69 +300,71 @@ char *test_gensvm_train_seed_kernel()
 	// compare with absolute values because of variability in the 
 	// eigendecomposition.
 	eps = 1e-7;
+
 	mu_assert(fabs(fabs(matrix_get(model->V, model->K-1, 0, 0)) -
-				fabs(5.0555413160638665)) < eps,
+				fabs(1.3968329665264863)) < eps,
 			"Incorrect model->V at 0, 0");
 	mu_assert(fabs(fabs(matrix_get(model->V, model->K-1, 0, 1)) -
-				fabs(-2.2586632211763198)) < eps,
+				fabs(-0.4491223112772532)) < eps,
 			"Incorrect model->V at 0, 1");
 	mu_assert(fabs(fabs(matrix_get(model->V, model->K-1, 0, 2)) -
-				fabs(-4.5572671806963143)) < eps,
+				fabs(-1.2044427235549637)) < eps,
 			"Incorrect model->V at 0, 2");
 	mu_assert(fabs(fabs(matrix_get(model->V, model->K-1, 1, 0)) -
-				fabs(-1.9627432869558412)) < eps,
+				fabs(-1.2834234211019704)) < eps,
 			"Incorrect model->V at 1, 0");
 	mu_assert(fabs(fabs(matrix_get(model->V, model->K-1, 1, 1)) -
-				fabs(0.9934555242449399)) < eps,
+				fabs(0.6330939040375793)) < eps,
 			"Incorrect model->V at 1, 1");
 	mu_assert(fabs(fabs(matrix_get(model->V, model->K-1, 1, 2)) -
-				fabs(1.7855287218670219)) < eps,
+				fabs(1.2876548429115076)) < eps,
 			"Incorrect model->V at 1, 2");
 	mu_assert(fabs(fabs(matrix_get(model->V, model->K-1, 2, 0)) -
-				fabs(1.9393083227054353)) < eps,
+				fabs(2.0023377286211428)) < eps,
 			"Incorrect model->V at 2, 0");
 	mu_assert(fabs(fabs(matrix_get(model->V, model->K-1, 2, 1)) -
-				fabs(-1.1958487809502740)) < eps,
+				fabs(-1.5454495147993872)) < eps,
 			"Incorrect model->V at 2, 1");
 	mu_assert(fabs(fabs(matrix_get(model->V, model->K-1, 2, 2)) -
-				fabs(2.1140967864804359)) < eps,
+				fabs(1.8380262406111434)) < eps,
 			"Incorrect model->V at 2, 2");
 	mu_assert(fabs(fabs(matrix_get(model->V, model->K-1, 3, 0)) -
-				fabs(2.3909204618652535)) < eps,
+				fabs(1.8873525552961188)) < eps,
 			"Incorrect model->V at 3, 0");
 	mu_assert(fabs(fabs(matrix_get(model->V, model->K-1, 3, 1)) -
-				fabs(-0.2834554569573399)) < eps,
+				fabs(-0.5671111794102348)) < eps,
 			"Incorrect model->V at 3, 1");
 	mu_assert(fabs(fabs(matrix_get(model->V, model->K-1, 3, 2)) -
-				fabs(1.0926232371314393)) < eps,
+				fabs(1.3530484176263944)) < eps,
 			"Incorrect model->V at 3, 2");
 	mu_assert(fabs(fabs(matrix_get(model->V, model->K-1, 4, 0)) -
-				fabs(3.3374545494113272)) < eps,
+				fabs(2.9991675684385952)) < eps,
 			"Incorrect model->V at 4, 0");
 	mu_assert(fabs(fabs(matrix_get(model->V, model->K-1, 4, 1)) -
-				fabs(1.6699291195221897)) < eps,
+				fabs(1.6232323178615611)) < eps,
 			"Incorrect model->V at 4, 1");
 	mu_assert(fabs(fabs(matrix_get(model->V, model->K-1, 4, 2)) -
-				fabs(-1.4345249893609275)) < eps,
+				fabs(-1.0853101351516645)) < eps,
 			"Incorrect model->V at 4, 2");
 	mu_assert(fabs(fabs(matrix_get(model->V, model->K-1, 5, 0)) -
-				fabs(-0.0221825925355533)) < eps,
+				fabs(-0.2735156994082831)) < eps,
 			"Incorrect model->V at 5, 0");
 	mu_assert(fabs(fabs(matrix_get(model->V, model->K-1, 5, 1)) -
-				fabs(-0.1216077739550210)) < eps,
+				fabs(-0.2154874773946488)) < eps,
 			"Incorrect model->V at 5, 1");
 	mu_assert(fabs(fabs(matrix_get(model->V, model->K-1, 5, 2)) -
-				fabs(-0.7900947982642630)) < eps,
+				fabs(-0.9036193937904904)) < eps,
 			"Incorrect model->V at 5, 2");
 	mu_assert(fabs(fabs(matrix_get(model->V, model->K-1, 6, 0)) -
-				fabs(-0.0076471781062262)) < eps,
+				fabs(-0.1010202110238350)) < eps,
 			"Incorrect model->V at 6, 0");
 	mu_assert(fabs(fabs(matrix_get(model->V, model->K-1, 6, 1)) -
-				fabs(-0.8781872510019056)) < eps,
+				fabs(-1.7921615999242961)) < eps,
 			"Incorrect model->V at 6, 1");
 	mu_assert(fabs(fabs(matrix_get(model->V, model->K-1, 6, 2)) -
-				fabs(-0.2782284589344380)) < eps,
+				fabs(-0.6850178130530472)) < eps,
 			"Incorrect model->V at 6, 2");
+
 	// end test code //
 
 	gensvm_free_model(model);