int main(int argc, char * argv[])
{
using namespace Eigen;
using namespace std;
using namespace igl;
VectorXd D;
if(!read_triangle_mesh("../shared/beetle.off",V,F))
{
cout<<"failed to load mesh"<<endl;
}
twod = V.col(2).minCoeff()==V.col(2).maxCoeff();
bbd = (V.colwise().maxCoeff()-V.colwise().minCoeff()).norm();
SparseMatrix<double> L,M;
cotmatrix(V,F,L);
L = (-L).eval();
massmatrix(V,F,MASSMATRIX_TYPE_DEFAULT,M);
const size_t k = 5;
if(!eigs(L,M,k+1,EIGS_TYPE_SM,U,D))
{
cout<<"failed."<<endl;
}
U = ((U.array()-U.minCoeff())/(U.maxCoeff()-U.minCoeff())).eval();
igl::viewer::Viewer viewer;
viewer.callback_key_down = [&](igl::viewer::Viewer & viewer,unsigned char key,int)->bool
{
switch(key)
{
default:
return false;
case ' ':
{
U = U.rightCols(k).eval();
// Rescale eigen vectors for visualization
VectorXd Z =
bbd*0.5*U.col(c);
Eigen::MatrixXd C;
igl::parula(U.col(c).eval(),false,C);
c = (c+1)%U.cols();
if(twod)
{
V.col(2) = Z;
}
viewer.data.set_mesh(V,F);
viewer.data.compute_normals();
viewer.data.set_colors(C);
return true;
}
}
};
viewer.callback_key_down(viewer,' ',0);
viewer.core.show_lines = false;
viewer.launch();
}
void ActivationFunctionsTestCase::linear()
{
const int N = 1000;
Eigen::MatrixXd a = Eigen::VectorXd::Random(N) * 10.0;
Eigen::MatrixXd z = Eigen::VectorXd::Zero(N);
OpenANN::linear(a, z);
ASSERT_EQUALS(a.minCoeff(), z.minCoeff());
ASSERT_EQUALS(a.maxCoeff(), z.maxCoeff());
Eigen::MatrixXd gd = Eigen::VectorXd::Zero(N);
Eigen::MatrixXd expected = Eigen::VectorXd::Ones(N);
OpenANN::linearDerivative(gd);
ASSERT_EQUALS(gd.sum(), expected.sum());
}
void ActivationFunctionsTestCase::normaltanh()
{
const int N = 1000;
Eigen::MatrixXd a = Eigen::VectorXd::Random(N) * 10.0;
Eigen::MatrixXd z = Eigen::VectorXd::Zero(N);
OpenANN::normaltanh(a, z);
ASSERT_WITHIN(z.minCoeff(), -1.0, -0.5);
ASSERT_WITHIN(z.maxCoeff(), 0.5, 1.0);
Eigen::MatrixXd gd = Eigen::VectorXd::Zero(N);
OpenANN::normaltanhDerivative(z, gd);
ASSERT_WITHIN(gd.minCoeff(), 0.0, 1.0);
ASSERT_WITHIN(gd.maxCoeff(), 0.0, 1.0);
}
void ActivationFunctionsTestCase::logistic()
{
const int N = 1000;
Eigen::MatrixXd a = Eigen::VectorXd::Random(N) * 10.0;
Eigen::MatrixXd z = Eigen::VectorXd::Zero(N);
OpenANN::logistic(a, z);
ASSERT_WITHIN(z.minCoeff(), 0.0, 0.2);
ASSERT_WITHIN(z.maxCoeff(), 0.8, 1.0);
Eigen::MatrixXd gd = Eigen::VectorXd::Zero(N);
OpenANN::logisticDerivative(z, gd);
ASSERT_WITHIN(gd.minCoeff(), 0.0, 1.0);
ASSERT_WITHIN(gd.maxCoeff(), 0.0, 1.0);
}
void ActivationFunctionsTestCase::softmax()
{
const int N = 1000;
Eigen::MatrixXd a = Eigen::VectorXd::Random(N).transpose();
OpenANN::softmax(a);
ASSERT_EQUALS_DELTA(1.0, a.sum(), 1e-3);
ASSERT_WITHIN(a.minCoeff(), 0.0, 1.0);
ASSERT_WITHIN(a.maxCoeff(), 0.0, 1.0);
}
void scaleData(Eigen::MatrixXd& data, double min, double max)
{
if(min >= max)
throw OpenANNException("Scaling failed: max has to be greater than min!");
const double minData = data.minCoeff();
const double maxData = data.maxCoeff();
const double dataRange = maxData - minData;
const double desiredRange = max - min;
const double scaling = desiredRange / dataRange;
data = data.array() * scaling + (min - minData * scaling);
}
void ActivationFunctionsTestCase::rectifier()
{
const int N = 1000;
Eigen::MatrixXd a = Eigen::MatrixXd::Random(1, N) * 10.0;
Eigen::MatrixXd z = Eigen::MatrixXd::Zero(1, N);
OpenANN::rectifier(a, z);
ASSERT_EQUALS(0.0, z.minCoeff());
ASSERT_EQUALS(a.maxCoeff(), z.maxCoeff());
Eigen::MatrixXd gd = Eigen::MatrixXd::Zero(1, N);
Eigen::MatrixXd expected = Eigen::MatrixXd::Ones(1, N);
for(int i = 0; i < N; i++)
expected(i) *= (double)(z(i) > 0.0);
OpenANN::rectifierDerivative(z, gd);
ASSERT_EQUALS(gd.sum(), expected.sum());
}