Skip to content

add ridge regularization in the normal equation #166

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
abhiyanpaudel wants to merge 6 commits into from
Closed

add ridge regularization in the normal equation #166

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
826e82c
add ridge regularization
abhiyan123 Mar 18, 2025
c5de3e5
update lambda parameter in function argument
abhiyan123 Mar 18, 2025
7dc4669
add defaualt argument for lambda parameter
abhiyan123 Mar 18, 2025
1b73884
update lambda
abhiyan123 Mar 18, 2025
beb74d7
update expected solution
abhiyan123 Mar 18, 2025
2ecc821
Merge branch 'develop' of github.com:SCOREC/pcms into mls_interpolation
abhiyan123 Mar 18, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 5 additions & 4 deletions src/pcms/interpolator/mls_interpolation.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -105,7 +105,8 @@ Write<Real> mls_interpolation(const Reals source_values,
const Reals source_coordinates,
const Reals target_coordinates,
const SupportResults& support, const LO& dim,
const LO& degree, RadialBasisFunction bf)
const LO& degree, RadialBasisFunction bf,
double lambda_factor)
{

const auto nvertices_target = target_coordinates.size() / dim;
Expand All @@ -116,19 +117,19 @@ Write<Real> mls_interpolation(const Reals source_values,
case RadialBasisFunction::RBF_GAUSSIAN:
interpolated_values = detail::mls_interpolation(
source_values, source_coordinates, target_coordinates, support, dim,
degree, RBF_GAUSSIAN{});
degree, RBF_GAUSSIAN{}, lambda_factor);
break;

case RadialBasisFunction::RBF_C4:
interpolated_values = detail::mls_interpolation(
source_values, source_coordinates, target_coordinates, support, dim,
degree, RBF_C4{});
degree, RBF_C4{}, lambda_factor);
break;

case RadialBasisFunction::RBF_CONST:
interpolated_values = detail::mls_interpolation(
source_values, source_coordinates, target_coordinates, support, dim,
degree, RBF_CONST{});
degree, RBF_CONST{}, lambda_factor);
break;
}

Expand Down
3 changes: 2 additions & 1 deletion src/pcms/interpolator/mls_interpolation.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -19,6 +19,7 @@ Write<Real> mls_interpolation(const Reals source_values,
const Reals source_coordinates,
const Reals target_coordinates,
const SupportResults& support, const LO& dim,
const LO& degree, RadialBasisFunction bf);
const LO& degree, RadialBasisFunction bf,
double lambda_factor = 0);
Copy link
Collaborator

@jacobmerson jacobmerson Mar 18, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can you add a docstring to this function explaining all of the inputs? And, what the lambda factor is?

} // namespace pcms
#endif
27 changes: 21 additions & 6 deletions src/pcms/interpolator/mls_interpolation_impl.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -231,6 +231,14 @@ void scale_column_trans_matrix(const ScratchMatView& matrix,
}
}

KOKKOS_INLINE_FUNCTION
void add_regularization(const member_type& team, ScratchMatView& square_matrix,
Real lambda_factor)
{
Kokkos::parallel_for(Kokkos::TeamThreadRange(team, square_matrix.extent(0)),
[=](int i) { square_matrix(i, i) += lambda_factor; });
}

/**
* @struct ResultConvertNormal
* @brief Stores the results of matrix and vector transformations.
Expand Down Expand Up @@ -283,7 +291,8 @@ KOKKOS_INLINE_FUNCTION
ResultConvertNormal convert_normal_equation(const ScratchMatView& matrix,
const ScratchVecView& weight_vector,
const ScratchVecView& rhs,
member_type team)
member_type team,
double lambda_factor)
{

int m = matrix.extent(0);
Expand Down Expand Up @@ -319,6 +328,10 @@ ResultConvertNormal convert_normal_equation(const ScratchMatView& matrix,

team.team_barrier();

add_regularization(team, square_matrix, lambda_factor);

team.team_barrier();

KokkosBlas::Experimental::
Gemv<KokkosBlas::Mode::Team, KokkosBlas::Algo::Gemv::Unblocked>::invoke(
team, 'N', 1.0, scaled_matrix, rhs, 0.0, transformed_rhs);
Expand Down Expand Up @@ -419,7 +432,8 @@ void mls_interpolation(RealConstDefaultScalarArrayView source_values,
RealConstDefaultScalarArrayView target_coordinates,
const SupportResults& support, const LO& dim,
const LO& degree, Func rbf_func,
RealDefaultScalarArrayView approx_target_values)
RealDefaultScalarArrayView approx_target_values,
double lambda_factor)
{
PCMS_FUNCTION_TIMER;
Copy link
Collaborator

@jacobmerson jacobmerson Mar 18, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Are there any limits on the range of the lambda? I.e., does it need to be positive? If so, we should check those preconditions with an assert here.

static_assert(std::is_invocable_r_v<double, Func, double, double>,
Expand Down Expand Up @@ -550,8 +564,8 @@ void mls_interpolation(RealConstDefaultScalarArrayView source_values,

team.team_barrier();

auto result = convert_normal_equation(vandermonde_matrix, phi_vector,
support_values, team);
auto result = convert_normal_equation(
vandermonde_matrix, phi_vector, support_values, team, lambda_factor);

team.team_barrier();

Expand Down Expand Up @@ -593,7 +607,8 @@ Write<Real> mls_interpolation(const Reals source_values,
const Reals source_coordinates,
const Reals target_coordinates,
const SupportResults& support, const LO& dim,
const LO& degree, Func rbf_func)
const LO& degree, Func rbf_func,
double lambda_factor)
{
const auto nvertices_source = source_coordinates.size() / dim;

Expand All @@ -619,7 +634,7 @@ Write<Real> mls_interpolation(const Reals source_values,

mls_interpolation(source_values_array_view, source_coordinates_array_view,
target_coordinates_array_view, support, dim, degree,
rbf_func, interpolated_values_array_view);
rbf_func, interpolated_values_array_view, lambda_factor);

return interpolated_values;
}
Expand Down
27 changes: 14 additions & 13 deletions test/test_linear_solver.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -103,8 +103,9 @@ TEST_CASE("solver test")

team.team_barrier();

double lambda = 0.5;
auto result = convert_normal_equation(vandermonde_matrix, phi,
support_values, team);
support_values, team, lambda);

team.team_barrier();

Expand Down Expand Up @@ -159,10 +160,10 @@ TEST_CASE("solver test")
Kokkos::View<double**, Kokkos::HostSpace> expected_solution(
Copy link
Collaborator

@jacobmerson jacobmerson Mar 18, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

why are the expected results changing so much here? Shouldn't we get the same values when lambda=0?

"expected solution", nvertices_target, size);

expected_lhs_matrix(0, 0, 0) = 35.0;
expected_lhs_matrix(0, 0, 0) = 35.5;
expected_lhs_matrix(0, 0, 1) = 44.0;
expected_lhs_matrix(0, 1, 0) = 44.0;
expected_lhs_matrix(0, 1, 1) = 56.0;
expected_lhs_matrix(0, 1, 1) = 56.5;

expected_rhs_vector(0, 0) = 76.0;
expected_rhs_vector(0, 1) = 100.0;
Expand All @@ -174,13 +175,13 @@ TEST_CASE("solver test")
expected_scaled_matrix(0, 1, 1) = 4.0;
expected_scaled_matrix(0, 1, 2) = 6.0;

expected_solution(0, 0) = -6.0;
expected_solution(0, 1) = 6.5;
expected_solution(0, 0) = -1.519713;
expected_solution(0, 1) = 2.953405;

expected_lhs_matrix(1, 0, 0) = 30.0;
expected_lhs_matrix(1, 0, 0) = 30.5;
expected_lhs_matrix(1, 0, 1) = 20.0;
expected_lhs_matrix(1, 1, 0) = 20.0;
expected_lhs_matrix(1, 1, 1) = 30.0;
expected_lhs_matrix(1, 1, 1) = 30.5;

expected_rhs_vector(1, 0) = 70.0;
expected_rhs_vector(1, 1) = 40.0;
Expand All @@ -192,8 +193,8 @@ TEST_CASE("solver test")
expected_scaled_matrix(1, 1, 1) = -2.0;
expected_scaled_matrix(1, 1, 2) = 1.0;

expected_solution(1, 0) = 2.6;
expected_solution(1, 1) = -0.4;
expected_solution(1, 0) = 2.517680;
expected_solution(1, 1) = -0.339463;

for (int i = 0; i < nvertices_target; ++i) {
for (int j = 0; j < size; ++j) {
Expand All @@ -203,7 +204,7 @@ TEST_CASE("solver test")
i, j, l);
REQUIRE_THAT(
expected_scaled_matrix(i, j, l),
Catch::Matchers::WithinAbs(host_result_scaled(i, j, l), 1E-10));
Catch::Matchers::WithinAbs(host_result_scaled(i, j, l), 1E-6));
}
}
for (int j = 0; j < size; ++j) {
Expand All @@ -213,22 +214,22 @@ TEST_CASE("solver test")
k);
REQUIRE_THAT(
expected_lhs_matrix(i, j, k),
Catch::Matchers::WithinAbs(host_result_lhs(i, j, k), 1E-10));
Catch::Matchers::WithinAbs(host_result_lhs(i, j, k), 1E-6));
}
}
for (int j = 0; j < size; ++j) {
printf("A^T Q b: (%f,%f) at (%d,%d)\n", expected_rhs_vector(i, j),
host_result_rhs(i, j), i, j);
REQUIRE_THAT(expected_rhs_vector(i, j),
Catch::Matchers::WithinAbs(host_result_rhs(i, j), 1E-10));
Catch::Matchers::WithinAbs(host_result_rhs(i, j), 1E-6));
}

for (int j = 0; j < size; ++j) {
printf("A^T Q b: (%f,%f) at (%d,%d)\n", expected_solution(i, j),
host_result_solution(i, j), i, j);
REQUIRE_THAT(
expected_solution(i, j),
Catch::Matchers::WithinAbs(host_result_solution(i, j), 1E-10));
Catch::Matchers::WithinAbs(host_result_solution(i, j), 1E-6));
}
}
}
Expand Down