add ridge regularization in the normal equation#166
add ridge regularization in the normal equation#166abhiyanpaudel wants to merge 6 commits intodevelopfrom
Conversation
jacobmerson
left a comment
jacobmerson
left a comment
There was a problem hiding this comment.
A few quick questions/changes. Then I can merge this change. Does it seem to help with the quality of the interpolation at all?
| const SupportResults& support, const LO& dim, | ||
| const LO& degree, RadialBasisFunction bf); | ||
| const LO& degree, RadialBasisFunction bf, | ||
| double lambda_factor = 0); |
There was a problem hiding this comment.
Can you add a docstring to this function explaining all of the inputs? And, what the lambda factor is?
| RealDefaultScalarArrayView approx_target_values, | ||
| double lambda_factor) | ||
| { | ||
| PCMS_FUNCTION_TIMER; |
There was a problem hiding this comment.
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.
| @@ -159,10 +160,10 @@ TEST_CASE("solver test") | |||
| Kokkos::View<double**, Kokkos::HostSpace> expected_solution( | |||
There was a problem hiding this comment.
why are the expected results changing so much here? Shouldn't we get the same values when lambda=0?
|
We decided to remove the normal equation based routines because they are horribly conditioned and do not provide good solutions for most of our test data. The SVD and QR based methods are much more effective. |
3 participants
No description provided.