Closed schmidtjonathan closed 1 year ago
Base: 94.32% // Head: 78.25% // Decreases project coverage by -16.07%
:warning:
Coverage data is based on head (
e866231
) compared to base (eac3538
). Patch coverage: 100.00% of modified lines in pull request are covered.
:umbrella: View full report at Codecov.
:loudspeaker: Do you have feedback about the report comment? Let us know in this issue.
Hi @schmidtjonathan . Thanks for opening this PR. Have you had a chance to look at the docs regarding how we handle lengthscales in general in KernelFunctions? (I suspect you may have missed them 😄 )
Hi @willtebbutt , thanks for your quick reply! Yes, I absolutely missed that, sorry. Thanks for the hint.
Np! Enjoy using KernelFunctions!
Summary
This PR adds a
lengthscale
parameter to theSqExponentialKernel
, making it more flexible.[1] C. E. Rasmussen and C. K. I. Williams, Gaussian processes for machine learning. Cambridge, Mass: MIT Press, 2006.
Proposed changes
According to Eq. (4.9) in [1], the Squared-Exponential Kernel (a.k.a. Gaussian Kernel, etc.) has a characteristic lengthscale as a parameter, s.t.
, where
r = d(x, y)
is the distance between two pointsx
andy
. This PR adds this parameter, wherelengthscale = 1.0
by default, recovering the previous state and thus making it compatible with applications already using it.Breaking changes
None.