add gibbskernel example to prior example script #375
Conversation
| @@ -120,6 +120,7 @@ kernels = [ | |||
| LinearKernel(), | |||
| compose(PeriodicKernel(), ScaleTransform(0.2)), | |||
| NeuralNetworkKernel(), | |||
| GibbsKernel(x -> exp(sin.(x))), | |||
There was a problem hiding this comment.
Usually we prefer the keyword argument constructor since it is a bit clearer. You can also simplify the lengthscale function a bit (broadcasting is only useful if x are vectors - in which case exp(sin.(x)) would fail though):
| GibbsKernel(x -> exp(sin.(x))), | |
| GibbsKernel(; lengthscale=exp ∘ sin), |
There was a problem hiding this comment.
I think we should generally handle vectors so something like exp(sum(sin, x)) maybe?
There was a problem hiding this comment.
It seemed the example only deals with scalar inputs but maybe I didn't check it carefully enough. If vectors occur as inputs, then I agree something like x -> exp(sum(sin, x)) or x -> sum(exp ∘ sin, x) should be used.
There was a problem hiding this comment.
Thanks both for the helpful information and suggestions. I went with x -> sum(exp ∘ sin, x) and checked locally that it works in a notebook so hopefully the CI will be happy.
make the GibbsKernel example more robust with thanks to @devmotion and @theogf
|
Can this be merged? |
|
@Cyberface, unfortunately since this is from a fork we cannot visualize the result... Could you run the docs locally and post here a screenshot of the resulting GP prior? :) |
|
@theogf thanks for letting me know what to do, I managed to build the docs using the README in the docs dir. The GP prior example page is quite long but there should only be two difference.
|
|
That looks great! Thanks for the addition |
Codecov Report
@@ Coverage Diff @@
## master #375 +/- ##
===========================================
- Coverage 89.23% 69.54% -19.69%
===========================================
Files 52 52
Lines 1198 1192 -6
===========================================
- Hits 1069 829 -240
- Misses 129 363 +234
Continue to review full report at Codecov.
|
Summary
Adding an example how to use the Gibbs Kernel in the prior example.
The Gibbs kernel has a lengthscale function as it's parameter.
I just chose a simple function for demonstration purposes i.e.
l(x) = exp(sin(x)).Proposed changes
examples/gaussian-process-priors/script.jlWhat alternatives have you considered?
Could make a separate example and will probably do that when I have a working version of the Gibbs kernel that also learns the lengthscale function.
Breaking changes
None