JaxGaussianProcesses / GPJax

Gaussian processes in JAX.
https://docs.jaxgaussianprocesses.com/
Apache License 2.0
436 stars 51 forks source link

bug: Default Parameter in PoweredExponential #427

Open ChrisBoettner opened 8 months ago

ChrisBoettner commented 8 months ago

Bug Report

GPJax version:

0.8.0

Current behavior:

Hi, I'm new to GPJax (and Jax) in general, so please excuse me if I'm being naive here, but I think there is some weird behavior in the default parameter of the PoweredExponential kernel. Specifically, the power gets set to 1 by default, with a default sigmoid bijector. This leads to the parameter to be inf, when running model.unconstrain(), which makes the fit fail with no obvious reason. It's definitely something that can be fixed very easily by setting a different initial value to power, but I doubt this is the intended behaviour.

Best wishes, Chris

github-actions[bot] commented 1 week ago

This issue has been marked as stale because it has been open for 7 days with no activity.