Daniel Dodd
Daniel Dodd
The guide for the custom likelihood is rather thorough in explaining keys concepts -- but an explicit example showing some actual code would be nice here.
Though for many applications in ML / Spatial statistics, the Matern kernel with smoothness parameter of 1/2, 3/2, 5/2, is sufficient, it is important in other applications such as oceanography...
We have a convenient `fit` function to train GPs against objectives. It would be good though e.g., in the regression notebook to show a simple python for loop, and a...
# Feature Request Introduce Vecchia approximation that is popular in spatial statistics https://en.wikipedia.org/wiki/Vecchia_approximation Like our `RFF` kernel, have a `Vecchia` kernel with a Vecchia compute engine to do the approximation....
Post #370, akin to kernel / likelihood quadrature computation, would be good to have on the top level default methods for these defined on the `Prior` that say what solver...
It seems possible to convert tensor flow probability models to NumyPyro models, so would be nice to have a notebook demonstrating this https://num.pyro.ai/en/0.7.0/_modules/numpyro/contrib/tfp/distributions.html Perhaps, without requiring this as a dependency,...
In relation to this #326, would be nice if there was some API for a user to build linear model via defining some basis functions (that may have hyperparams), which...
The proposal here is to extend and structure #195 as a module than _just_ a method on the GP. It is still nice to have quick access to this, but...
Thankfully beartype told me there was an issue with my code during refactoring to CoLA #370, but we need to be really careful with shapes in the integrator for quadrature....
Post #370, migrate `cross_covariance` in kernel computations be a LinearOperator rather than a dense array (this would match the signature of gram). This is beneficial in sparse situations.