Ross Viljoen
Ross Viljoen
# 🐛 Bug I have been trying to modify the [Batch Independent Multioutput GP](https://docs.gpytorch.ai/en/latest/examples/03_Multitask_Exact_GPs/Batch_Independent_Multioutput_GP.html#Batch-Independent-Multioutput-GP) example to use a single kernel with shared hyperparameters across the independent tasks (as opposed to...
**NOT INTENDED FOR MERGE** This PR is meant to give an idea of what work is needed to allow AbstractGPs to run on GPU (although with no regard for performance)....
There are some example code snippets in the docs (particularly `docs/src/concrete_features.md`) which aren't run as doctests. As noted [here](https://github.com/JuliaGaussianProcesses/AbstractGPs.jl/pull/194#discussion_r686261164), some of these examples didn't even run, so it's probably worth...
Would it make sense to implement the internal API for `cov` & `var` and remove the definitions for `FiniteBLR`? It would add some currently missing methods like `cov(f, x, y)`
Relaxes the type signature of `elbo` to make it possible to extend
Supersedes #65, Closes #7 Implements pathwise sampling of functions from sparse GP posteriors from [1] [1] https://arxiv.org/abs/2002.09309 --- TODO: - Remove example before merging and move it to a new...
It would be nice to eventually add pathwise posterior sampling from [1]. Essentially, reimplement https://github.com/aterenin/SparseGaussianProcesses.jl but using AbstractGPs etc. [1] Wilson, James, et al. "Efficiently sampling functions from Gaussian process...
It takes a bit of work to figure out how the sparse GP posterior equations map to the actual code, so it would be good to write this up in...
The issues I know of that currently prevent GPU support are: - [ ] KernelFunctions uses hardcoded vectors for many kernels (e.g. ScaleTransform) https://github.com/JuliaGaussianProcesses/KernelFunctions.jl/issues/299 - [ ] `Distances.jl` doesn't work...