Will Tebbutt
Will Tebbutt
If someone has some time, a `softplus` Bijector would be cool to have.
[This](https://github.com/TuringLang/DistributionsAD.jl/blob/2d566ae83d0ca57966e1f69e3de469da8f9f10ec/src/zygote.jl#L78) definition is very optimistic about the things that it thinks that it can handle. In particular, it hijacks control away from [this method](https://github.com/willtebbutt/Stheno.jl/blob/e09858d408f775e3277594fd2ab5037c9ed60910/src/abstract_gp.jl#L197) in Stheno, and causes AD to...
Currently, we have ```julia approx_posterior(approximation, fx, y, u) ``` As pointed out by @st-- and @rossviljoen in [SparseGPs.jl](https://github.com/rossviljoen/SparseGPs.jl/issues/13), we should consider reducing this to a 3-arg function in which `approximation`...
**Summary** All of our work involving the LatentGP and friends type now resides in ApproximateGPs.jl. Moreover, literally nothing in this package uses that type. Given that we're going to need...
As discussed in https://github.com/JuliaGaussianProcesses/AbstractGPs.jl/issues/202 I don't want to merge this until we're ready to make another breaking release (I couldn't figure out how to do this without being breaking). edit:...
#201 made me realise that we can't currently handle input-specific likelihoods in our LatentGP. In particular, consider this line: https://github.com/JuliaGaussianProcesses/AbstractGPs.jl/blob/83dc67dda55ef9c1013131dfe04fc9babb9059c9/src/latent_gp.jl#L29 The likelihood passes straight through to the LatentFiniteGP, meaning that...
**Summary** What: Proposal for a vector-valued GP. Why: It might be a convenient thing to have around for some use-cases in which people really do think about their multi-output GPs...
The public API tests don't currently check the correctness of whatever is produced by `posterior`, we just check that it's some kind of `AbstractGP`. We could, however, check that e.g....
@theogf pointed out [here](https://github.com/JuliaGaussianProcesses/AbstractGPs.jl/pull/93#issuecomment-756046955) that we've got some missing examples, and @devmotion pointed out that some consolidation / extension of our existing examples would be helpful. Moreover, we've known for...
As discussed elsewhere, there is some consensus that it would be very helpful to provide some kind of user-facing front-end which sits on top of the low-level researcher-focused functionality provided...