st--

ndiagquad cannot cope with Y with several columns (lacks broadcasting capability)

6

The use of `gpflow.quadrature.ndiagquad()` in the quadrature fallback of `Likelihood.variational_expectations()` currently fails if you want a custom likelihood that manually handles additional columns in `Y` (e.g. known noise variance). Ideally,...

gpflow.conditionals.util.independent_interdomain_conditional not covered by tests

1

Noticed as part of #1219

ChangePoints kernel does not support active_dims, which means that it can only be used when the *entire* input is 1D, whereas it should be possible to have a changepoints kernel...

**BREAKING** Fix loss of structure in cov() and invcov() of AbstractMvNormal subtypes

4

Resolves #572. Distributions.jl has a lower bound of PDMats=0.10, in which PDMat and PDiagMat are subtypes of AbstractMatrix, so no need to call Matrix() anymore. What, if anything, do I...

Add JuliaFormatter to workflow

6

> yeah we should add JuliaFormatter to this. It would be good to merge as many PRs as possible before because it will create conflicts all over _Originally posted by...

incongruency in `params` of MvNormalCanon vs NormalCanon

1

In contrast to the (co)variance vs standard deviation inconsistency between Normal and MvNormal (https://github.com/JuliaStats/Distributions.jl/issues/584), NormalCanon and MvNormalCanon both are defined in terms of the natural parameters, \eta = \mu *...

_map -> Base.map

1

Again, curious to see what breaks

remove zygoterules

9

Curious to see what breaks...