`rand(MvNormalCanon(Q))` gives bad realizations

ElOceanografo / SPDEPrecisionMatrices.jl

Sparse precision matrices for random spatial fields using SPDE approximation

MIT License

1 stars 1 forks source link

`rand(MvNormalCanon(Q))` gives bad realizations #2

Open jkbest2 opened 3 years ago

jkbest2 commented 3 years ago

This is an upstream problems with JuliaStats/PDMats.jl#120 (also JuliaStats/Distributions.jl#1200), going back to your Discourse thread. I haven't checked whether this also affects e.g. logpdf, but I suspect it does.

Just a reminder that once it's fixed we'll want to put a lower compat bound on either of these packages that we depend on.

ElOceanografo commented 3 years ago

Fortunately, it doesn't affect logpdf. Roger on the compat bounds.

ElOceanografo commented 3 years ago

Will be fixed by https://github.com/JuliaStats/Distributions.jl/pull/1218