Rafael Schouten

Montreal is probably out for me. I might see some of you at JuliaCon - I should be talking about DynamicGrids.jl and Dispersal.jl there, which are also looking like cutting...

Maybe relevent, [DynamicGrids.jl](https://github.com/cesaraustralia/DynamicGrids.jl) can run direct convolutions on GPU using KernelAbstractions.jl. A simple direct windowing implementation is pretty easy to write as a toy example. But something faster or other...

Have you tried a much larger kernel with a larger image? These sizes are not favourable for FFTs or GPUs. I think `imfilter` chooses FFT when the kernel size is...

Im interested in adding GPU FFT to DynamicGrids.jl at some stage. It would be good if we could share a base package somehow, eventually... DG convolutions code is even more...

If people here are interested (@johnnychen94 ?), I'm extracting the neighborhoods from DynamicGrids.jl into a KernelAbstractions.jl based neighborhoods/windowing package. https://github.com/rafaqz/Neighborhoods.jl Its threaded using KernelAbstractions, and runs on GPU as well...

This is more of a Plots.jl than a GeoInterface.jl issue. Maybe we can get Makie.jl working with holes.

Another option is to use a sentinel approach with a wrapper array type, and replace the missing values on the fly in `getindex`. This may have less performance cost for...

Yeah, its a bit annoying because of the manual definitions required. But `Base.convert` is a bit of a special case. Its the generic way of converting things in Julia, and...

It seems the spec doesn't say either way. So mixed names are probably acceptable. Your problem comes from our maybe overly simplistic implementation of the Tables.jl interface. We can instead...

We can at least fix this error, we should check the feature property exists before calling `getproperty` on line 46 of features.jl. And return missing if not. Currently that check...