Wrapping issue around pi boundary

[dehann]

Samples close to wrap boundary pull the distribution onto the boundary:

using ApproxManifoldProducts, KernelDensityEstimatePlotting, Gadfly

PTS = Vector{Array{Float64, 2}}()

for th in range(-pi, pi, length=30)
         abs(th) < pi/2 ? continue : nothing
         push!(PTS, 0.5.*randn(1,100) .+ th)
end

BEL = BallTreeDensity[]

for pts in PTS
       push!(BEL, manikde!(pts, (:Circular,)))
end

col = getColorsByLength(length(BEL))
plotKDECircular(BEL, c=col)

• Example gets worse for wide beliefs.
• I was on branch feature/3Q19/basicproduct.
• NOTE, was also using upstream branch KernelDensityEstimate.jl:feature/3Q19/sensitivityentropy.
• Might be related to ensure circular products split balltree works for exact opposite kernels #8.

[dehann]

Looks like it might be a know "feature", user must now set explicitly from KernelDensityEstimate:

setForceEvalDirect!(true)

[dehann]