Update calcCovarianceBasic

[Affie]

ApproxManifoldProducts.jl/src/CommonUtils.jl

Lines 63 to 76 in 1169d4c

	# Returns the covariance (square), not deviation
	function calcCovarianceBasic(M::AbstractManifold, ptsArr::Vector{P}) where P
	#TODO double check the maths,. it looks like its working at least for groups
	μ = mean(M, ptsArr)
	Xcs = vee.(Ref(M), Ref(μ), log.(Ref(M), Ref(μ), ptsArr))
	Σ = mean(Xcs .* transpose.(Xcs))
	@debug "calcCovarianceBasic" μ
	@debug "calcCovarianceBasic" Σ
	# TODO don't know what to do here so keeping as before, #FIXME it will break
	# a change between this and previous is that a full covariance matrix is returned
	msst = Σ
	msst_ = 0 < sum(1e-10 .< msst) ? maximum(msst) : 1.0
	return msst_
	end

Manifolds.jl now has a cov function that this can be updated to use: cov(M, ptsArr; basis, kwargs... )

We should also consider changing the name to reflect the fact that it is a variance and not a covariance. We can then also use the manifolds var function.

[dehann]

I just ran into this again, going to add this to the v0.6.1 planning.

[Affie]

See

https://github.com/JuliaRobotics/IncrementalInference.jl/blob/d8c82570bdf7ae0c092f391f66deeff4fd9fab1b/src/VariableStatistics.jl#L13-L19

[dehann]

quick work to close this:

• consolidation towards Manifolds.jl cov #220

[dehann]

Ongoing tech debt is IIF.nullhypo code should spread on each dimension specifically, not just reduced to one number as these legacy function. We can open this as a separate ongoing issue, and close this one faster.

Also see:

• calcStandardDev