[TESTING] improve tree testing of factors per clique

[dehann]

Tasks,

• 1. Second example from Confirm clique sub-graphs only use priors once  #458 (resolved by ex2 test from #458 #516)
• 2. Add more Factor checks to tree tests (partially resolved ex2 test from #458 #516)
• 3. Additional consolidation of previous half-baked tests
• 4. Check to ensure Previous Factor and Subgraph Factors match (partial ex2 test from #458 #516 )

1: Include second example from #458

Example 2 from #458 (comment) should be included here:

IncrementalInference.jl/test/testCliqueFactors.jl

Line 100 in f6f164e

@warn "Test for Example 2 from 458 must still be coded."

Note that Example 1 is already built just above:

IncrementalInference.jl/test/testCliqueFactors.jl

Line 9 in f6f164e

@testset "Test clique factors, #458 Example 1" begin

2. Add more Factor checks to tree tests

Many tests on trees but these do not yet check which factors are in which clique (as was done in example above):

IncrementalInference.jl/test/testJunctionTreeConstruction.jl

Line 19 in f6f164e

@testset "Test Caesar Ring 1D symbolic tree construction" begin

IncrementalInference.jl/test/testJunctionTreeConstruction.jl

Line 11 in f6f164e

vo = [:l1, :l2, :x1, :x2, :x3]

IncrementalInference.jl/test/testJunctionTreeConstruction.jl

Line 75 in f6f164e

eo = [:x1; :l3; :l1; :x5; :x2; :l2; :x4; :x3]

3. Additional consolidation of previous half-baked tests

previous Kaess test case:

IncrementalInference.jl/test/testBayesTreeiSAM2Example.jl

Line 53 in f6f164e

# Michael reference -- x2->x1, x2->x3, x2->x4, x2->l1, x4->x3, l1->x3, l1->x4

Can readily be merged with this newer and improved testing here:

IncrementalInference.jl/test/testJunctionTreeConstruction.jl

Line 11 in f6f164e

vo = [:l1, :l2, :x1, :x2, :x3]

4. Check to ensure Previous Factor and Subgraph Factors match

This "newer" factor per clique test:

IncrementalInference.jl/test/testCliqueFactors.jl

Lines 62 to 68 in f6f164e

	C3_fg = buildCliqSubgraph(fg, tree, getCliq(tree, :x0) )
	# drawGraph(C3_fg, show=true)
	C3_fcts = [:x0l0f1;:x0l1f1;:x0x1f1;:x0f1]
	@test intersect(ls(C3_fg), [:x0; :x1; :l0; :l1]) |> length == 4
	@test intersect(lsf(C3_fg), C3_fcts) |> length == length(C3_fcts)

does not yet cover the previous main use case where factors were separately searched and described in the BayesTreeNodeData.potentials field:

IncrementalInference.jl/src/JunctionTree.jl

Lines 810 to 827 in f6f164e

	fctsyms = getFactorsAmongVariablesOnly(dfg, varlist, unused=true )
	# filter only factors connected to frontals (for upward)
	frtfcts = union(map(x->ls(dfg, x), getCliqFrontalVarIds(cliq))...)
	fctsyms = intersect(fctsyms, frtfcts)
	getData(cliq).potentials = fctsyms
	getData(cliq).partialpotential = Vector{Bool}(undef, length(fctsyms))
	fcts = map(x->getFactor(dfg, x), fctsyms)
	getData(cliq).partialpotential = map(x->isPartial(x), fcts)
	for fct in fcts
	solverData(fct).potentialused = true
	end
	@info "finding all frontals for down WIP"
	ffctsyms = getCliqFactorsFromFrontals(dfg, cliq, Symbol[], inseparator=false, unused=false, solvable=solvable)
	# fnsyms = getCliqVarsWithFrontalNeighbors(csmc.dfg, csmc.cliq)
	getData(cliq).dwnPotentials = ffctsyms
	getData(cliq).dwnPartialPotential = map(x->isPartial(getFactor(dfg,x)), ffctsyms )

Clearly some consolidation is required between new CSM code, and previous .potentials list approach -- but that can be done later after sufficient tests are in place that ensure a few test trees are being built properly. Starting with a simple tests that compares buildCliqSubgraph with existing BTND.potentials is probably easiest way to go.

[dehann]

Task 1 ( and partial Task 4) is completed by #516