Fix CompositeDataSet constructor

Author: RafaelArutjunjan

ext/InformationGeometryProfileLikelihoodJLext.jl

@@ -15,18 +15,16 @@ using Optimization: AbstractADType
 Constructs `LikelihoodProblem` struct from given `DataModel`.
 """
 function ProfileLikelihood.LikelihoodProblem(DM::AbstractDataModel, Mle::AbstractVector=MLE(DM);
-                adtype::AbstractADType=Optimization.AutoForwardDiff(), Domain::Union{HyperCube,Nothing}=GetDomain(DM),
-                syms::AbstractVector{<:Symbol}=Pnames(DM), maxval::Real=1e5,
+                adtype::AbstractADType=Optimization.AutoForwardDiff(), maxval::Real=1e5, Domain::Union{HyperCube,Nothing}=GetDomain(DM)∩FullDomain(length(Mle),maxval),
+                syms::AbstractVector{<:Symbol}=Pnames(DM), LogLikelihooodFn::Function=InformationGeometry.loglikelihood(DM),
                 lb=(!isnothing(Domain) ? Domain.L : fill(-maxval,length(Mle))), ub=(!isnothing(Domain) ? Domain.U : fill(maxval,length(Mle))),
                 cons=nothing, lcons=nothing, ucons=nothing, f_kwargs=NamedTuple(), prob_kwargs=NamedTuple(), kwargs...)
 
-    L = InformationGeometry.loglikelihood(DM)
-
     f_kwargs = (; adtype=adtype, cons=cons, f_kwargs...)
     # any(isfinite, lb) && any(isfinite, lb) && 
     prob_kwargs = (; lb=lb, ub=ub, prob_kwargs...)
 
-    ProfileLikelihood.LikelihoodProblem((p,x)->L(p), Mle; syms, f_kwargs=f_kwargs, prob_kwargs=prob_kwargs, kwargs...)
+    ProfileLikelihood.LikelihoodProblem((p,x)->LogLikelihooodFn(p), Mle; syms, f_kwargs=f_kwargs, prob_kwargs=prob_kwargs, kwargs...)
 end
 
 
@@ -35,7 +33,7 @@ end
 Computes profiles for given `idxs` up to given confidence threshold `Confnum` in units of `σ` via the `ProfileLikelihood.jl` package.
 """
 function ProfileLikelihood.profile(DM::AbstractDataModel, Confnum::Real=1; idxs=1:pdim(DM), N::Int=31, meth=Optim.LBFGS(), OptimMeth=meth, alg=OptimMeth, 
-                                    parallel::Bool=true, resolution=N, Domain::Union{HyperCube,Nothing}=GetDomain(DM), maxval::Real=1e5, Mle::AbstractVector=MLE(DM),
+                                    parallel::Bool=true, resolution=N, maxval::Real=1e5, Mle::AbstractVector=MLE(DM), Domain::Union{HyperCube,Nothing}=GetDomain(DM)∩FullDomain(length(Mle),maxval), 
                                     lb=(!isnothing(Domain) ? Domain.L : fill(-maxval,length(Mle))), ub=(!isnothing(Domain) ? Domain.U : fill(maxval,length(Mle))), kwargs...)
     prob = ProfileLikelihood.LikelihoodProblem(DM; Domain, maxval, lb, ub)
     sol = ProfileLikelihood.mle(prob, alg)
@@ -49,7 +47,7 @@ end
 Computes bivariate profiles for given `idxs` up to given confidence threshold `Confnum` in units of `σ`, where pairs of parameters are fixed at different values and the remaining parameters are re-optimized.
 """
 function ProfileLikelihood.bivariate_profile(DM::AbstractDataModel, Confnum::Real=1; idxs=nothing, N::Int=31, meth=Optim.LBFGS(), OptimMeth=meth, alg=OptimMeth, 
-                                        parallel::Bool=true, resolution=N, Domain::Union{HyperCube,Nothing}=GetDomain(DM), maxval::Real=1e5, Mle::AbstractVector=MLE(DM),
+                                        parallel::Bool=true, resolution=N, maxval::Real=1e5, Mle::AbstractVector=MLE(DM), Domain::Union{HyperCube,Nothing}=GetDomain(DM)∩FullDomain(length(Mle),maxval),
                                         lb=(!isnothing(Domain) ? Domain.L : fill(-maxval,length(Mle))), ub=(!isnothing(Domain) ? Domain.U : fill(maxval,length(Mle))), kwargs...)
     prob = ProfileLikelihood.LikelihoodProblem(DM; Domain, maxval, lb, ub)
     sol = ProfileLikelihood.mle(prob, alg)

src/DataStructures/CompositeDataSet.jl

@@ -138,7 +138,7 @@ InvCov::AbstractMatrix=Diagonal([1,2.]),
 logdetInvCov::Real=-Inf,
 WoundX::AbstractVector=[0.],
 SharedYdim::Val=Val(true),
-name::StringOrSymb=Symbol()) = CompositeDataSet(DSs, logdetInvCov, WoundX, SharedYdim, name)
+name::StringOrSymb=Symbol()) = CompositeDataSet(DSs, InvCov, logdetInvCov, WoundX, SharedYdim, name)
 
 
 Data(CDS::CompositeDataSet) = CDS.DSs

src/Subspaces.jl

@@ -436,9 +436,10 @@ Base.BigFloat(C::HyperCube) = HyperCube(BigFloat.(C.L), BigFloat.(C.U))
     intersect(Cubes::AbstractVector{<:HyperCube}) -> HyperCube
 Returns new `HyperCube` which is the intersection of the given `HyperCube`s.
 """
-function Base.intersect(A::HyperCube, B::HyperCube)
-    HyperCube(map(max, A.L, B.L), map(min, A.U, B.U))
-end
+Base.intersect(A::HyperCube, B::HyperCube) = HyperCube(map(max, A.L, B.L), map(min, A.U, B.U))
+Base.intersect(N::Nothing, C::HyperCube) = C
+Base.intersect(C::HyperCube, N::Nothing) = C
+
 function Base.intersect(Cubes::AbstractVector{<:HyperCube})
     LowerMatrix = [Cubes[i].L for i in eachindex(Cubes)] |> Unpack
     UpperMatrix = [Cubes[i].U for i in eachindex(Cubes)] |> Unpack