added more debug info & updated mdl precision, drop dependency

Project.toml

@@ -1,5 +1,5 @@
 name = "LMCLUS"
-version="0.4.1"
+version="0.4.2"
 uuid = "173e3a66-dd8a-5f2e-8a96-8b7ddc4d534d"
 
 [deps]
@@ -13,14 +13,15 @@ MultivariateStats = "6f286f6a-111f-5878-ab1e-185364afe411"
 Clustering = "aaaa29a8-35af-508c-8bc3-b662a17a0fe5"
 
 [compat]
-MultivariateStats = ">= 0.6.0"
+Clustering = "^0.14"
+MultivariateStats = "^0.7"
+StatsBase  = "^0.33"
 
 [extras]
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 DelimitedFiles = "8bb1440f-4735-579b-a4ab-409b98df4dab"
-Combinatorics = "861a8166-3701-5b0c-9a16-15d98fcdc6aa"
 Distributions = "31c24e10-a181-5473-b8eb-7969acd0382f"
 Serialization = "9e88b42a-f829-5b0c-bbe9-9e923198166b"
 
 [targets]
-test = ["Test", "DelimitedFiles", "Combinatorics", "Distributions", "Serialization"]
+test = ["Test", "DelimitedFiles", "Distributions", "Serialization"]

src/LMCLUS.jl

@@ -140,7 +140,7 @@ function find_manifold(X::AbstractMatrix{T}, index::Vector{Int},
     filtered = Int[]
     selected = copy(index)
     N = size(X,1) # full space dimension
-    best_manifold = Manifold{T}(params.max_dim, index)
+    best_manifold = Manifold{T}(params.min_dim, index)
     best_separation = Separation()
 
     sep_dim = params.min_dim
@@ -166,14 +166,14 @@ function find_manifold(X::AbstractMatrix{T}, index::Vector{Int},
                 # check if the adjusted basis provides better separation
                 previous_outdim = outdim(best_manifold)
                 adjustbasis!(best_manifold, X, adjust_dim=params.dim_adjustment, adjust_dim_ratio=params.dim_adjustment_ratio)
-                @debug "manifold: perform basis adjustment..." outdim=outdim(best_manifold) previous_outdim
                 origin, basis = mean(best_manifold), projection(best_manifold)
+                @debug "manifold: perform basis adjustment..." outdim=outdim(best_manifold) previous_outdim origin basis
                 idxs = points(best_manifold)
                 sep = find_separation(view(X, :, idxs), origin, basis, params, prngs[1])
             elseif state == :BOUND && params.bounded_cluster && size(best_manifold) > 0
                 # check if the bounded cluster provides better separation
-                @debug "manifold: separating within manifold subspace..."
                 origin, basis = mean(best_manifold), projection(best_manifold)
+                @debug "manifold: separating within manifold subspace..." origin basis
                 idxs = points(best_manifold)
                 sep = find_separation(view(X, :, idxs), origin, basis, params, prngs[1], ocss = true)
             else

src/mdl.jl

@@ -331,28 +331,28 @@ end
 
 "Calculate MDL for the manifold"
 function calculate(::Type{MT}, M::Manifold{T}, X::AbstractMatrix{T}, Pm::Int, Pd::Int;
-                   ɛ::T = 1e-2, tot::Int = 1000, tol = 1e-8) where {MT<:MethodType, T<:AbstractFloat}
+                   ɛ::T = 1e-3, tot::Int = 1000, tol = 1e-18) where {MT<:MethodType, T<:AbstractFloat}
     return modeldl(MT, M, X, Pm) + datadl(MT, M, X, Pd, ɛ, tot, tol)
 end
 
 "Calculate MDL for the clustering"
 function calculate(::Type{MT}, Ms::Vector{Manifold}, X::AbstractMatrix{T}, Pm::Int, Pd::Int;
-             ɛ::T=1e-2, tot::Int = 1000, tol = 1e-8) where {MT<:MethodType, T<:AbstractFloat}
+             ɛ::T=1e-3, tot::Int = 1000, tol = 1e-18) where {MT<:MethodType, T<:AbstractFloat}
     return sum(calculate(MT, m, X[:,points(m)], Pm, Pd, ɛ=ɛ, tot=tot, tol=tol) for m in Ms)
 end
 
 end
 
 # Compatibility
 function mdl(M::Manifold{T}, X::AbstractMatrix{T}, Pm::Int, Pd::Int;
-             dist::Symbol = :OptimalQuant, ɛ::T = 1e-2,
+             dist::Symbol = :OptimalQuant, ɛ::T = 1e-3,
              tot::Int = 1000, tol = 1e-8) where T<:AbstractFloat
     mdltype = Core.eval(MDL, dist)
     return MDL.calculate(mdltype, M, X, Pm, Pd, ɛ=ɛ)
 end
 
 function mdl(Ms::Vector{Manifold}, X::AbstractMatrix{T}, Pm::Int, Pd::Int;
-             dist::Symbol = :OptimalQuant, ɛ::T = 1e-2,
+             dist::Symbol = :OptimalQuant, ɛ::T = 1e-3,
              tot::Int = 1000, tol = 1e-8) where T<:AbstractFloat
     return sum(mdl(m,X,Pm,Pd,dist=dist,ɛ=ɛ,tot=tot,tol=tol) for m in Ms)
 end

test/lmclus_test.jl

@@ -1,9 +1,8 @@
 using LMCLUS
-using Combinatorics
 using Test
 using Statistics
 using DelimitedFiles
-import Random
+using Random
 
 @testset "Clustering" begin
 
@@ -54,7 +53,8 @@ import Random
 	@test sum(counts(res)) == size(ds, 1)
 
 	cnts = counts(res)
-	@testset "Label Match" for idxs in combinations(1:nclusters(res),2)
+	nc = nclusters(res)
+	@testset "Label Match" for idxs in [(i,j) for i in 1:nc, j in 1:nc if i < j]
 		i = idxs[1]
 		j = idxs[2]
 		@test length(symdiff(points(manifold(res,i)), points(manifold(res, j)))) == cnts[i] + cnts[j]