Merge pull request #1 from simonschoelly/v1.0_fix

Fixes for Julia v1.0 and v0.7

Author: sbromberger

.travis.yml

@@ -4,7 +4,8 @@ os:
   - linux
 #  - osx
 julia:
- - 0.6
+ - 0.7
+ . 1.0
  - nightly
 
 matrix:
@@ -35,4 +36,4 @@ git:
 #  - julia -e 'Pkg.clone(pwd()); Pkg.build("CommunityDetection"); Pkg.test("CommunityDetection"; coverage=true)'
 after_success:
   # push coverage results to Codecov
-  - julia -e 'cd(Pkg.dir("CommunityDetection")); Pkg.add("Coverage"); using Coverage; Codecov.submit(Codecov.process_folder())'
+  - julia -e 'cd(using Pkg; Pkg.dir("CommunityDetection")); Pkg.add("Coverage"); using Coverage; Codecov.submit(Codecov.process_folder())'

REQUIRE

@@ -1,3 +1,4 @@
-julia 0.6
-LightGraphs 0.9
+julia 0.7
+LightGraphs 1.1.0
 Clustering
+ArnoldiMethod

src/CommunityDetection.jl

@@ -1,7 +1,8 @@
-__precompile__(true)
 module CommunityDetection
 using LightGraphs
-import Clustering: kmeans
+using ArnoldiMethod: LR, SR
+using LinearAlgebra: I, Diagonal
+using Clustering: kmeans
 
 export community_detection_nback, community_detection_bethe
 
@@ -19,7 +20,9 @@ non-backtracking matrix of `g`.
 function community_detection_nback(g::AbstractGraph, k::Int)
     #TODO insert check on connected_components
     ϕ = real(nonbacktrack_embedding(g, k))
-    if k==2
+    if k == 1
+        c = fill(1, nv(g))
+    elseif k==2
         c = community_detection_threshold(g, ϕ[1,:])
     else
         c = kmeans(ϕ, k).assignments
@@ -54,8 +57,8 @@ See `Nonbacktracking` for details.
 """
 function nonbacktrack_embedding(g::AbstractGraph, k::Int)
     B = Nonbacktracking(g)
-    λ, eigv, conv = eigs(B, nev=k+1, v0=ones(Float64, B.m))
-    ϕ = zeros(Complex64, nv(g), k-1)
+    λ, eigv = LightGraphs.LinAlg.eigs(B, nev=k+1, which=LR())
+    ϕ = zeros(ComplexF32, nv(g), k-1)
     # TODO decide what to do with the stationary distribution ϕ[:,1]
     # this code just throws it away in favor of eigv[:,2:k+1].
     # we might also use the degree distribution to scale these vectors as is
@@ -84,19 +87,23 @@ Return a vector containing the vertex assignments.
 """
 function community_detection_bethe(g::AbstractGraph, k::Int=-1; kmax::Int=15)
     A = adjacency_matrix(g)
-    D = diagm(degree(g))
+    D = Diagonal(degree(g))
     r = (sum(degree(g)) / nv(g))^0.5
 
-    Hr = (r^2-1)*eye(nv(g))-r*A+D;
-    # Hmr = (r^2-1)*eye(nv(g))+r*A+D;
+    Hr = Matrix((r^2-1)*I, nv(g), nv(g)) - r*A + D;
+    #Hmr = Matrix((r^2-1)*I, nv(g), nv(g)) + r*A + D;
     k >= 1 && (kmax = k)
-    λ, eigv = eigs(Hr, which=:SR, nev=min(kmax, nv(g)))
-    q = findlast(x -> x<0, λ)
-    k > q && warn("Using eigenvectors with positive eigenvalues,
+    λ, eigv = LightGraphs.LinAlg.eigs(Hr, which=SR(), nev=min(kmax, nv(g)))
+
+    # TODO eps() is chosen quite arbitrarily here, because some of eigenvalues
+    # don't convert exactly to zero as they should. Some analysis could show
+    # what threshold should be used instead
+    q = something(findlast(x -> (x < -eps()), λ), 0)
+    k > q && @warn("Using eigenvectors with positive eigenvalues,
                     some communities could be meaningless. Try to reduce `k`.")
     k < 1 && (k = q)
-    k < 1 && return fill(1, nv(g))
-    labels = kmeans(eigv[:,2:k]', k).assignments
+    k <= 1 && return fill(1, nv(g))
+    labels = kmeans(collect(transpose(eigv[:,2:k])), k).assignments
     return labels
 end
 

test/runtests.jl

@@ -1,6 +1,9 @@
 using CommunityDetection
 using LightGraphs
-using Base.Test
+using LinearAlgebra: I, norm
+using ArnoldiMethod: LR
+using SparseArrays: sparse
+using Test
 
 """ Spectral embedding of the non-backtracking matrix of `g`
 (see [Krzakala et al.](http://www.pnas.org/content/110/52/20935.short)).
@@ -12,8 +15,8 @@ return : a matrix ϕ where ϕ[:,i] are the coordinates for vertex i.
 """
 function nonbacktrack_embedding_dense(g::AbstractGraph, k::Int)
     B, edgeid = non_backtracking_matrix(g)
-    λ,eigv,conv = eigs(B, nev=k+1, v0=ones(Float64, size(B,1)))
-    ϕ = zeros(Complex64, k-1, nv(g))
+    λ, eigv = LightGraphs.LinAlg.eigs(B, nev=k+1, which=LR())
+    ϕ = zeros(ComplexF32, k-1, nv(g))
     # TODO decide what to do with the stationary distribution ϕ[:,1]
     # this code just throws it away in favor of eigv[:,2:k+1].
     # we might also use the degree distribution to scale these vectors as is
@@ -30,6 +33,8 @@ function nonbacktrack_embedding_dense(g::AbstractGraph, k::Int)
     return ϕ
 end
 
+@testset "CommunityDetection" begin
+
 n = 10; k = 5
 pg = PathGraph(n)
 ϕ1 = CommunityDetection.nonbacktrack_embedding(pg, k)'
@@ -46,48 +51,72 @@ z = B * x
 @test norm(y-z) < 1e-8
 
 #check that matmat works and full(nbt) == B
-@test norm(nbt*eye(nbt.m) - B) < 1e-8
+@test norm(nbt*Matrix{Float64}(I, nbt.m, nbt.m) - B) < 1e-8
 
 #check that we can use the implicit matvec in nonbacktrack_embedding
 @test size(y) == size(x)
 ϕ2 = nonbacktrack_embedding_dense(pg, k)'
 @test size(ϕ2) == size(ϕ1)
 
 #check that this recovers communities in the path of cliques
-n=10
-g10 = CompleteGraph(n)
-z = copy(g10)
-for k=2:5
-    z = blkdiag(z, g10)
-    add_edge!(z, (k-1)*n, k*n)
-
-    c = community_detection_nback(z, k)
-    @test sort(union(c)) == [1:k;]
-    a = collect(n:n:k*n)
-    @test length(c[a]) == length(unique(c[a]))
-    for i=1:k
-        for j=(i-1)*n+1:i*n
-            @test c[j] == c[i*n]
+@testset "community_detection_nback(z, k)" begin
+    n=10
+    g10 = CompleteGraph(n)
+    z = copy(g10)
+    for k=2:5
+        z = blockdiag(z, g10)
+        add_edge!(z, (k-1)*n, k*n)
+
+        c = community_detection_nback(z, k)
+        @test sort(union(c)) == [1:k;]
+        a = collect(n:n:k*n)
+        @test length(c[a]) == length(unique(c[a]))
+        for i=1:k
+            cluster_range = (1:n) .+ (i-1)*n
+            @test length(unique(c[cluster_range])) == 1
         end
     end
+end
 
-    c = community_detection_bethe(z, k)
-    @test sort(union(c)) == [1:k;]
-    a = collect(n:n:k*n)
-    @test length(c[a]) == length(unique(c[a]))
-    for i=1:k
-        for j=(i-1)*n+1:i*n
-            @test c[j] == c[i*n]
+@testset "community_detection_bethe(z, k)" begin
+    n=10
+    g10 = CompleteGraph(n)
+    z = copy(g10)
+    for k=2:5
+        z = blockdiag(z, g10)
+        add_edge!(z, (k-1)*n, k*n)
+
+        c = community_detection_bethe(z, k)
+        @test sort(union(c)) == [1:k;]
+        a = collect(n:n:k*n)
+        @test length(c[a]) == length(unique(c[a]))
+
+        for i=1:k
+            cluster_range = (1:n) .+ (i-1)*n
+            @test length(unique(c[cluster_range])) == 1
         end
+
     end
+end
+
+@testset "community_detection_bethe(z)" begin
+    n=10
+    g10 = CompleteGraph(n)
+    z = copy(g10)
+    for k=2:5
+        z = blockdiag(z, g10)
+        add_edge!(z, (k-1)*n, k*n)
+
+        c = community_detection_bethe(z)
+        @test sort(union(c)) == [1:k;]
+        a = collect(n:n:k*n)
+        @test length(c[a]) == length(unique(c[a]))
 
-    c = community_detection_bethe(z)
-    @test sort(union(c)) == [1:k;]
-    a = collect(n:n:k*n)
-    @test length(c[a]) == length(unique(c[a]))
-    for i=1:k
-        for j=(i-1)*n+1:i*n
-            @test c[j] == c[i*n]
+        for i=1:k
+            cluster_range = (1:n) .+ (i-1)*n
+            @test length(unique(c[cluster_range])) == 1
         end
     end
 end
+
+end