Merge pull request #122 from alyst/more-0.7

More Julia v0.7/1.0 fixes

.travis.yml

@@ -5,6 +5,7 @@ os:
   - osx
 julia:
   - 0.7
+  - 1.0
   - nightly
 matrix:
   allow_failures:
@@ -14,6 +15,6 @@ notifications:
 # uncomment the following lines to override the default test script
 #script:
 #  - if [[ -a .git/shallow ]]; then git fetch --unshallow; fi
-#  - julia -e 'Pkg.clone(pwd()); Pkg.build("Clustering"); Pkg.test("Clustering"; coverage=true)'
+#  - julia -e 'using Pkg; Pkg.add(pwd()); Pkg.build("Clustering"); Pkg.test("Clustering"; coverage=true)'
 after_success:
-  - julia -e 'cd(Pkg.dir("Clustering")); Pkg.add("Coverage"); using Coverage; Coveralls.submit(Coveralls.process_folder())';
+  - julia -e 'using Pkg, Clustering; cd(joinpath(dirname(pathof(Clustering)), "..")); Pkg.add("Coverage"); using Coverage; Coveralls.submit(Coveralls.process_folder())';

README.md

@@ -2,8 +2,10 @@
 
 This package provides a set of algorithms for data clustering.
 
-[![0.5](http://pkg.julialang.org/badges/Clustering_0.5.svg)](http://pkg.julialang.org/?pkg=Clustering)
-[![0.6](http://pkg.julialang.org/badges/Clustering_0.6.svg)](http://pkg.julialang.org/?pkg=Clustering)
+[![0.6](http://pkg.julialang.org/badges/Clustering_0.6.svg)](http://pkg.julialang.org/?pkg=Clustering&ver=0.6)
+[![0.7](http://pkg.julialang.org/badges/Clustering_0.7.svg)](http://pkg.julialang.org/?pkg=Clustering&ver=0.7)
+[![1.0](http://pkg.julialang.org/badges/Clustering_1.0.svg)](http://pkg.julialang.org/?pkg=Clustering&ver=1.0)
+
 [![Travis](https://travis-ci.org/JuliaStats/Clustering.jl.svg?branch=master)](https://travis-ci.org/JuliaStats/Clustering.jl)
 [![Coveralls](https://coveralls.io/repos/github/JuliaStats/Clustering.jl/badge.svg?branch=master)](https://coveralls.io/github/JuliaStats/Clustering.jl?branch=master)
 
@@ -37,5 +39,3 @@ Pkg.add("Clustering")
 ## Resources
 
 **Documentation:** http://clusteringjl.readthedocs.org/en/latest/
-
-

REQUIRE

@@ -1,4 +1,4 @@
-julia 0.7.0-alpha
+julia 0.7
 Distances 0.3.1
 NearestNeighbors 0.0.3
 StatsBase 0.9.0

src/Clustering.jl

@@ -1,5 +1,3 @@
-__precompile__()
-
 module Clustering
     using Distances
     using NearestNeighbors
@@ -8,9 +6,7 @@ module Clustering
     using Printf
     using LinearAlgebra
     using SparseArrays
-    if VERSION >= v"0.7.0-beta.85"
-        using Statistics
-    end
+    using Statistics
 
     import Base: show
     import StatsBase: IntegerVector, RealVector, RealMatrix, counts

src/hclust.jl

@@ -375,14 +375,13 @@ function cutree(hclust::Hclust; k::Int=1,
         push!(clusters, new)
         i += 1
     end
-    all = vcat(clusters, nodes)
-    all = all[map(length, all) .> 0]
+    all = filter!(!isempty, vcat(clusters, nodes))
     ## convert to a single array of cluster indices
-    res = Vector{Int}(undef, nnodes)
+    res = fill(0, nnodes)
     for (i, cl) in enumerate(all)
-        res[cl] = i
+        res[cl] .= i
     end
-    res
+    return res
 end
 
 ## some diagnostic functions, not exported

src/kmeans.jl

@@ -86,7 +86,7 @@ function _kmeans!(
     dmat = pairwise(distance, centers, x)
     dmat = convert(Array{T}, dmat) #Can be removed if one day Distance.result_type(SqEuclidean(), T, T) == T
     update_assignments!(dmat, true, assignments, costs, counts, to_update, unused)
-    objv = w == nothing ? sum(costs) : dot(w, costs)
+    objv = w === nothing ? sum(costs) : dot(w, costs)
 
     # main loop
     t = 0
@@ -111,7 +111,7 @@ function _kmeans!(
         # update pairwise distance matrix
 
         if !isempty(unused)
-            to_update[unused] = true
+            to_update[unused] .= true
         end
 
         if t == 1 || num_affected > 0.75 * k
@@ -120,7 +120,7 @@ function _kmeans!(
             # if only a small subset is affected, only compute for that subset
             affected_inds = findall(to_update)
             dmat_p = pairwise(distance, centers[:, affected_inds], x)
-            dmat[affected_inds, :] = dmat_p
+            dmat[affected_inds, :] .= dmat_p
         end
 
         # update assignments
@@ -131,7 +131,7 @@ function _kmeans!(
         # compute change of objective and determine convergence
 
         prev_objv = objv
-        objv = w == nothing ? sum(costs) : dot(w, costs)
+        objv = w === nothing ? sum(costs) : dot(w, costs)
         objv_change = objv - prev_objv
 
         if objv_change > tol
@@ -297,18 +297,15 @@ function update_centers!(
     assignments::Vector{Int},       # in: assignments (n)
     to_update::Vector{Bool},        # in: whether a center needs update (k)
     centers::Matrix{T},             # out: updated centers (d x k)
-    cweights::Vector) where T<:AbstractFloat               # out: updated cluster weights (k)
+    cweights::Vector                # out: updated cluster weights (k)
+) where T<:AbstractFloat
 
     d::Int = size(x, 1)
     n::Int = size(x, 2)
     k::Int = size(centers, 2)
 
     # initialize center weights
-    for i = 1 : k
-        if to_update[i]
-            cweights[i] = 0.
-        end
-    end
+    cweights[to_update] .= 0.0
 
     # accumulate columns
     # accumulate_cols_u!(centers, cweights, x, assignments, weights, to_update)
@@ -323,13 +320,9 @@ function update_centers!(
                 rj = view(centers, :, cj)
                 xj = view(x, :, j)
                 if cweights[cj] > 0
-                    for i = 1:d
-                        @inbounds rj[i] += xj[i] * wj
-                    end
+                    @inbounds rj .+= xj * wj
                 else
-                    for i = 1:d
-                        @inbounds rj[i] = xj[i] * wj
-                    end
+                    @inbounds rj .= xj * wj
                 end
                 cweights[cj] += wj
             end
@@ -339,11 +332,7 @@ function update_centers!(
     # sum ==> mean
     for j = 1:k
         if to_update[j]
-            @inbounds cj::T = 1 / cweights[j]
-            vj = view(centers,:,j)
-            for i = 1:d
-                @inbounds vj[i] *= cj
-            end
+            @inbounds centers[:, j] .*= 1 / cweights[j]
         end
     end
 end

src/mcl.jl

@@ -1,7 +1,7 @@
 # MCL (Markov CLustering algorithm)
 
 """
-    immutable MCLResult <: ClusteringResult
+    struct MCLResult <: ClusteringResult
 
 Result returned by `mcl()`.
 """
@@ -21,16 +21,12 @@ end
 # `zero_tol` is a minimal value to consider as an element-to-cluster assignment
 function _mcl_clusters(mcl_adj::AbstractMatrix, allow_singles::Bool, zero_tol::Float64 = 1E-20)
     # remove rows containing only zero elements and convert into a mask of nonzero elements
-    el2clu_mask = mcl_adj[squeeze(sum(mcl_adj, dims=2), dims=2) .> zero_tol, :] .> zero_tol
+    el2clu_mask = mcl_adj[dropdims(sum(mcl_adj, dims=2), dims=2) .> zero_tol, :] .> zero_tol
 
     # assign cluster indexes to each node
     # cluster index is the index of the first TRUE in a given column
-    @static if VERSION >= v"0.7.0-beta.73"
-        _ms = mapslices(el_mask->isempty(el_mask) ? 0 : argmax(el_mask), el2clu_mask, dims=1)
-    else
-        _ms = mapslices(el_mask->isempty(el_mask) ? 0 : argmax(el_mask), el2clu_mask, 1)
-    end
-    clu_ixs = squeeze(_ms, dims=1)
+    _ms = mapslices(el_mask->isempty(el_mask) ? 0 : argmax(el_mask), el2clu_mask, dims=1)
+    clu_ixs = dropdims(_ms, dims=1)
     clu_sizes = zeros(Int, size(el2clu_mask, 1))
     unassigned_count = 0
     @inbounds for clu_ix in clu_ixs
@@ -155,7 +151,7 @@ function mcl(adj::AbstractMatrix{T};
     # initialize the MCL adjacency matrix by normalized `adj` weights
     mcl_adj = copy(adj)
     # normalize in columns
-    rmul!(mcl_adj, Diagonal(map(x -> x != 0.0 ?  1.0/x : x, squeeze(sum(mcl_adj, dims=1), dims=1))))
+    rmul!(mcl_adj, Diagonal(map(x -> x != 0.0 ?  1.0/x : x, dropdims(sum(mcl_adj, dims=1), dims=1))))
     mcl_norm = norm(mcl_adj)
     if !isfinite(mcl_norm)
         throw(OverflowError("The norm of the input adjacency matrix is not finite"))
@@ -176,15 +172,15 @@ function mcl(adj::AbstractMatrix{T};
 
         # normalize in columns
         rmul!(next_mcl_adj, Diagonal(map(x -> x != 0.0 ? 1.0/x : x,
-                                         squeeze(sum(next_mcl_adj, dims=1), dims=1))))
+                                         dropdims(sum(next_mcl_adj, dims=1), dims=1))))
 
         next_mcl_norm = norm(next_mcl_adj)
         if !isfinite(next_mcl_norm)
             @warn("MCL adjacency matrix norm is not finite")
             break
         end
         rel_delta = euclidean(next_mcl_adj, mcl_adj)/mcl_norm
-        (display == :verbose) && info("MCL iter. #$niter: rel.Δ=", rel_delta)
+        (display == :verbose) && @info("MCL iter. #$niter: rel.Δ=", rel_delta)
         (converged = rel_delta <= tol) && break
         # update (swap) MCL adjacency
         niter += 1

test/affprop.jl

@@ -6,7 +6,9 @@ using Clustering
 using LinearAlgebra
 using Random
 
-srand(34568)
+@testset "affinityprop() (affinity propagation)" begin
+
+Random.seed!(34568)
 
 d = 10
 n = 500
@@ -31,3 +33,4 @@ for i = 1:k
     @test R.counts[i] == count(==(i), R.assignments)
 end
 
+end