Merge pull request #267 from JuliaStats/ast/seeding_cleanup

Cleanup kmeans() seeding

src/seeding.jl

@@ -180,9 +180,8 @@ function initseeds!(iseeds::AbstractVector{<:Integer}, alg::KmppAlg,
             iseeds[j] = p
 
             # update mincosts
-            c = view(X, :, p)
             colwise!(metric, tmpcosts, X, view(X, :, p))
-            updatemin!(mincosts, tmpcosts)
+            mincosts .= min.(mincosts, tmpcosts)
             mincosts[p] = 0
         end
     end
@@ -211,7 +210,7 @@ function initseeds_by_costs!(iseeds::AbstractVector{<:Integer}, alg::KmppAlg,
             iseeds[j] = p
 
             # update mincosts
-            updatemin!(mincosts, view(costs, :, p))
+            mincosts .= min.(mincosts, view(costs, :, p))
             mincosts[p] = 0
         end
     end
@@ -240,21 +239,11 @@ function initseeds_by_costs!(iseeds::AbstractVector{<:Integer}, alg::KmCentralit
     k = length(iseeds)
     check_seeding_args(n, k)
 
-    # compute score for each item
-    coefs = vec(sum(costs, dims=2))
-    for i = 1:n
-        @inbounds coefs[i] = inv(coefs[i])
-    end
-
     # scores[j] = \sum_j costs[i,j] / (\sum_{j'} costs[i,j'])
-    #           = costs[i,j] * coefs[i]
-    scores = costs'coefs
+    scores = costs'vec(mapslices(inv∘sum, costs, dims=2))
 
     # lower score indicates better seeds
-    sp = sortperm(scores)
-    for i = 1:k
-        @inbounds iseeds[i] = sp[i]
-    end
+    copyto!(iseeds, 1, sortperm(scores), 1, k)
     return iseeds
 end
 

src/utils.jl

@@ -64,20 +64,6 @@ display_level(s::Symbol) = get(DisplayLevels, s) do
     throw(ArgumentError("Invalid option display=:$s ($(join(valid_vals, ", ", ", or ")) expected)"))
 end
 
-##### update minimum value
-
-function updatemin!(r::AbstractArray, x::AbstractArray)
-    n = length(r)
-    length(x) == n || throw(DimensionMismatch("Inconsistent array lengths."))
-    @inbounds for i = 1:n
-        xi = x[i]
-        if xi < r[i]
-            r[i] = xi
-        end
-    end
-    return r
-end
-
 function check_assignments(assignments::AbstractVector{<:Integer}, nclusters::Union{Integer, Nothing})
     nclu = nclusters === nothing ? maximum(assignments) : nclusters
     for (j, c) in enumerate(assignments)