WIP: Refactor metric & broken lloyd

Author: PyDataBlog

src/kmeans.jl

@@ -130,7 +130,7 @@ design matrix(x), centroids (centre), and the number of desired groups (k).
 
 A Float type representing the computed metric is returned.
 """
-function sum_of_squares(metric, containers, x, labels, centre, weights, r, idx)
+function sum_of_squares(containers, x, labels, centre, weights, metric, r, idx)
     s = zero(eltype(x))
 
     @inbounds for i in r

src/lloyd.jl

@@ -30,7 +30,7 @@ function kmeans!(alg::Lloyd, containers, X, k, weights;
 
     # Update centroids & labels with closest members until convergence
     while niters <= max_iters
-        @parallelize n_threads ncol chunk_update_centroids(alg, metric, containers, centroids, X, weights)
+        @parallelize n_threads ncol chunk_update_centroids(alg, containers, centroids, X, weights, metric)
         collect_containers(alg, containers, centroids, n_threads)
         J = sum(containers.J)
 
@@ -49,7 +49,7 @@ function kmeans!(alg::Lloyd, containers, X, k, weights;
         niters += 1
     end
 
-    @parallelize n_threads ncol sum_of_squares(metric, containers, X, containers.labels, centroids, weights)
+    @parallelize n_threads ncol sum_of_squares(containers, X, containers.labels, centroids, weights, metric)
     totalcost = sum(containers.sum_of_squares)
 
     # Terminate algorithm with the assumption that K-means has converged
@@ -104,7 +104,7 @@ function create_containers(::Lloyd, X, k, nrow, ncol, n_threads)
 end
 
 
-function chunk_update_centroids(::Lloyd, metric, containers, centroids, X, weights, r, idx)
+function chunk_update_centroids(::Lloyd, containers, centroids, X, weights, metric, r, idx)
     # unpack containers for easier manipulations
     centroids_new = containers.centroids_new[idx]
     centroids_cnt = containers.centroids_cnt[idx]
@@ -118,7 +118,7 @@ function chunk_update_centroids(::Lloyd, metric, containers, centroids, X, weigh
         min_dist = distance(metric, X, centroids, i, 1)
         label = 1
         for j in 2:size(centroids, 2)
-            dist = distance(metric, X, centroids, i, j)
+            dist = distance(metric, X, centroids, i, 1)
             label = dist < min_dist ? j : label
             min_dist = dist < min_dist ? dist : min_dist
         end

src/seeding.jl

@@ -9,7 +9,7 @@ centroid vector `y[:, i]`.
 UnitRange argument `r` select subarray of original design matrix `x` that is going
 to be processed.
 """
-function chunk_colwise(metric, target, x, y, i, weights, r, idx)
+function chunk_colwise(target, x, y, i, weights, metric, r, idx)
     T = eltype(x)
     @inbounds for j in r
         dist = distance(metric, x, y, j, i)
@@ -18,6 +18,7 @@ function chunk_colwise(metric, target, x, y, i, weights, r, idx)
     end
 end
 
+
 """
     smart_init(X, k; init="k-means++")
 
@@ -56,7 +57,7 @@ function smart_init(X, k, n_threads = Threads.nthreads(), weights = nothing,
         distances = fill(T(Inf), ncol)
 
         # compute distances from the first centroid chosen to all the other data points
-        @parallelize n_threads ncol chunk_colwise(metric, distances, X, centroids, 1, weights)
+        @parallelize n_threads ncol chunk_colwise(distances, X, centroids, 1, weights, metric)
         distances[rand_idx] = zero(T)
 
         for i = 2:k
@@ -72,7 +73,7 @@ function smart_init(X, k, n_threads = Threads.nthreads(), weights = nothing,
             i == k && break
 
             # compute distances from the centroids to all data points
-            @parallelize n_threads ncol chunk_colwise(metric, distances, X, centroids, i, weights)
+            @parallelize n_threads ncol chunk_colwise(distances, X, centroids, i, weights, metric)
 
             distances[r_idx] = zero(T)
         end

test/test01_distance.jl

@@ -10,7 +10,7 @@ using Test
     r = fill(Inf, ncol)
     n_threads = 1
 
-    @parallelize n_threads ncol chunk_colwise(Euclidean(), r, X, y, 1, nothing)
+    @parallelize n_threads ncol chunk_colwise(r, X, y, 1, nothing, Euclidean())
     @test all(r .≈ [0.0, 13.0, 25.0])
 end
 
@@ -21,7 +21,7 @@ end
     r = fill(Inf, ncol)
     n_threads = 2
 
-    @parallelize n_threads ncol chunk_colwise(Euclidean(), r, X, y, 1, nothing)
+    @parallelize n_threads ncol chunk_colwise(r, X, y, 1, nothing, Euclidean())
 
     @test all(r .≈ [0.0, 13.0, 25.0])
 end