Set default preference to be median similarity of all pairs

Project.toml

@@ -8,10 +8,12 @@ Clustering = "aaaa29a8-35af-508c-8bc3-b662a17a0fe5"
 Distances = "b4f34e82-e78d-54a5-968a-f98e89d6e8f7"
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
 MLJModelInterface = "e80e1ace-859a-464e-9ed9-23947d8ae3ea"
+StatsBase = "2913bbd2-ae8a-5f71-8c99-4fb6c76f3a91"
 
 [compat]
 Clustering = "0.15"
 Distances = "0.9, 0.10"
 LinearAlgebra = "1.11.0"
 MLJModelInterface = "1.4"
+StatsBase = "0.34.3"
 julia = "1.6"

src/MLJClusteringInterface.jl

@@ -14,6 +14,7 @@ import MLJModelInterface: Continuous, Count, Finite, Multiclass, Table, OrderedF
 
 using Distances
 using LinearAlgebra
+using StatsBase
 
 # ===================================================================
 ## EXPORTS
@@ -224,11 +225,18 @@ function MMI.predict(model::AffinityPropagation, ::Nothing, X)
     # Compute similarity matrix using negative pairwise distances
     S = -pairwise(model.metric, Xarray, dims=2)
 
-    # Set preferences on diagonal if specified
-    if !isnothing(model.preference)
-        fill!(view(S, diagind(S)), model.preference)
+    diagonal_element = if !isnothing(model.preference)
+        model.preference
+    else
+        # Get the median out of all pairs of similarity, that is, values above
+        # the diagonal line.
+        # Such default choice is mentioned in the algorithm's wiki article
+        iuppertri = triu!(trues(size(S)),1)
+        median(S[iuppertri])
     end
 
+    fill!(view(S, diagind(S)), diagonal_element)
+
     result = Cl.affinityprop(
         S,
         maxiter=model.maxiter,