affinity propagation result is not consistent with sklearn in python

[xiuliren]

the two clustering results are different. Julia version did not do any clustering since the assignment is just the index of each object! My similarity matrix is too large to show here.

using Clustering

@time affinityPropResult = Clustering.affinityprop(similarityMatrix)

affinityPropResult.assignments

using PyCall

@pyimport sklearn.cluster as cl
af = cl.AffinityPropagation(affinity="precomputed")[:fit](similarityMatrix)

labels = af[:labels_]

The travis test also did not verify the correctness of the result.

[alyst]

Thanks for the report! You are most welcome to submit a fix. Otherwise, if you have a small (<100 entities, the smaller the better) reproducible example, I can look into that.

[xiuliren]

I double checked the results, they are same with some random tests. I did not add 1 for python result since python start from 0.

labels = af[:labels_] .+ 1

[xiuliren]

this issue still exist. The tests shows that it requires that the diagnal value should be the median value of the similarity matrix, otherwise the result is not consistent with python!

This is my test code:

using Distances
using Clustering
using LinearAlgebra
using Random
using Statistics 
Random.seed!(123)

    d = 10
    n = 44
    x = rand(d, n)
    S = -pairwise(Euclidean(), x, x)

    # set diagonal value to median value
#     S = S - diagm(0 => diag(S)) + median(S)*I

    R = affinityprop(S)

    k = length(R.exemplars)

using PyCall

@pyimport sklearn.cluster as cl
af = cl.AffinityPropagation(affinity="precomputed")[:fit]( S )
ref_assignments = af[:labels_] .+ 1

@assert randindex(R.assignments, ref_assignments)[2]==1.0