DimensionMismatch: The size of a distance matrix ((1, 440)) doesn't match the length of assignment vector (440)

Specific Domains Machine Learning
1

How to fix this:

using Clustering
using VectorizedStatistics

function MyKmeans(Train::Vector{Float64}, k::Int64)
    the_mat = Train'
    model = kmeans(the_mat, k)

    return vmean(silhouettes(assignments(model), counts(model), the_mat))
end

julia> MyKmeans(rand(440), 2)
ERROR: DimensionMismatch: The size of a distance matrix ((1, 440)) doesn't match the length of assignment vector 
(440).
Stacktrace:
 [1] silhouettes(assignments::Vector{Int64}, counts::Vector{Int64}, dists::Adjoint{Float64, Vector{Float64}})    
   @ Clustering C:\Users\Julia\.julia\packages\Clustering\eBCMN\src\silhouette.jl:55
 [2] MyKmeans(Train::Vector{Float64}, k::Int64, kind::Silhouette)
   @ Main e:\newapproach.jl:155
 [3] top-level scope
   @ REPL[33]:1
1 Like
2

the_mat needs to be a distance matrix. So, it can be initialized as:

    the_mat = abs.(Train .- Train')

and then the function returns without an error.

Another option for getting the same distance matrix (in a more forward looking programming style):

    the_mat = pairwise(Euclidean(), Train)
1 Like
3

Thanks,
Should I give the result of abs.(Train .- Train') to kmeans function rather than Train'? Or just in case of silhouettes?

4

Just silhouettes function, kmeans likes Train'.
BTW, if you are new you might have missed out on the easy ‘help’ feature of the REPL: Just pressing ? and a function gives useful info for reputable packages.

1 Like