Adding StatsBase.jl weights methods to a function in Julia

Question

I am trying to add methods to a function similar to the example function my_function below. These methods should be dispatched when any of the subtypes of AbstractWeights from the StatsBase.jl package are passed.

I don't encounter any problem when writing the example function with Abstract and Primitive types from the Base package. E.g.

function my_function(v::Array{<:Real,1})
    return sum(v .* 3)/length(v)
end

function my_function(v::Array{<:Real,1}, w::Array{<:Real,1})
    return  sum(v .* w .* 3)/sum(w)
 end

v = [1,2,3]

w = [3,2,1]

my_function(v)
# 6.0
my_function(v, w)
# 5.0

However, when adding the methods for the types from StatsBase.jl I get MethodError errors:

using StatsBase

my_function(v::Array{<:Real,1}, w::Array{<:AbstractWeights,1}) = my_function(v,w)

my_function(v::Array{<:Real,1}, w::Array{Weights,1}) = my_function(v,w)

my_function(v, pweights(w))
# ERROR: LoadError: MethodError: no method matching my_function(::Vector{Int64}, ::ProbabilityWeights{Int64, Int64, Vector{Int64}})

my_function(v, weights(w))
# ERROR: LoadError: MethodError: no method matching my_function(::Vector{Int64}, ::Weights{Int64, Int64, Vector{Int64}})

How can I write the methods for the StatsBase.jl weights types above?

If the function worked well, the following should be true

my_function(v, w) == my_function(v, weights(w)) == my_function(v, pweights(w)) == my_function(v, fweights(w))
# true

Answer 1

There are several issues here:

• AbstractWeights is already a vector, so you do not need to wrap it in a vector;
• your implementation is recursive (the method calls itself).

So the way you should implement the code is:

my_function(v::Vector{<:Real}, w::AbstractWeights) = sum(v .* w .* 3)/sum(w)