preprocess_mean throws TypeError when handling Integer types

[Heladio-ac]

I found this error while working with integer data, it seems to occur with every type of integer.

using MultivariateStats

int_matrix = rand(Int, (2, 2))
fit(PCA, int_matrix)

Error:

ERROR: TypeError: in typeassert, expected Array{Int8,1}, got Array{Float64,1}
Stacktrace:
 [1] preprocess_mean(::Array{Int8,2}, ::Nothing) at /home/heladioac/.julia/packages/MultivariateStats/BYMwD/src/common.jl:25
 [2] #fit#10(::Symbol, ::Int64, ::Float64, ::Nothing, ::Function, ::Type{PCA}, ::Array{Int8,2}) at /home/heladioac/.julia/packages/MultivariateStats/BYMwD/src/pca.jl:148
 [3] fit(::Type{PCA}, ::Array{Int8,2}) at /home/heladioac/.julia/packages/MultivariateStats/BYMwD/src/pca.jl:138
 [4] top-level scope at none:0

The problem seems to come specifically from preprocess_mean trying to return a Vector{Int} when X contains integers, while the expression vec(mean(X, dims=2)) always returns an array of type Float64 in my computer.

preprocess_mean(X::AbstractMatrix{T}, m) where T<:Real =
    (m == nothing ? vec(mean(X, dims=2)) : m == 0 ? T[] :  m)::Vector{T}

[nalimilan]

Can you try removing ::Vector{T} and see whether it works? Replacing == nothing with === nothing should allow the compiler to infer the type correctly without this.