Package for KPCA

Specific Domains Statistics
1

Hi, I am a newbie to Julia. I am looking for a package for KPCA please. I tried using MultivariateStats, unfortunately, ran into the below issue

https://github.com/JuliaStats/MultivariateStats.jl/issues/164

I couldn’t find any other packages which would offer similar functionality - any ideas?

thanks
Rohit

2

Hi

I can’t reproduce your issue with julia 1.7.1.

Can you run the following in a fresh julia session and report the full error you get?

using Pkg; Pkg.activate(temp = true); Pkg.add.(["MultivariateStats", "RDatasets"])
using MultivariateStats, RDatasets

# load iris dataset
iris = dataset("datasets", "iris")

Xtr = Matrix(iris[!,1:4])'
Xtr_labels = convert(Array, iris[!,5])

model = fit(
KernelPCA,
Xtr;
kernel = (x, y) -> x'y,
maxoutdim = 3,
inverse = true,
)
1 Like
3

@jbrea : Thanks mate. it works fine now. I should have tried with Julia 1.7.1.

ta!

1 Like
4

You may also try package Jchemo.jl

using Jchemo, JchemoData, 
using JLD2, CairoMakie, StatsBase

path_jdat = dirname(dirname(pathof(JchemoData)))
db = joinpath(path_jdat, "data/iris.jld2")
@load db dat
pnames(dat)
summ(dat.X)

X = dat.X[:, 1:4]
n = nro(X)

ntrain = 120
s = sample(1:n, ntrain; replace = false)
Xtrain = X[s, :]
Xtest = rmrow(X, s)

nlv = 3 ; gamma = 1e-2
fm = kpca(Xtrain; nlv = nlv, gamma = gamma) ;
pnames(fm)
fm.T             # Scores
fm.P             # Loadings
fm.T' * fm.T
fm.P' * fm.P

group = recodcat2int(dat.X.species[s])
plotxy(fm.T[:, 1], fm.T[:, 2], group).f

Jchemo.transform(fm, Xtest)  # projection of new samples

res = Base.summary(fm) ;
pnames(res)
res.explvarx

Other kernel methods are available, for instance for kernel PLS.

1 Like