standardize data before pca

[gdkrmr]

Currently there is no way to standardize data when doing a PCA, only to centralize.

[wildart]

Usually, you would standardize data prior to fitting and do not mix two operations together. It's better not to mix two operations together.

FYI, there is data transformation PR pending: JuliaStats/StatsBase.jl#85

[gdkrmr]

What about the centering step, are you going to remove that?

[gdkrmr]

you should probably allow something like:

x = rand(3, 10)
z = fit(ZScoreTransform, x)
p = fit(PCA, transform(z, x))
transform((z, p), xnew)

or maybe even make the objects callable:

xnew |> z |>  p

and for the reconstruction:

y |> inv(p) |> inv(z)

[wildart]

What about the centering step, are you going to remove that?

No, not really.

Transformation and reconstructions are following:

x = rand(3, 10)
# standardize data
T = fit(ZScoreTransform, x)
xstd = transform(Z, x)
# perform PCA
M = fit(PCA, xstd)
ysub = transform(M, xstd)
# reconstruct to original space
ystd = reconstruct(M, ysub)
# reconstruct to original scale
y = reconstruct(T, ystd)
x ≈ y

Even thought pipeline X |> ZScoreTransform |> PCA looks appealing. It is beyond the scope of this package.

[gdkrmr]

What about the centering step, are you going to remove that?

No, not really.

This seems like an arbitrary choice.