Variance and predictability describe different properties of a time series.
Principal Components Analysis finds components ordered by their variance, but
Predictable Components Analysis finds components ordered by their predictability. The first predictable component is more predictable than any other component that can be extracted from a field.
This is work in progress. The main function gKPrCA
finds components with maximal predictability in a linear or nonlinear sense.
Pkg.clone("https://github.com/Mattriks/PredictableComponents.jl")