cvFromInterpolsvd: cross-validation from interpolsvd_em algorithm
In proto4426/ValUSunSSN: Val-U-Sun project functions

Description Usage Arguments Value Author(s) Examples

It aims to compare

The three tuneable (hyper)parameters are :

ncomp
nsmo
nembed

1
2
3

cvFromInterpolsvd(x, comp_max = 4, nembed = 2, nsmo = 81,
  method = "splines", niter = 5, min_keep_frac = 0.1, seed = 123,
  brow = F)

`x`	the numeric matrix of SSN with days in rows and stations in columns for which we want to compute the cross-validation based on inteprolsvd_em()
`comp_max`	Maximum number of component we want to test
`method`	Controls the method used for interpolation. Thus either "smooth_gauss" or "splines" is allowed so far.
`niter`	The number of iterations of the algorithm.
`min_keep_frac`	Real between 0 and 1 controlling the station that has the highest number of NA. This will determine the number of fynthetic gaps
`seed`	Controls the seed of the random index sampling pf the synthetic gaps

A grid of 2 ggplot representing the error and the cross-validation error with respect to the number of components retained from the SVD. And a list containing the following elements:

y.filled
errorByComp
CVerrorByComp

Antoine Pissoort, antoine.pissoort@student.uclouvain.be

library(ValUSunSSN)
data("data.mat2.fin")
y <- data.mat2.fin

y_obsToNA <- cvFromInterpolsvd(x = y, comp_max = 10,
                              niter = 30, min_keep_frac = 0.2)