cv.scca.Rd

% Generated by roxygen2: do not edit by hand
% Please edit documentation in R/scca.R
\name{cv.scca}
\alias{cv.scca}
\title{Cross-validated grid search over SCCA penalties}
\usage{
cv.scca(
  X,
  Y,
  lambda1 = seq(1e-06, 0.001, length = 5),
  lambda2 = seq(1e-06, 0.001, length = 5),
  ndim = 3,
  nfolds = 10,
  folds = NULL,
  opt.dim = 1,
  parallel = FALSE,
  init = TRUE,
  verbose = FALSE,
  ...
)
}
\arguments{
\item{X}{A numeric matrix. The use of PLINK datasets is currently not
supported.}

\item{Y}{A numeric matrix}

\item{lambda1}{A numeric vector of non-negative penalties}

\item{lambda2}{A numeric vector of non-negative penalties}

\item{ndim}{Integer. The number of dimensions to infer.}

\item{nfolds}{Integer. The number of cross-validation folds.}

\item{folds}{A numeric vector of integers. The indices of the folds to use}

\item{opt.dim}{Integer. Which dimension to select the optimal penalties
for.}

\item{parallel}{Logical. Whether to parallelise the cross-validation using
the foreach package.}

\item{init}{Logical. Whether to initialise SCCA with the SVD of X'Y.}

\item{verbose}{Logical. Whether to print more information.}

\item{...}{Other arguments that will be passed to \code{scca}.}
}
\value{
\code{cv.scca} returns an array containing the average
cross-validated canonical Pearson correlations between X and Y, as follows:
   Dimension 1: the ndim different canonical dimensions;
   Dimension 2: along the lambda1 penalties;
   Dimension 3: along the lambda2 penalties.
}
\description{
Cross-validated grid search over SCCA penalties
}
\details{
Note that the default penalties may not be appropriate for every dataset
and for some values the algorithm may not converge (especially for small
penalties).
}
\examples{

#######################
## HapMap3 chr1 example
data(hm3.chr1)
X <- scale2(hm3.chr1$bed)
n <- nrow(X)
m <- ncol(X)
k <- 5
B <- matrix(stats::rnorm(m * k), m, k)
Y <- X \%*\% B + stats::rnorm(n * k)

r <- cv.scca(X, Y,
   standx = "sd", standy = "sd", nfolds = 3, ndim = 2,
   lambda1 = seq(1e-3, 1e-1, length = 10),
   lambda2 = seq(1e-4, 0.5, length = 8)
)

par(mfrow = c(1, 2))
plot(r, dim = 1)
plot(r, dim = 2)

}