\name{bandwidthCV}
\alias{bandwidthCV}
\title{Bandwidth cross-validation}
\usage{
  bandwidthCV(well, fileParser = readYeastGrower,
    getWellIds = getWellIdsTecan,
    bandwidths = seq(0.5 * 3600, 10 * 3600, length.out = 30),
    nFold = 10, nWell = 100, cutoff = 0.95,
    calibration = identity, scaleY = log2)
}
\arguments{
  \item{well}{well dataframe. See
  \code{\link{wellDataFrame}}.}

  \item{fileParser}{Converts the file generated by the
  machine to proper R format. See
  \code{\link{readYeastGrower}} for details.}

  \item{getWellIds}{function or vector. If function its
  parameter is the return value of fileParser. It should
  return a vector containing the well ids (e.g. A01, A02,
  ...). You can set the well ids vector directly. See
  \code{\link{getWellIdsTecan}}.}

  \item{bandwidths}{vector of bandwidths to test on}

  \item{nFold}{\code{integer}. In how many parts is the
  sample divided for cross-validation?}

  \item{nWell}{\code{integer}. How many wells out of the
  well dataframe will be used for cross validation?}

  \item{cutoff}{\code{scalar} between 0 and 1. See
  details.}

  \item{calibration}{\code{function} or \code{list} of
  \code{functions}. If function, calibration is applied to
  all raw data. If list, the well dataframe must contain a
  column \code{machine}. Depending on that column the
  according function in the list is applied to the raw
  data. See details}

  \item{scaleY}{\code{function} applied to the calibrated
  data.}
}
\value{
  list with entries \item{bandwidth}{"optimal" bandwidth}
  \item{well}{well dataframe} \item{bandwidths}{tested
  bandwidths} \item{err2}{squared error}
  \item{err2std}{Standard deviation of squared error}
  \item{muStd}{Standard deviation of max growth rate}
  \item{oneStdOfMini}{bandwidths within one std of best}
}
\description{
  Perform cross-validation to detect optimal bandwidth
}
\details{
  This function needs a few minutes time. The "optimal"
  bandwidth is the largest bandwidth which is in 95\%
  (cutoff parameter) of all cases within one standard
  deviation of the best bandwidth. This should make the
  derivative of the fitted curve more robust. The raw
  values from the machine might not be directly optical
  densities (OD), which is needed to infer doubling time.
  Calibration functions for each machine can be provided to
  map raw values into OD using the \code{calibration}
  parameter.
}
\examples{
folder <- system.file("extdata", package="cellGrowth")
well <- wellDataFrame(file.path(folder, "plateLayout.txt"), file.path(folder,"machineRun.txt"))

## for a fast example, we use nWell = 1 here. Use a large number (default 100) for real life applications
bw <- bandwidthCV(well, nWell=1)
}
\author{
  Julien Gagneur and Andreas Neudecker
}