diff options
author | Suraj Yerramilli | 2015-01-29 19:32:25 +0530 |
---|---|---|
committer | Suraj Yerramilli | 2015-01-29 19:32:25 +0530 |
commit | bf782db25276d2b5e1cba9db0a0dc861cefe8f0b (patch) | |
tree | b2ee872c32c0ca93bf231cff5fe28f3687897d20 /R | |
parent | 27d4517777cac4ac7409a2c28728a1f8105dcda8 (diff) | |
download | SysID-R-code-bf782db25276d2b5e1cba9db0a0dc861cefe8f0b.tar.gz SysID-R-code-bf782db25276d2b5e1cba9db0a0dc861cefe8f0b.tar.bz2 SysID-R-code-bf782db25276d2b5e1cba9db0a0dc861cefe8f0b.zip |
added documentation for the reading functions
Diffstat (limited to 'R')
-rw-r--r-- | R/readData.R | 60 |
1 files changed, 54 insertions, 6 deletions
diff --git a/R/readData.R b/R/readData.R index 910bc94..24bb25d 100644 --- a/R/readData.R +++ b/R/readData.R @@ -1,4 +1,22 @@ -# Reading data from dataframes +#' Data input into a idframe object +#' +#' Read the contents of an file in table format into a \code{idframe} object. +#' +#' @param data a \code{data.frame} object +#' @param ninputs the number of input columns. (Default: 1) +#' @param type indicates the domain of the data (Default:\code{"time"}) +#' @param Ts sampling interval (Default: 1) +#' @param freqData a logical value indicating whether the file contains the list of +#' frequencies. If \code{TRUE}, they need to be present in the first column. +#' (Default: \code{idframe}) +#' +#' @details +#' +#' If \code{type="freq"} and \code{freqData = TRUE}, then the first column in the file +#' should contain the frequencies. +#' +#' @return an idframe object +#' @export read.idframe <- function(data,ninputs=1,type=c("time","freq")[1], Ts = 1,freqData=FALSE){ @@ -22,8 +40,38 @@ read.idframe <- function(data,ninputs=1,type=c("time","freq")[1], return(out) } -# Reading from table-formatted files -readtab.idframe <- function(file,header=TRUE,sep=",",ninputs=1, +#' Data input into a idframe object +#' +#' Read the contents of an file in table format into a \code{idframe} object. +#' +#' @param file the path to the file to read +#' @param sep the field separator character. Values on each line of the file are +#' separated by this character. (Default: \code{","}) +#' @param header a logical value indicating whether the first row corresponding to +#' the first element of the rowIndex vector contains the names of the variables. +#' (Default: \code{TRUE}) +#' @param ninputs the number of input columns. (Default: 1) +#' @param type indicates the domain of the data (Default:\code{"time"}) +#' @param Ts sampling interval (Default: 1) +#' @param freqData a logical value indicating whether the file contains the list of +#' frequencies. If \code{TRUE}, they need to be present in the first column. +#' (Default: \code{idframe}) +#' @param ... additional arguments to be passed to the \code{\link[utils]{read.table}} function +#' +#' @details +#' +#' The \code{readxlsx.idframe} function uses the \code{\link[utils]{read.table}} function, +#' provided by the \pkg{xlsx} package, to read data from an excel file and then calls the +#' \code{\link{read.idframe}} function to read the data into a idframe object +#' +#' +#' If \code{type="freq"} and \code{freqData = TRUE}, then the first column in the file +#' should contain the frequencies. +#' +#' @return an idframe object +#' @export +#' @seealso \code{\link[utils]{read.table}} +read.table.idframe <- function(file,header=TRUE,sep=",",ninputs=1, type=c("time","freq")[1],Ts = 1,freqData=FALSE,...){ # Read from file (default: csv file) @@ -44,7 +92,7 @@ readtab.idframe <- function(file,header=TRUE,sep=",",ninputs=1, #' the first element of the rowIndex vector contains the names of the variables. #' (Default: \code{TRUE}) #' @param ninputs the number of input columns. (Default: 1) -#' @param type ndicates the domain of the data (Default:\code{"time"}) +#' @param type indicates the domain of the data (Default:\code{"time"}) #' @param Ts sampling interval (Default: 1) #' @param freqData a logical value indicating whether the file contains the list of #' frequencies. If \code{TRUE}, they need to be present in the first column. @@ -53,7 +101,7 @@ readtab.idframe <- function(file,header=TRUE,sep=",",ninputs=1, #' #' @details #' -#' The \code{readxlsx.idframe} function uses the \code{\link[xlsx]{read.xlsx2}} function, +#' The \code{read.xlsx.idframe} function uses the \code{\link[xlsx]{read.xlsx2}} function, #' provided by the \pkg{xlsx} package, to read data from an excel file and then calls the #' \code{\link{read.idframe}} function to read the data into a idframe object #' @@ -68,7 +116,7 @@ readtab.idframe <- function(file,header=TRUE,sep=",",ninputs=1, #' @return an idframe object #' @export #' @seealso \code{\link[xlsx]{read.xlsx2}} -readxls.idframe <- function(file,sheetName,header=TRUE,ninputs=1, +read.xls.idframe <- function(file,sheetName,header=TRUE,ninputs=1, type=c("time","freq")[1],Ts = 1,freqData=FALSE,...){ require(xlsx) |