R/drewcurves.R

In gbasulto/drewcurves: Produce fast Andrews plots with 'ggplot2'

##' Andrews Plot
##'
##' This function computes the Andrews plot of a dataframe.
##'
##' The default output is the Andrews plot generated with ggplot2.
##' @param df An nxp dataframe or matrix.
##' @param type Type of Fourier series to be computed. See vignette.
##' @param group Integer indicating the number of column that
##'     represents the group, if any.
##' @param resolution The number of regularly spaced points where the
##'     Fourier series will be evaluated.
##' @param return_dataframe If TRUE, the function will return the
##'     dataframe used to generate the Andrews plot rather than the
##'     plot itself.
##' @param minmax_scaling If TRUE, it substract the mimimum and divide
##'     by the length of the range of every column.
##' @param xlim A vector of size two with the limits where the Fourier
##'     series will be evaluated. The default limits are are -pi and
##'     pi.
##' @return A ggplot object or a dataframe (depending on the parameter
##'     'return_dataframe').
##' @author Guillermo Basulto-Elias
##' @export 
drewcurves <- function (df, type = 1, group = NULL, resolution = 100,
                        return_dataframe = FALSE,
                        minmax_scaling = TRUE, xlim = c(-pi, pi)) {

    ## Extract column names
    var_names <- colnames(df)

    ## Read df (which could be a matrix) as dataframe
    df <- dplyr::tbl_df(df)
    
    ## Remove group variable if it is provided
    if (!is.null(group)) {
        grp <- df[[group]]              # As vector
        if (is.character(group)) {
            ## Change group to numeric if necessary.
            group <- which(var_names == group) 
        }
        grp_name <- names(var_names[group])
        df <- df[, -group]
    }

    ## Convert the rest of the variables to 
    df <- as.matrix(df)

    ## Scale if required.
    if (minmax_scaling) df <- minmax(df)

    ## Compute Fourier Series
    t <- seq(xlim[1], xlim[2], length.out = resolution)
    fourier_series <- compute_fourier_series(df, t = t, type = type)
    
    
    ## Create 'long format' dataframe with no group (at this point).
    key <- value <- NULL                      # Set to null kill NOTE
    out <- as.data.frame(fourier_series) # Cast to dataframe
    out <- dplyr::mutate(out, t = t)     # Add time column
    out <- tidyr::gather(out, key, value, -t)  # Cast to long format
                                        # dataframe.

    ## Add group if it exists.
    if (!is.null(group)) {
        group_name <- var_names[group]
        out <- dplyr::mutate(out, group = rep(grp, each = resolution))
        new_names <- names(out)
        cat("Group name = ", group_name, "\n")
        new_names[length(new_names)] <- group_name
        out <- stats::setNames(out, new_names)
    }

    ## Return dataframe instead of plot?
    if (return_dataframe) return (out)

    ## Generate NULL plot.
    out <- ggplot2::ggplot(data = out,
                           ggplot2::aes(t, value, group = key))
    
    ## If no group column is given, it returns a plot with no color
    ## coding.
    if (is.null(group)) {
        out <- out + ggplot2::geom_line()
    } else {
        out <- out +
            ggplot2::geom_line(ggplot2::aes_string(color = group_name))
        }
    
    out
}