R/ezr_correlations.R

In jmp1989/easyr: Helpful wrappers for common EDA, Data Manipulation, & Modeling

#' Correlation
#'
#' Correlation function.
#'
#' @param dataset dataframe
#' @param target optional.  The target y value you care about, providing cleaner, streamlined results
#' @param only_variables  optional.  Only compute correlation with these variables
#' @param exclude_variables optional.  Do not compute correlation with these variables
#' @param only_numeric Default TRUE.  Only numeric values are considered.  Right now the rest is not yet implemented.
#' @param corr_method Default 'pearson'.  Other valid values are 'spearman' and 'kendall'
#' @param return_raw_numbers Default is TRUE. Return the raw correlation matrix
#' @param return_plot Default is TRUE.  Makes a correlation plot.
#' @param title Default is NULL.  Only applies when you are making a plot.
#'
#' @return If both return_raw_numbers and return_plot are true then it will return a list of both.  Otherwise just returns what was indicated.
#' @export
#'
#' @examples

ezr.correlation = function (dataset, target = NULL, only_variables = NULL, exclude_variables = NULL,
only_numeric = TRUE, corr_method = "pearson", return_raw_numbers = TRUE,
return_plot = TRUE, title = NULL, n_limit_using_target=15) {
    if (only_numeric == TRUE) {
        dataset = dataset %>% dplyr::select_if(base::is.numeric)
    }
    if (is.null(only_variables) == FALSE) {
        retain_vars = intersect(names(dataset), only_variables)
    }else {
        retain_vars = names(dataset)
    }
    retain_vars = setdiff(retain_vars, exclude_variables)
    dataset = dataset %>% select(retain_vars)


    if (is.null(target)==FALSE){
    result = round(stats::cor(dataset, y = dataset[[target]], use = "pairwise.complete.obs",
    method = corr_method), 2)
    } else {
        result = round(stats::cor(dataset, y =target, use = "pairwise.complete.obs",
                                  method = corr_method), 2)
    }
    # removing leasting

    if(is.null(target)==TRUE){
        plt =  ggcorrplot::ggcorrplot(result, hc.order = TRUE, type = "lower",
        outline.col = "white", lab = TRUE)+theme_classic()+theme(axis.title = element_text(face = "bold", size = rel(1)),
        plot.title = element_text(face = "bold", size = rel(1.25)))+labs(title=title, y=element_blank(), x=element_blank())
    }



    if (is.null(target) == FALSE) {
        plot_data = as.data.frame(result)
        plot_data = tibble::rownames_to_column(plot_data, var = "feature")
        names(plot_data) = c("feature", "correlation")
        plot_data = plot_data %>% filter(feature != target)
        plot_data = plot_data %>% arrange(desc(correlation))
        plot_data = plot_data %>% mutate(abs_value = abs(correlation))
        if (is.null(title) == TRUE) {
            title = paste0("Correlation with ", target)
        }
        plt = plot_data %>% ggplot(aes(feature)) + geom_bar(aes(y = correlation),
        stat = "identity") + scale_y_continuous(limits = c(-1,1),breaks= seq(-1,1,0.2)) + labs(y=element_blank(), title=title) +theme_Publication()+scale_x_discrete(limits = c(plot_data$feature))


        plot_data = plot_data %>% mutate(label_for_plt = ifelse(correlation<0, correlation-.05, correlation+.05))



        plt = plt + geom_text(stat='identity', data=plot_data,
                  aes(label=correlation, y = label_for_plt))



n_limit_using_target

    }

    if (return_plot == TRUE & return_raw_numbers == TRUE) {
        result = list(correlation_raw_numbers = result, plot = plt)
    } else if(return_raw_numbers==TRUE & return_plot==FALSE) {
        result = result
    } else if(return_raw_numbers==FALSE & return_plot==TRUE){
        result = plt
    }
    return(result)
}