Mercurial > repos > ecology > ecology_link_between_var
view graph_homogeneity_normality.r @ 0:c7dd4706f982 draft
"planemo upload for repository https://github.com/Marie59/Data_explo_tools commit 2f883743403105d9cac6d267496d985100da3958"
author | ecology |
---|---|
date | Tue, 27 Jul 2021 16:55:49 +0000 |
parents | |
children |
line wrap: on
line source
#Rscript ####################################### ## Homogeneity and normality ## ####################################### #####Packages : car # ggplot2 # ggpubr # Cowplot #####Load arguments args <- commandArgs(trailingOnly = TRUE) if (length(args) == 0) { stop("This tool needs at least one argument") }else{ table <- args[1] hr <- args[2] date <- as.numeric(args[3]) spe <- as.numeric(args[4]) var <- as.numeric(args[5]) } if (hr == "false") { hr <- FALSE }else{ hr <- TRUE } #####Import data data <- read.table(table, sep = "\t", dec = ".", header = hr, fill = TRUE, encoding = "UTF-8") data <- na.omit(data) coldate <- colnames(data)[date] colspe <- colnames(data)[spe] colvar <- colnames(data)[var] #####Your analysis ####Homogeneity of the variance#### ##Test of Levene## testlevene <- function(data, col1, col2) { data[, col1] <- as.numeric(data[, col1]) data[, col2] <- as.factor(data[, col2]) tb_levene <- car::leveneTest(y = data[, col1], group = data[, col2]) return(tb_levene) } levene <- capture.output(testlevene(data = data, col1 = colvar, col2 = colspe)) cat("\nwrite table with levene test. \n--> \"", paste(levene, "\"\n", sep = ""), file = "levene.txt", sep = "", append = TRUE) ##Two boxplots to visualize it## homog_var <- function(data, col1, col2, col3, mult) { data[, col1] <- as.factor(data[, col1]) if (mult) { for (spe in unique(data[, col2])) { data_cut <- data[data[, col2] == spe, ] graph_2 <- ggplot2::ggplot(data_cut, ggplot2::aes_string(x = col1, y = col3, color = col1)) + ggplot2::geom_boxplot() + ggplot2::theme(legend.position = "none", axis.text.x = ggplot2::element_text(angle = 90, vjust = 0.5, hjust = 1), panel.background = ggplot2::element_rect(fill = "#d9d4c5", colour = "#d9d4c5", linetype = "solid"), panel.grid.major = ggplot2::element_line(linetype = "solid", colour = "white"), panel.grid.minor = ggplot2::element_line(linetype = "solid", colour = "white")) ggplot2::ggsave(paste("Homogeneity_of_", spe, ".png"), graph_2, width = 16, height = 9, units = "cm") } }else{ graph_1 <- ggplot2::ggplot(data, ggplot2::aes_string(x = col1, y = col3, color = col1)) + ggplot2::geom_boxplot() + ggplot2::theme(legend.position = "none", axis.text.x = ggplot2::element_text(angle = 90, vjust = 0.5, hjust = 1)) #Put multiple panels graph_2 <- graph_1 + ggplot2::facet_grid(rows = ggplot2::vars(data[, col2]), scales = "free") + ggplot2::theme(panel.background = ggplot2::element_rect(fill = "#d9d4c5", colour = "#d9d4c5", linetype = "solid"), panel.grid.major = ggplot2::element_line(linetype = "solid", colour = "white"), panel.grid.minor = ggplot2::element_line(linetype = "solid", colour = "white")) ggplot2::ggsave("Homogeneity.png", graph_2, width = 16, height = 9, units = "cm") } } ####Normality of the distribution#### # Kolmogorov-Smirnov test ks <- capture.output(ks.test(x = data[, var], y = "pnorm", alternative = "two.sided")) cat("\nwrite table with Kolmogorov-Smirnov test. \n--> \"", paste(ks, "\"\n", sep = ""), file = "ks.txt", sep = "", append = TRUE) #Histogramm with distribution line graph_hist <- function(data, var1) { graph_hist <- ggplot2::ggplot(data) + ggplot2::geom_histogram(ggplot2::aes_string(x = var1), binwidth = 2, color = "black", fill = "white") + ggplot2::geom_density(ggplot2::aes_string(var1), alpha = 0.12, fill = "red") + ggplot2::ggtitle("Distribution histogram") return(graph_hist) } #Add the mean dashed line add_mean <- function(graph, var1) { graph_mean <- graph + ggplot2::geom_vline(xintercept = mean(data[, var1]), color = "midnightblue", linetype = "dashed", size = 1) return(graph_mean) } #Adding a QQplot graph_qqplot <- function(data, var1) { graph2 <- ggpubr::ggqqplot(data, var1, color = "midnightblue") + ggplot2::ggtitle("Q-Q plot") return(graph2) } #On suppose que les données sont distribuées normalement lorsque les points suivent approximativement la ligne de référence à 45 degrés. graph_fin <- function(graph1, graph2) { graph <- cowplot::plot_grid(graph1, graph2, ncol = 2, nrow = 1) ggplot2::ggsave("Normal_distribution.png", graph, width = 10, height = 7, units = "cm") } mult <- ifelse(length(unique(data[, colspe])) == 2, FALSE, TRUE) homog_var(data, col1 = coldate, col2 = colspe, col3 = colvar, mult = mult) graph_hist1 <- graph_hist(data, var1 = colvar) graph_mean <- add_mean(graph = graph_hist1, var1 = colvar) graph_fin(graph1 = graph_mean, graph2 = graph_qqplot(data, var1 = colvar))