R/POST_FATE.graphic_validationStatistics.R
In MayaGueguen/RFate: Vegetation modelling with the FATE model
Defines functions
POST_FATE.graphic_validationStatistics
Documented in
POST_FATE.graphic_validationStatistics
### HEADER #####################################################################
##' @title Create a graphical representation of several statistics for each PFG
##' to asses the quality of the model for one (or several) specific year of a
##' \code{FATE} simulation
##'
##' @name POST_FATE.graphic_validationStatistics
##'
##' @author Maya Guéguen
##'
##' @description This script is designed to produce a graphical representation
##' of several statistics (sensitivity, specificity, TSS, AUC) for quality
##' assessment for one (or several) specific \code{FATE} simulation year.
##'
##' @param name.simulation a \code{string} corresponding to the main directory
##' or simulation name of the \code{FATE} simulation
##' @param file.simulParam default \code{NULL}. \cr A \code{string}
##' corresponding to the name of a parameter file that will be contained into
##' the \code{PARAM_SIMUL} folder of the \code{FATE} simulation
##' @param years an \code{integer}, or a \code{vector} of \code{integer},
##' corresponding to the simulation year(s) that will be used to extract PFG
##' relative abundance maps
##' @param mat.PFG.obs a \code{data.frame} with 4 columns : \code{PFG},
##' \code{X}, \code{Y}, \code{obs}
##' @param opt.ras_habitat (\emph{optional}) default \code{NULL}. \cr
##' A \code{string} corresponding to the file name of a raster mask, with an
##' \code{integer} value within each pixel, corresponding to a specific habitat
##' @param opt.doPlot (\emph{optional}) default \code{TRUE}. \cr If \code{TRUE},
##' plot(s) will be processed, otherwise only the calculation and reorganization
##' of outputs will occur, be saved and returned
##'
##'
##' @details
##'
##' This function allows to obtain, for a specific \code{FATE} simulation and
##' a specific parameter file within this simulation, \strong{PFG validation
##' statistic values} and one preanalytical graphic. \cr \cr
##'
##'
##' Observation records (presences and absences) are required for each PFG
##' within the \code{mat.PFG.obs} object :
##'
##' \describe{
##' \item{\code{PFG}}{the concerned plant functional group}
##' \item{\code{X, Y}}{the coordinates of each observation, matching with the
##' projection of the mask of \code{name.simulation}}
##' \item{\code{obs}}{either \code{0} or \code{1} to indicate presence or
##' absence \cr \cr}
##' }
##'
##'
##' For each PFG and each selected simulation year, raster maps are retrieved
##' from the results folder \code{ABUND_REL_perPFG_allStrata} and lead to the
##' production of one table :
##'
##' \itemize{
##' \item{the value of \strong{several statistics to evaluate the predictive
##' quality of the model} for each plant functional group \cr
##' (\code{\link[PresenceAbsence]{sensitivity}},
##' \code{\link[PresenceAbsence]{specificity}},
##' \code{\link[PresenceAbsence]{auc}},
##' \code{TSS = sensitivity + specificity - 1})}
##' }
##'
##' \strong{If a raster mask for habitat has been provided}, the values and
##' graphics will be also calculated per habitat. \cr \cr
##'
##' \strong{It requires} that the \code{\link{POST_FATE.relativeAbund}}
##' function has been run and that the folder \code{ABUND_REL_perPFG_allStrata}
##' exists. \cr \cr
##'
##' \strong{This \code{.csv} file can then be used by other functions} :
##'
##' \itemize{
##' \item to produce maps of PFG \emph{presence / absence} from modelled
##' abundances \cr (see \code{\link{POST_FATE.binaryMaps}})
##' }
##'
##'
##' @return A \code{list} containing one \code{data.frame} object with the
##' following columns, and one \code{ggplot2} object :
##'
##' \describe{
##' \item{tab}{
##' \describe{
##' \item{\code{PFG}}{concerned plant functional group}
##' \item{\code{AUC.sd}}{standard deviation of the AUC values}
##' \item{\code{sensitivity.sd}}{standard deviation of the sensitivity
##' values}
##' \item{\code{specificity.sd}}{standard deviation of the specificity
##' values}
##' \item{\code{variable}}{name of the calculated statistic among
##' \code{sensitivity}, \code{specificity}, \code{TSS} and \code{AUC}}
##' \item{\code{value}}{value of the corresponding statistic}
##' }
##' }
##' \item{plot}{\code{ggplot2} object, representing the values for each PFG
##' of these four validation statistics (sensitivity, specificity, TSS, AUC)
##' \cr \cr}
##' }
##'
##' One \file{POST_FATE_TABLE_YEAR_[...].csv} file is created :
##' \describe{
##' \item{\file{validationStatistics}}{containing the \code{data.frame}
##' detailed above}
##' }
##'
##' One \file{POST_FATE_[...].pdf} file is created :
##' \describe{
##' \item{\file{GRAPHIC_B \cr validationStatistics}}{to assess the modeling
##' quality of each PFG based on given observations within the studied area}
##' }
##'
##'
##' @keywords FATE, outputs, area under curve, sensitivity, specificity,
##' true skill statistic
##'
##' @seealso \code{\link{POST_FATE.relativeAbund}},
##' \code{\link[PresenceAbsence]{cmx}},
##' \code{\link[PresenceAbsence]{sensitivity}},
##' \code{\link[PresenceAbsence]{specificity}},
##' \code{\link[PresenceAbsence]{auc}},
##' \code{\link{.getCutoff}},
##' \code{\link{POST_FATE.binaryMaps}}
##'
##' @examples
##'
##'
##'
##' ## ----------------------------------------------------------------------------------------- ##
##' ## Load example data
##'
##' @export
##'
##' @importFrom foreach foreach %do%
##' @importFrom reshape2 melt
##' @importFrom raster raster stack cellFromXY
##' @importFrom grid unit
##' @importFrom PresenceAbsence cmx sensitivity specificity auc
##'
##' @importFrom ggplot2 ggplot aes_string
##' geom_bar geom_point geom_hline geom_errorbar annotate
##' scale_fill_gradientn scale_y_continuous
##' labs ylim theme element_text element_blank
##' @importFrom ggthemes theme_fivethirtyeight
##' @importFrom ggExtra ggMarginal
##' @importFrom gridExtra grid.arrange
##' @importFrom cowplot get_legend
##' @importFrom RColorBrewer brewer.pal
##' @importFrom grDevices pdf
##' @importFrom graphics plot
##'
## END OF HEADER ###############################################################
POST_FATE.graphic_validationStatistics = function(
name.simulation
, file.simulParam = NULL
, years
, mat.PFG.obs
, opt.ras_habitat = NULL
, opt.doPlot = TRUE
){
#############################################################################
## CHECK parameter name.simulation
.testParam_existFolder(name.simulation, "PARAM_SIMUL/")
.testParam_existFolder(name.simulation, "RESULTS/")
.testParam_existFolder(name.simulation, "DATA/")
name.simulation = sub("/", "", name.simulation)
## CHECK parameter file.simulParam
abs.simulParams = .getParam_abs.simulParams(file.simulParam, name.simulation)
## CHECK parameter years
.testParam_notInteger.m("years", years)
## CHECK parameter mat.PFG.obs
if (.testParam_notDf(mat.PFG.obs))
{
.stopMessage_beDataframe("mat.PFG.obs")
} else
{
if (nrow(mat.PFG.obs) == 0 || ncol(mat.PFG.obs) != 4)
{
.stopMessage_numRowCol("mat.PFG.obs", c("PFG", "X", "Y", "obs"))
} else
{
if(.testParam_notColnames(mat.PFG.obs, c("PFG", "X", "Y", "obs")))
{
.stopMessage_columnNames("mat.PFG.obs", c("PFG", "X", "Y", "obs"))
}
}
mat.PFG.obs$PFG = as.character(mat.PFG.obs$PFG)
mat.PFG.obs = na.exclude(mat.PFG.obs)
if (nrow(mat.PFG.obs) == 0)
{
stop("Missing data!\n Too many NA values. Please check.")
}
.testParam_notChar.m("mat.PFG.obs$PFG", mat.PFG.obs$PFG)
.testParam_notNum.m("mat.PFG.obs$X", mat.PFG.obs$X)
.testParam_notNum.m("mat.PFG.obs$Y", mat.PFG.obs$Y)
.testParam_notInValues.m("mat.PFG.obs$obs", mat.PFG.obs$obs, c(0, 1))
}
## CHECK parameter opt.ras_habitat
if (!.testParam_notDef(opt.ras_habitat))
{
.testParam_notChar.m("opt.ras_habitat", opt.ras_habitat)
.testParam_existFile(opt.ras_habitat)
ras.habitat = raster(opt.ras_habitat)
}
cat("\n\n #------------------------------------------------------------#")
cat("\n # POST_FATE.graphic_validationStatistics")
cat("\n #------------------------------------------------------------# \n")
#############################################################################
res = foreach (abs.simulParam = abs.simulParams) %do%
{
cat("\n+++++++\n")
cat("\n Simulation name : ", name.simulation)
cat("\n Simulation file : ", abs.simulParam)
cat("\n")
## Get results directories ------------------------------------------------
.getGraphics_results(name.simulation = name.simulation
, abs.simulParam = abs.simulParam)
## Get number of PFGs -----------------------------------------------------
## Get PFG names ----------------------------------------------------------
.getGraphics_PFG(name.simulation = name.simulation
, abs.simulParam = abs.simulParam)
## Get raster mask --------------------------------------------------------
.getGraphics_mask(name.simulation = name.simulation
, abs.simulParam = abs.simulParam)
## Get PFG observations ---------------------------------------------------
mat.PFG.obs = mat.PFG.obs[which(mat.PFG.obs$PFG %in% PFG), ]
if (nrow(mat.PFG.obs) == 0)
{
stop(paste0("Missing data!\n The names of PFG within `mat.PFG.obs`"
, " is different from the names of PFG contained from "
, name.simulation, "/DATA/PFGS/SUCC/"))
}
## Get habitat information ------------------------------------------------
no_hab = 1
hab_names = "ALL"
if (exists("ras.habitat"))
{
ras.habitat = ras.habitat * ras.mask
df.habitat = data.frame(ID = cellFromXY(ras.habitat, xy.1)
, stringsAsFactors = FALSE)
df.habitat$HAB = ras.habitat[df.habitat$ID]
hab_names = c(hab_names, unique(df.habitat$HAB))
hab_names = hab_names[which(!is.na(hab_names))]
no_hab = length(hab_names)
}
## Get list of arrays and extract years of simulation ---------------------
years = sort(unique(as.numeric(years)))
## UNZIP the raster saved -------------------------------------------------
raster.perPFG.allStrata.REL = .getRasterNames(years, "allStrata", "REL")
## get the data inside the rasters ----------------------------------------
cat("\n ---------- GETTING STATISTICS for")
mat.valid_list = list()
plot_list = foreach (y = years) %do%
{
cat("\n> year", y)
file_name = paste0(dir.output.perPFG.allStrata.REL
, "Abund_relative_YEAR_"
, y
, "_"
, PFG
, "_STRATA_all.tif")
gp = PFG[which(file.exists(file_name))]
file_name = file_name[which(file.exists(file_name))]
if (length(file_name) > 0)
{
ras = stack(file_name) * ras.mask
names(ras) = gp
mat.y = mat.PFG.obs
mat.y$ID = as.numeric(as.factor(paste0(mat.y$X, "_", mat.y$Y)))
mat.y$CELL = cellFromXY(ras, mat.y[, c("X", "Y")])
mat.y = na.exclude(mat.y)
if (exists("df.habitat"))
{
mat.y = merge(mat.y, df.habitat, by.x = "CELL", by.y = "ID", all.x = TRUE)
} else
{
mat.y$HAB = "ALL"
}
mat.y.split = split(mat.y, mat.y$PFG)
mat.valid = foreach(i = 1:length(mat.y.split)
, .combine = "rbind"
) %do%
{
fg = names(mat.y.split)[i]
mat = mat.y.split[[i]]
isThereValues = TRUE
cat("\n PFG ", fg)
if (fg %in% gp)
{
if (nrow(mat) > 0)
{
mat$fg = ras[[fg]][mat$CELL]
mat = mat[, c("ID", "CELL", "obs", "fg", "HAB")]
mat = na.exclude(mat)
}
if (nrow(mat) == 0)
{
isThereValues = FALSE
} else
{
cat(", habitat")
mat.split = split(mat, mat$HAB)
mat.res = foreach (habi = hab_names, .combine = "rbind") %do%
{
mat.res.habi = data.frame(PFG = fg, HAB = habi
, auc = NA, sens = NA
, spec = NA, TSS = NA
, stringsAsFactors = FALSE)
cat(" ", habi)
if (habi == "ALL")
{
tmp = mat
} else
{
tmp = mat.split[[as.character(habi)]]
}
tmp = tmp[, -which(colnames(tmp) %in% c("Row.names", "HAB")), drop = FALSE]
if (nrow(tmp) > 0 &&
length(which(tmp$obs == 0)) > 0 &&
length(which(tmp$obs == 1)) > 0)
{
if (habi == "ALL")
{
assign("cutoff", NULL, envir = .GlobalEnv)
cutoff <<- .getCutoff(Obs = tmp[, "obs"], Fit = tmp[, "fg"])
}
if (exists("cutoff") && !(is.na(cutoff[1])))
{
## Calculate validation statistics ------------------------
mat.bin = tmp
mat.bin$fg = ifelse(mat.bin$fg >= cutoff$Cut, 1, 0)
mat.conf = cmx(mat.bin[, c("ID", "obs", "fg")])
sens = sensitivity(mat.conf)
spec = specificity(mat.conf)
TSS = sens$sensitivity + spec$specificity - 1
auc = auc(mat.bin[, c("ID", "obs", "fg")])
mat.res.habi = data.frame(PFG = fg, HAB = habi
, cutoff = cutoff$Cut
, auc, sens, spec, TSS
, stringsAsFactors = FALSE)
}
}
return(mat.res.habi)
}
if (nrow(na.exclude(mat.res)) == 0)
{
isThereValues = FALSE
}
}
} else
{
isThereValues = FALSE
}
if (!isThereValues)
{
warning(paste0("Missing data!\n No values for PFG ", fg
, ".\n No validation statistics will be produced!"))
return(data.frame(PFG = fg
, cutoff = NA
, HAB = "ALL"
, AUC = NA, AUC.sd = NA
, sensitivity = NA, sensitivity.sd = NA
, specificity = NA, specificity.sd = NA
, TSS = NA
, stringsAsFactors = FALSE))
} else
{
return(mat.res)
}
} ## END mat.valid
cat("\n")
if (nrow(na.exclude(mat.valid)) > 0)
{
## save table -------------------------------------------------------
write.csv(mat.valid
, file = paste0(name.simulation
, "/RESULTS/POST_FATE_TABLE_YEAR_"
, y
, "_validationStatistics_"
, basename(dir.save)
, ".csv")
, row.names = FALSE)
message(paste0("\n The output file POST_FATE_TABLE_YEAR_"
, y
, "_validationStatistics_"
, basename(dir.save)
, ".csv has been successfully created !\n"))
## prepare the plot -------------------------------------------------
mat.valid = melt(mat.valid
, id.vars = c("PFG", "HAB", "AUC.sd"
, "sensitivity.sd", "specificity.sd"))
mat.valid$variable = factor(mat.valid$variable
, c("sensitivity", "TSS"
, "specificity", "AUC"))
mat.valid$AUC.sd[which(mat.valid$variable != "AUC")] = NA
mat.valid$sensitivity.sd[which(mat.valid$variable != "sensitivity")] = NA
mat.valid$specificity.sd[which(mat.valid$variable != "specificity")] = NA
mat.valid_list[[as.character(y)]] = mat.valid
mat.valid$hline = 0.5
mat.valid$hline[which(mat.valid$variable == "AUC")] = 0.8
mat.valid$hline[which(mat.valid$variable == "TSS")] = 0.4
## produce the plot -------------------------------------------------
if (opt.doPlot)
{
cat("\n ---------- PRODUCING PLOT(S)")
plot_list.hab = foreach(habi = hab_names) %do%
{
cat("\n> Preparing for habitat ", habi)
mat.plot = mat.valid[which(mat.valid$HAB == habi), ]
## 1. get the legend
pp = ggplot(mat.plot, aes_string(x = "PFG", y = "value", fill = "value")) +
scale_fill_gradientn(""
, colors = brewer.pal(9, "RdYlGn")
, breaks = seq(0, 1, 0.2)
, limits = c(0, 1)) +
geom_bar(stat = "identity", na.rm = TRUE) +
ylim(0, 1) +
.getGraphics_theme() +
theme(legend.key.width = unit(2, "lines"))
pp_leg = suppressWarnings(get_legend(pp))
## 2. get one plot for the title and for each statistic
pp_list = foreach(vari = c("all", "sensitivity", "specificity", "TSS", "AUC")
) %do%
{
if (vari == "all"){
pp = ggplot(mat.plot, aes_string(x = "PFG"
, y = "value"
, fill = "value")) +
labs(x = "", y = ""
, title = paste0("GRAPH B : validation statistics"
, " - Simulation year : "
, y, " - Habitat ", habi)
, subtitle = paste0("Sensitivity (or specificity) measures "
, "the proportion of actual positives "
, "(or negatives) that are correctly "
, "identified as such.\n"
, "True skill statistic (TSS) values "
, "of -1 indicate predictive abilities "
, "of not better than a random model,\n"
, "0 indicates an indiscriminate model "
, "and +1 a perfect model.\n"
, "AUC corresponds to the area under "
, "the ROC curve (Receiver Operating "
, "Characteristic).\n")) +
.getGraphics_theme() +
theme(panel.grid = element_blank()
, axis.text = element_blank())
} else
{
if (vari == "sensitivity") subti = "Sensitivity - True positive rate"
if (vari == "specificity") subti = "Specificity - True negative rate"
if (vari == "TSS") subti = "True Skill Statistic (TSS)"
if (vari == "AUC") subti = "Area Under Curve (AUC)"
pp = ggplot(mat.plot[which(mat.plot$variable == vari), ]
, aes_string(x = "PFG", y = "value", fill = "value")) +
scale_fill_gradientn(guide = F
, colors = brewer.pal(9, "RdYlGn")
, breaks = seq(0, 1, 0.2)
, limits = c(0, 1)) +
scale_y_continuous(breaks = seq(0, 1, 0.2), limits = c(0, 1.08)) +
geom_point(alpha = 0) +
geom_bar(stat = "identity", na.rm = TRUE) +
geom_hline(aes_string(yintercept = "hline")
, lty = 2, color = "grey30") +
geom_errorbar(aes_string(ymin = "value - sensitivity.sd"
, ymax = "value + sensitivity.sd")
, color = "grey30", na.rm = TRUE) +
geom_errorbar(aes_string(ymin = "value - specificity.sd"
, ymax = "value + specificity.sd")
, color = "grey30", na.rm = TRUE) +
geom_errorbar(aes_string(ymin = "value - AUC.sd"
, ymax = "value + AUC.sd")
, color = "grey30", na.rm = TRUE) +
annotate(geom = "text"
, x = no_PFG / 2
, y = 1.05
, label = subti
, size = 4) +
.getGraphics_theme() +
theme(axis.text.x = element_text(angle = 90))
pp = suppressWarnings(ggMarginal(pp
, type = "boxplot"
, margins = "y"
, size = 7))
}
return(pp)
}
## 3. gather everything
pp_list[[6]] = pp_leg
pp_final = grid.arrange(grobs = pp_list
, layout_matrix = matrix(c(1,1,2,3,2,3,4,5,4,5,6,6)
, ncol = 2, byrow = TRUE)
, newpage = ifelse(y == years[1], FALSE, TRUE))
pp_final = cowplot::ggdraw(pp_final) +
theme(plot.background = element_rect(fill = "transparent", color = NA))
# require(patchwork)
# pp_final <- pp_list[[1]] / (pp_list[[2]] | pp_list[[3]]) / (pp_list[[4]] | pp_list[[5]]) / pp_list[[6]]
return(pp_final)
} ## END loop on hab_names
names(plot_list.hab) = hab_names
return(plot_list.hab)
} ## END opt.doPlot
} else
{
warning(paste0("Missing data!\n No validation has been calculated for year ", y, "!"))
return(NULL)
}
} ## END condition file_name
} ## END loop on years
names(plot_list) = years
cat("\n")
## SAVE plots into file ---------------------------------------------------
if (opt.doPlot && sum(sapply(plot_list, is.null)) < length(plot_list))
{
pdf(file = paste0(name.simulation
, "/RESULTS/POST_FATE_GRAPHIC_B_validationStatistics_"
, basename(dir.save), ".pdf")
, width = 12, height = 10)
for (y in years)
{
for (habi in hab_names)
{
pp = plot_list[[as.character(y)]][[habi]]
if (!is.null(pp)) plot(pp)
}
}
dev.off()
}
## ------------------------------------------------------------------------
cat("\n> Done!\n")
return(list(tab = mat.valid_list, plot = plot_list))
} ## END loop on abs.simulParams
names(res) = abs.simulParams
return(res)
}
MayaGueguen/RFate documentation built on Oct. 17, 2020, 8:06 a.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Defines functions POST_FATE.graphic_validationStatistics
Documented in POST_FATE.graphic_validationStatistics
### HEADER #####################################################################
##' @title Create a graphical representation of several statistics for each PFG
##' to asses the quality of the model for one (or several) specific year of a
##' \code{FATE} simulation
##'
##' @name POST_FATE.graphic_validationStatistics
##'
##' @author Maya Guéguen
##'
##' @description This script is designed to produce a graphical representation
##' of several statistics (sensitivity, specificity, TSS, AUC) for quality
##' assessment for one (or several) specific \code{FATE} simulation year.
##'
##' @param name.simulation a \code{string} corresponding to the main directory
##' or simulation name of the \code{FATE} simulation
##' @param file.simulParam default \code{NULL}. \cr A \code{string}
##' corresponding to the name of a parameter file that will be contained into
##' the \code{PARAM_SIMUL} folder of the \code{FATE} simulation
##' @param years an \code{integer}, or a \code{vector} of \code{integer},
##' corresponding to the simulation year(s) that will be used to extract PFG
##' relative abundance maps
##' @param mat.PFG.obs a \code{data.frame} with 4 columns : \code{PFG},
##' \code{X}, \code{Y}, \code{obs}
##' @param opt.ras_habitat (\emph{optional}) default \code{NULL}. \cr
##' A \code{string} corresponding to the file name of a raster mask, with an
##' \code{integer} value within each pixel, corresponding to a specific habitat
##' @param opt.doPlot (\emph{optional}) default \code{TRUE}. \cr If \code{TRUE},
##' plot(s) will be processed, otherwise only the calculation and reorganization
##' of outputs will occur, be saved and returned
##'
##'
##' @details
##'
##' This function allows to obtain, for a specific \code{FATE} simulation and
##' a specific parameter file within this simulation, \strong{PFG validation
##' statistic values} and one preanalytical graphic. \cr \cr
##'
##'
##' Observation records (presences and absences) are required for each PFG
##' within the \code{mat.PFG.obs} object :
##'
##' \describe{
##' \item{\code{PFG}}{the concerned plant functional group}
##' \item{\code{X, Y}}{the coordinates of each observation, matching with the
##' projection of the mask of \code{name.simulation}}
##' \item{\code{obs}}{either \code{0} or \code{1} to indicate presence or
##' absence \cr \cr}
##' }
##'
##'
##' For each PFG and each selected simulation year, raster maps are retrieved
##' from the results folder \code{ABUND_REL_perPFG_allStrata} and lead to the
##' production of one table :
##'
##' \itemize{
##' \item{the value of \strong{several statistics to evaluate the predictive
##' quality of the model} for each plant functional group \cr
##' (\code{\link[PresenceAbsence]{sensitivity}},
##' \code{\link[PresenceAbsence]{specificity}},
##' \code{\link[PresenceAbsence]{auc}},
##' \code{TSS = sensitivity + specificity - 1})}
##' }
##'
##' \strong{If a raster mask for habitat has been provided}, the values and
##' graphics will be also calculated per habitat. \cr \cr
##'
##' \strong{It requires} that the \code{\link{POST_FATE.relativeAbund}}
##' function has been run and that the folder \code{ABUND_REL_perPFG_allStrata}
##' exists. \cr \cr
##'
##' \strong{This \code{.csv} file can then be used by other functions} :
##'
##' \itemize{
##' \item to produce maps of PFG \emph{presence / absence} from modelled
##' abundances \cr (see \code{\link{POST_FATE.binaryMaps}})
##' }
##'
##'
##' @return A \code{list} containing one \code{data.frame} object with the
##' following columns, and one \code{ggplot2} object :
##'
##' \describe{
##' \item{tab}{
##' \describe{
##' \item{\code{PFG}}{concerned plant functional group}
##' \item{\code{AUC.sd}}{standard deviation of the AUC values}
##' \item{\code{sensitivity.sd}}{standard deviation of the sensitivity
##' values}
##' \item{\code{specificity.sd}}{standard deviation of the specificity
##' values}
##' \item{\code{variable}}{name of the calculated statistic among
##' \code{sensitivity}, \code{specificity}, \code{TSS} and \code{AUC}}
##' \item{\code{value}}{value of the corresponding statistic}
##' }
##' }
##' \item{plot}{\code{ggplot2} object, representing the values for each PFG
##' of these four validation statistics (sensitivity, specificity, TSS, AUC)
##' \cr \cr}
##' }
##'
##' One \file{POST_FATE_TABLE_YEAR_[...].csv} file is created :
##' \describe{
##' \item{\file{validationStatistics}}{containing the \code{data.frame}
##' detailed above}
##' }
##'
##' One \file{POST_FATE_[...].pdf} file is created :
##' \describe{
##' \item{\file{GRAPHIC_B \cr validationStatistics}}{to assess the modeling
##' quality of each PFG based on given observations within the studied area}
##' }
##'
##'
##' @keywords FATE, outputs, area under curve, sensitivity, specificity,
##' true skill statistic
##'
##' @seealso \code{\link{POST_FATE.relativeAbund}},
##' \code{\link[PresenceAbsence]{cmx}},
##' \code{\link[PresenceAbsence]{sensitivity}},
##' \code{\link[PresenceAbsence]{specificity}},
##' \code{\link[PresenceAbsence]{auc}},
##' \code{\link{.getCutoff}},
##' \code{\link{POST_FATE.binaryMaps}}
##'
##' @examples
##'
##'
##'
##' ## ----------------------------------------------------------------------------------------- ##
##' ## Load example data
##'
##' @export
##'
##' @importFrom foreach foreach %do%
##' @importFrom reshape2 melt
##' @importFrom raster raster stack cellFromXY
##' @importFrom grid unit
##' @importFrom PresenceAbsence cmx sensitivity specificity auc
##'
##' @importFrom ggplot2 ggplot aes_string
##' geom_bar geom_point geom_hline geom_errorbar annotate
##' scale_fill_gradientn scale_y_continuous
##' labs ylim theme element_text element_blank
##' @importFrom ggthemes theme_fivethirtyeight
##' @importFrom ggExtra ggMarginal
##' @importFrom gridExtra grid.arrange
##' @importFrom cowplot get_legend
##' @importFrom RColorBrewer brewer.pal
##' @importFrom grDevices pdf
##' @importFrom graphics plot
##'
## END OF HEADER ###############################################################
POST_FATE.graphic_validationStatistics = function(
name.simulation
, file.simulParam = NULL
, years
, mat.PFG.obs
, opt.ras_habitat = NULL
, opt.doPlot = TRUE
){
#############################################################################
## CHECK parameter name.simulation
.testParam_existFolder(name.simulation, "PARAM_SIMUL/")
.testParam_existFolder(name.simulation, "RESULTS/")
.testParam_existFolder(name.simulation, "DATA/")
name.simulation = sub("/", "", name.simulation)
## CHECK parameter file.simulParam
abs.simulParams = .getParam_abs.simulParams(file.simulParam, name.simulation)
## CHECK parameter years
.testParam_notInteger.m("years", years)
## CHECK parameter mat.PFG.obs
if (.testParam_notDf(mat.PFG.obs))
{
.stopMessage_beDataframe("mat.PFG.obs")
} else
{
if (nrow(mat.PFG.obs) == 0 || ncol(mat.PFG.obs) != 4)
{
.stopMessage_numRowCol("mat.PFG.obs", c("PFG", "X", "Y", "obs"))
} else
{
if(.testParam_notColnames(mat.PFG.obs, c("PFG", "X", "Y", "obs")))
{
.stopMessage_columnNames("mat.PFG.obs", c("PFG", "X", "Y", "obs"))
}
}
mat.PFG.obs$PFG = as.character(mat.PFG.obs$PFG)
mat.PFG.obs = na.exclude(mat.PFG.obs)
if (nrow(mat.PFG.obs) == 0)
{
stop("Missing data!\n Too many NA values. Please check.")
}
.testParam_notChar.m("mat.PFG.obs$PFG", mat.PFG.obs$PFG)
.testParam_notNum.m("mat.PFG.obs$X", mat.PFG.obs$X)
.testParam_notNum.m("mat.PFG.obs$Y", mat.PFG.obs$Y)
.testParam_notInValues.m("mat.PFG.obs$obs", mat.PFG.obs$obs, c(0, 1))
}
## CHECK parameter opt.ras_habitat
if (!.testParam_notDef(opt.ras_habitat))
{
.testParam_notChar.m("opt.ras_habitat", opt.ras_habitat)
.testParam_existFile(opt.ras_habitat)
ras.habitat = raster(opt.ras_habitat)
}
cat("\n\n #------------------------------------------------------------#")
cat("\n # POST_FATE.graphic_validationStatistics")
cat("\n #------------------------------------------------------------# \n")
#############################################################################
res = foreach (abs.simulParam = abs.simulParams) %do%
{
cat("\n+++++++\n")
cat("\n Simulation name : ", name.simulation)
cat("\n Simulation file : ", abs.simulParam)
cat("\n")
## Get results directories ------------------------------------------------
.getGraphics_results(name.simulation = name.simulation
, abs.simulParam = abs.simulParam)
## Get number of PFGs -----------------------------------------------------
## Get PFG names ----------------------------------------------------------
.getGraphics_PFG(name.simulation = name.simulation
, abs.simulParam = abs.simulParam)
## Get raster mask --------------------------------------------------------
.getGraphics_mask(name.simulation = name.simulation
, abs.simulParam = abs.simulParam)
## Get PFG observations ---------------------------------------------------
mat.PFG.obs = mat.PFG.obs[which(mat.PFG.obs$PFG %in% PFG), ]
if (nrow(mat.PFG.obs) == 0)
{
stop(paste0("Missing data!\n The names of PFG within `mat.PFG.obs`"
, " is different from the names of PFG contained from "
, name.simulation, "/DATA/PFGS/SUCC/"))
}
## Get habitat information ------------------------------------------------
no_hab = 1
hab_names = "ALL"
if (exists("ras.habitat"))
{
ras.habitat = ras.habitat * ras.mask
df.habitat = data.frame(ID = cellFromXY(ras.habitat, xy.1)
, stringsAsFactors = FALSE)
df.habitat$HAB = ras.habitat[df.habitat$ID]
hab_names = c(hab_names, unique(df.habitat$HAB))
hab_names = hab_names[which(!is.na(hab_names))]
no_hab = length(hab_names)
}
## Get list of arrays and extract years of simulation ---------------------
years = sort(unique(as.numeric(years)))
## UNZIP the raster saved -------------------------------------------------
raster.perPFG.allStrata.REL = .getRasterNames(years, "allStrata", "REL")
## get the data inside the rasters ----------------------------------------
cat("\n ---------- GETTING STATISTICS for")
mat.valid_list = list()
plot_list = foreach (y = years) %do%
{
cat("\n> year", y)
file_name = paste0(dir.output.perPFG.allStrata.REL
, "Abund_relative_YEAR_"
, y
, "_"
, PFG
, "_STRATA_all.tif")
gp = PFG[which(file.exists(file_name))]
file_name = file_name[which(file.exists(file_name))]
if (length(file_name) > 0)
{
ras = stack(file_name) * ras.mask
names(ras) = gp
mat.y = mat.PFG.obs
mat.y$ID = as.numeric(as.factor(paste0(mat.y$X, "_", mat.y$Y)))
mat.y$CELL = cellFromXY(ras, mat.y[, c("X", "Y")])
mat.y = na.exclude(mat.y)
if (exists("df.habitat"))
{
mat.y = merge(mat.y, df.habitat, by.x = "CELL", by.y = "ID", all.x = TRUE)
} else
{
mat.y$HAB = "ALL"
}
mat.y.split = split(mat.y, mat.y$PFG)
mat.valid = foreach(i = 1:length(mat.y.split)
, .combine = "rbind"
) %do%
{
fg = names(mat.y.split)[i]
mat = mat.y.split[[i]]
isThereValues = TRUE
cat("\n PFG ", fg)
if (fg %in% gp)
{
if (nrow(mat) > 0)
{
mat$fg = ras[[fg]][mat$CELL]
mat = mat[, c("ID", "CELL", "obs", "fg", "HAB")]
mat = na.exclude(mat)
}
if (nrow(mat) == 0)
{
isThereValues = FALSE
} else
{
cat(", habitat")
mat.split = split(mat, mat$HAB)
mat.res = foreach (habi = hab_names, .combine = "rbind") %do%
{
mat.res.habi = data.frame(PFG = fg, HAB = habi
, auc = NA, sens = NA
, spec = NA, TSS = NA
, stringsAsFactors = FALSE)
cat(" ", habi)
if (habi == "ALL")
{
tmp = mat
} else
{
tmp = mat.split[[as.character(habi)]]
}
tmp = tmp[, -which(colnames(tmp) %in% c("Row.names", "HAB")), drop = FALSE]
if (nrow(tmp) > 0 &&
length(which(tmp$obs == 0)) > 0 &&
length(which(tmp$obs == 1)) > 0)
{
if (habi == "ALL")
{
assign("cutoff", NULL, envir = .GlobalEnv)
cutoff <<- .getCutoff(Obs = tmp[, "obs"], Fit = tmp[, "fg"])
}
if (exists("cutoff") && !(is.na(cutoff[1])))
{
## Calculate validation statistics ------------------------
mat.bin = tmp
mat.bin$fg = ifelse(mat.bin$fg >= cutoff$Cut, 1, 0)
mat.conf = cmx(mat.bin[, c("ID", "obs", "fg")])
sens = sensitivity(mat.conf)
spec = specificity(mat.conf)
TSS = sens$sensitivity + spec$specificity - 1
auc = auc(mat.bin[, c("ID", "obs", "fg")])
mat.res.habi = data.frame(PFG = fg, HAB = habi
, cutoff = cutoff$Cut
, auc, sens, spec, TSS
, stringsAsFactors = FALSE)
}
}
return(mat.res.habi)
}
if (nrow(na.exclude(mat.res)) == 0)
{
isThereValues = FALSE
}
}
} else
{
isThereValues = FALSE
}
if (!isThereValues)
{
warning(paste0("Missing data!\n No values for PFG ", fg
, ".\n No validation statistics will be produced!"))
return(data.frame(PFG = fg
, cutoff = NA
, HAB = "ALL"
, AUC = NA, AUC.sd = NA
, sensitivity = NA, sensitivity.sd = NA
, specificity = NA, specificity.sd = NA
, TSS = NA
, stringsAsFactors = FALSE))
} else
{
return(mat.res)
}
} ## END mat.valid
cat("\n")
if (nrow(na.exclude(mat.valid)) > 0)
{
## save table -------------------------------------------------------
write.csv(mat.valid
, file = paste0(name.simulation
, "/RESULTS/POST_FATE_TABLE_YEAR_"
, y
, "_validationStatistics_"
, basename(dir.save)
, ".csv")
, row.names = FALSE)
message(paste0("\n The output file POST_FATE_TABLE_YEAR_"
, y
, "_validationStatistics_"
, basename(dir.save)
, ".csv has been successfully created !\n"))
## prepare the plot -------------------------------------------------
mat.valid = melt(mat.valid
, id.vars = c("PFG", "HAB", "AUC.sd"
, "sensitivity.sd", "specificity.sd"))
mat.valid$variable = factor(mat.valid$variable
, c("sensitivity", "TSS"
, "specificity", "AUC"))
mat.valid$AUC.sd[which(mat.valid$variable != "AUC")] = NA
mat.valid$sensitivity.sd[which(mat.valid$variable != "sensitivity")] = NA
mat.valid$specificity.sd[which(mat.valid$variable != "specificity")] = NA
mat.valid_list[[as.character(y)]] = mat.valid
mat.valid$hline = 0.5
mat.valid$hline[which(mat.valid$variable == "AUC")] = 0.8
mat.valid$hline[which(mat.valid$variable == "TSS")] = 0.4
## produce the plot -------------------------------------------------
if (opt.doPlot)
{
cat("\n ---------- PRODUCING PLOT(S)")
plot_list.hab = foreach(habi = hab_names) %do%
{
cat("\n> Preparing for habitat ", habi)
mat.plot = mat.valid[which(mat.valid$HAB == habi), ]
## 1. get the legend
pp = ggplot(mat.plot, aes_string(x = "PFG", y = "value", fill = "value")) +
scale_fill_gradientn(""
, colors = brewer.pal(9, "RdYlGn")
, breaks = seq(0, 1, 0.2)
, limits = c(0, 1)) +
geom_bar(stat = "identity", na.rm = TRUE) +
ylim(0, 1) +
.getGraphics_theme() +
theme(legend.key.width = unit(2, "lines"))
pp_leg = suppressWarnings(get_legend(pp))
## 2. get one plot for the title and for each statistic
pp_list = foreach(vari = c("all", "sensitivity", "specificity", "TSS", "AUC")
) %do%
{
if (vari == "all"){
pp = ggplot(mat.plot, aes_string(x = "PFG"
, y = "value"
, fill = "value")) +
labs(x = "", y = ""
, title = paste0("GRAPH B : validation statistics"
, " - Simulation year : "
, y, " - Habitat ", habi)
, subtitle = paste0("Sensitivity (or specificity) measures "
, "the proportion of actual positives "
, "(or negatives) that are correctly "
, "identified as such.\n"
, "True skill statistic (TSS) values "
, "of -1 indicate predictive abilities "
, "of not better than a random model,\n"
, "0 indicates an indiscriminate model "
, "and +1 a perfect model.\n"
, "AUC corresponds to the area under "
, "the ROC curve (Receiver Operating "
, "Characteristic).\n")) +
.getGraphics_theme() +
theme(panel.grid = element_blank()
, axis.text = element_blank())
} else
{
if (vari == "sensitivity") subti = "Sensitivity - True positive rate"
if (vari == "specificity") subti = "Specificity - True negative rate"
if (vari == "TSS") subti = "True Skill Statistic (TSS)"
if (vari == "AUC") subti = "Area Under Curve (AUC)"
pp = ggplot(mat.plot[which(mat.plot$variable == vari), ]
, aes_string(x = "PFG", y = "value", fill = "value")) +
scale_fill_gradientn(guide = F
, colors = brewer.pal(9, "RdYlGn")
, breaks = seq(0, 1, 0.2)
, limits = c(0, 1)) +
scale_y_continuous(breaks = seq(0, 1, 0.2), limits = c(0, 1.08)) +
geom_point(alpha = 0) +
geom_bar(stat = "identity", na.rm = TRUE) +
geom_hline(aes_string(yintercept = "hline")
, lty = 2, color = "grey30") +
geom_errorbar(aes_string(ymin = "value - sensitivity.sd"
, ymax = "value + sensitivity.sd")
, color = "grey30", na.rm = TRUE) +
geom_errorbar(aes_string(ymin = "value - specificity.sd"
, ymax = "value + specificity.sd")
, color = "grey30", na.rm = TRUE) +
geom_errorbar(aes_string(ymin = "value - AUC.sd"
, ymax = "value + AUC.sd")
, color = "grey30", na.rm = TRUE) +
annotate(geom = "text"
, x = no_PFG / 2
, y = 1.05
, label = subti
, size = 4) +
.getGraphics_theme() +
theme(axis.text.x = element_text(angle = 90))
pp = suppressWarnings(ggMarginal(pp
, type = "boxplot"
, margins = "y"
, size = 7))
}
return(pp)
}
## 3. gather everything
pp_list[[6]] = pp_leg
pp_final = grid.arrange(grobs = pp_list
, layout_matrix = matrix(c(1,1,2,3,2,3,4,5,4,5,6,6)
, ncol = 2, byrow = TRUE)
, newpage = ifelse(y == years[1], FALSE, TRUE))
pp_final = cowplot::ggdraw(pp_final) +
theme(plot.background = element_rect(fill = "transparent", color = NA))
# require(patchwork)
# pp_final <- pp_list[[1]] / (pp_list[[2]] | pp_list[[3]]) / (pp_list[[4]] | pp_list[[5]]) / pp_list[[6]]
return(pp_final)
} ## END loop on hab_names
names(plot_list.hab) = hab_names
return(plot_list.hab)
} ## END opt.doPlot
} else
{
warning(paste0("Missing data!\n No validation has been calculated for year ", y, "!"))
return(NULL)
}
} ## END condition file_name
} ## END loop on years
names(plot_list) = years
cat("\n")
## SAVE plots into file ---------------------------------------------------
if (opt.doPlot && sum(sapply(plot_list, is.null)) < length(plot_list))
{
pdf(file = paste0(name.simulation
, "/RESULTS/POST_FATE_GRAPHIC_B_validationStatistics_"
, basename(dir.save), ".pdf")
, width = 12, height = 10)
for (y in years)
{
for (habi in hab_names)
{
pp = plot_list[[as.character(y)]][[habi]]
if (!is.null(pp)) plot(pp)
}
}
dev.off()
}
## ------------------------------------------------------------------------
cat("\n> Done!\n")
return(list(tab = mat.valid_list, plot = plot_list))
} ## END loop on abs.simulParams
names(res) = abs.simulParams
return(res)
}
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.