R/f1ScoreVoting.R

Defines functions f1ScoreVoting

Documented in f1ScoreVoting

#' f1ScoreVoting
#'
#' Calculates the F1 score fore each group in a partition, when provided with a fuzzy classification.
#'
#' @param voting A list where each entry is a vote on the respective label.
#' @param clustering Labels of the partition.
#' @param cytometry Data of the clustering, where the last variable contains the original labels.
#' @param nivel_sup level of tolerance for assigning a hard clustering. Should be greater or equal than 1. Class A is assigned if class A > nivel_sup * Class B.
#' @param noise.cells An array of labels to be considered as noise.
#'
#' @return A matrix where the first row is the F1 score, the second row is the Precision and the third row is the Recall.
#'
#' @examples
#' # # We construct a simple database selecting only some of the Cytometries and some cell types for simplicity and for a better visualisation.
#' database <- buildDatabase(
#'   dataset_names = paste0('Cytometry', c(2:5, 7:9, 12:17, 19, 21)),
#'     population_ids = c('Monocytes', 'CD4+CD8-', 'Mature SIg Kappa', 'TCRgd-'))
#'
#' templates.optimalFlow <- optimalFlowTemplates(database = database, templates.number = 5,
#' cl.paral = 1)
#'
#' classification.optimalFlow <- optimalFlowClassification(as.data.frame(Cytometry1)[
#'  which(match(Cytometry1$`Population ID (name)`, c('Monocytes', 'CD4+CD8-',
#'                                                  'Mature SIg Kappa', 'TCRgd-'), nomatch = 0) > 0), 1:10], database, templates.optimalFlow,
#'  classif.method = 'matching', cost.function = 'ellipses', cl.paral = 1)
#'
#' f1ScoreVoting(classification.optimalFlow$cluster.vote, classification.optimalFlow$cluster,
#'               as.data.frame(Cytometry1)[which(match(Cytometry1$`Population ID (name)`,
#'                                                     c('Monocytes', 'CD4+CD8-', 'Mature SIg Kappa', 'TCRgd-'), nomatch = 0) > 0), ], 1.01, noise.types)
#' @import optimalFlowData
#' @import rlang
#' @import dplyr
#'
#' @export
#'
f1ScoreVoting <- function(voting, clustering, cytometry, nivel_sup, noise.cells) {
    correspondance <- array(dim = length(voting))
    clusters.to.eliminate <- sort(unique(clustering))[-which(sort(unique(clustering)) %in% names(voting))]
    if (length(clusters.to.eliminate) > 0) {
        indexes.to.eliminate <- which(clustering %in% clusters.to.eliminate)
        clustering <- clustering[-indexes.to.eliminate]
        cytometry <- cytometry[-indexes.to.eliminate, ]
    }
    dim.cyto <- dim(cytometry)[2]
    for (i in 1:length(voting)) {
        vote_new <- voting[[i]]
        noise <- which(vote_new$cell %in% noise.cells)
        if (length(noise) > 0) {
            if (length(noise) == length(vote_new$cell)) {
                correspondance[i] <- "noise"
            } else {
                ruido_s <- sum(vote_new$compound.proportion[noise])
                vote_new <- vote_new[-noise, ]
                if (ruido_s > nivel_sup * vote_new[1, 2]) {
                  correspondance[i] <- "noise"
                } else {
                  if (vote_new[1, 2] < nivel_sup * ruido_s) {
                    correspondance[i] <- paste("undecided", vote_new$cell[1], "noise")
                  } else {
                    if (dim(vote_new)[1] > 2) {
                      if (vote_new[1, 2] < nivel_sup * vote_new[2, 2]) {
                        correspondance[i] <- paste("undecided", vote_new$cell[1], vote_new$cell[2])
                      } else {
                        correspondance[i] <- paste(vote_new$cell[1])
                      }
                    } else {
                      correspondance[i] <- paste(vote_new$cell[1])
                    }
                  }
                }
            }

        } else {
            if (dim(vote_new)[1] > 2) {
                if (vote_new[1, 2] < nivel_sup * vote_new[2, 2]) {
                  correspondance[i] <- paste("undecided", vote_new$cell[1], vote_new$cell[2])
                } else {
                  correspondance[i] <- paste(vote_new$cell[1])
                }
            } else {
                correspondance[i] <- paste(vote_new$cell[1])
            }
        }
    }
    t <- 0
    aciertos <- array(dim = length(voting))
    for (j in names(voting)) {
        t <- t + 1
        if (correspondance[t] == "noise") {
            aciertos[t] <- sum(dplyr::pull(cytometry, dim.cyto)[clustering == j] %in% noise.cells)/length(dplyr::pull(cytometry,
                dim.cyto)[clustering == j])
        } else {
            aciertos[t] <- sum(dplyr::pull(cytometry, dim.cyto)[clustering == j] %in% correspondance[t])/length(dplyr::pull(cytometry,
                dim.cyto)[clustering == j])
        }
    }

    p <- array(dim = c(1, length(unique(correspondance))))
    r <- p
    t <- 0
    for (i in unique(correspondance)) {
        t <- t + 1
        if (i == "noise") {
            p[t] <- sum((aciertos * table(clustering))[correspondance %in% "noise"])/sum(table(clustering)[correspondance %in%
                "noise"])
            r[t] <- sum((aciertos * table(clustering))[correspondance %in% "noise"])/sum(dplyr::pull(cytometry, dim.cyto) %in%
                noise.cells)
        } else {
            p[t] <- sum((aciertos * table(clustering))[correspondance %in% i])/sum(table(clustering)[correspondance %in% i])
            r[t] <- sum((aciertos * table(clustering))[correspondance %in% i])/sum(dplyr::pull(cytometry, dim.cyto) %in% i)
        }
    }
    F1_score <- 2 * p * r/(p + r)
    F1_score_wass <- rbind(F1_score, p, r)
    cels_original <- names(table(dplyr::pull(cytometry, dim.cyto)))[which(table(dplyr::pull(cytometry, dim.cyto)) > 0)]
    cel_no_encontradas <- cels_original[!(cels_original %in% correspondance) & !(cels_original %in% noise.cells)]
    F1_score_wass <- cbind(F1_score_wass, matrix(0, ncol = length(cel_no_encontradas), nrow = 3))
    colnames(F1_score_wass) <- c(unique(correspondance), cel_no_encontradas)
    rownames(F1_score_wass) <- c("F1-score", "Precision", "Recall")
    return(list(F1_score = F1_score_wass, correspondance = correspondance))
}

Try the optimalFlow package in your browser

Any scripts or data that you put into this service are public.

optimalFlow documentation built on Nov. 8, 2020, 6:59 p.m.