R/s3methods.R

In mcgfa: Mixtures of Contaminated Gaussian Factor Analyzers

predict.mcgfa <- function(object, y, ...) {
    fit <- object
    if (ncol(y) != ncol(fit$X)) {
        y <- t(y)
    }
    if (ncol(y) != ncol(fit$X)) {
        stop("Input vector has incorrect number of dimensions.")
    }

    m <- nrow(y)

    if (!is.null(attr(fit$X, "scaled:scale"))) {
        y <- scale(y, center = attr(fit$X, "scaled:center"), scale = attr(fit$X, "scaled:scale"))
    }

    dens <- matrix(0,m,fit$G)
    for (g in 1:fit$G) {
        dens[,g] <- cn_dens(y, fit$mu[,g], fit$sigma[,,g], fit$alpha[g], fit$eta[g])
    }
    soft <- dens/matrix(rowSums(dens),m,fit$G)
    hard <- apply(soft,1,which.max)
    list(soft = soft, hard = hard)
}

plot.mcgfa <- function(x, vars, hideBadPoints = FALSE, ...) {
    fit <- x
    if (hideBadPoints) {
        pairs(fit$X[fit$isBad == 0,vars], col = fit$group)
    } else {
        pairs(fit$X[,vars], col = fit$group, pch = fit$isBad+1)
    }
}

cn_dens <- function(x,mu,sigma,alpha,eta) {
    p <- length(mu)
    good_dens <- (2*pi)^(-p/2) * det(sigma)^(-1/2) * exp(-0.5*mahalanobis(x,mu,sigma))
    bad_dens <- (2*pi)^(-p/2) * det(sigma*eta)^(-1/2) * exp(-0.5*mahalanobis(x,mu,sigma*eta))
    alpha*good_dens + (1-alpha) * bad_dens
}