#' rho : parametre de variance renormalisé, calculé par l'EM
#' pi : parametre des proportions renormalisé, calculé par l'EM
#' LLF : log vraisemblance associée à cet échantillon, pour les valeurs estimées des paramètres
-#' S : ... affec : ...
+#' S : ...
+#' affec : ...
#'
#' @export
EMGLLF <- function(phiInit, rhoInit, piInit, gamInit, mini, maxi, gamma, lambda,
- X, Y, eps, fast = TRUE)
+ X, Y, eps, fast)
{
if (!fast)
{
X, Y, eps, phi = double(p * m * k), rho = double(m * m * k), pi = double(k),
LLF = double(maxi), S = double(p * m * k), affec = integer(n), n, p, m, k,
PACKAGE = "valse")
+ list(phi = phi, rho = rho, pi = pi, llh = llh, S = S, affec=affec)
}
# R version - slow but easy to read
.EMGLLF_R <- function(phiInit, rhoInit, piInit, gamInit, mini, maxi, gamma, lambda,
- X2, Y, eps)
+ X, Y, eps)
{
# Matrix dimensions
- n <- dim(Y)[1]
- if (length(dim(phiInit)) == 2) {
- p <- 1
- m <- dim(phiInit)[1]
- k <- dim(phiInit)[2]
- } else {
- p <- dim(phiInit)[1]
- m <- dim(phiInit)[2]
- k <- dim(phiInit)[3]
- }
- X <- matrix(nrow = n, ncol = p)
- X[1:n, 1:p] <- X2
+ n <- nrow(X)
+ p <- ncol(X)
+ m <- ncol(Y)
+ k <- length(piInit)
+
+ # Adjustments required when p==1 or m==1 (var.sel. or output dim 1)
+ if (p==1 || m==1)
+ phiInit <- array(phiInit, dim=c(p,m,k))
+ if (m==1)
+ rhoInit <- array(rhoInit, dim=c(m,m,k))
# Outputs
- phi <- array(NA, dim = c(p, m, k))
- phi[1:p, , ] <- phiInit
+ phi <- phiInit
rho <- rhoInit
pi <- piInit
llh <- -Inf
ps2 <- array(0, dim = c(p, m, k))
X2 <- array(0, dim = c(n, p, k))
Y2 <- array(0, dim = c(n, m, k))
- EPS <- 1e-15
for (ite in 1:maxi)
{
# t(m) is the largest value in the grid O.1^k such that it is nonincreasing
while (kk < 1000 && -a/n + lambda * sum(pi^gamma * b) <
- -sum(gam2 * log(pi2))/n + lambda * sum(pi2^gamma * b))
+ # na.rm=TRUE to handle 0*log(0)
+ -sum(gam2 * log(pi2), na.rm=TRUE)/n + lambda * sum(pi2^gamma * b))
{
pi2 <- pi + 0.1^kk * (1/n * gam2 - pi)
kk <- kk + 1
{
for (mm in 1:m)
{
- S[j, mm, r] <- -rho[mm, mm, r] * ps2[j, mm, r]
- + sum(phi[-j, mm, r] * Gram2[j, -j, r])
+ S[j, mm, r] <- -rho[mm, mm, r] * ps2[j, mm, r] +
+ sum(phi[-j, mm, r] * Gram2[j, -j, r])
if (abs(S[j, mm, r]) <= n * lambda * (pi[r]^gamma)) {
phi[j, mm, r] <- 0
} else if (S[j, mm, r] > n * lambda * (pi[r]^gamma)) {
}
}
- ######## E step#
+ ## E step
# Precompute det(rho[,,r]) for r in 1...k
- detRho <- sapply(1:k, function(r) det(rho[, , r]))
- gam1 <- matrix(0, nrow = n, ncol = k)
+ detRho <- sapply(1:k, function(r) gdet(rho[, , r]))
+ sumLogLLH <- 0
for (i in 1:n)
{
- # Update gam[,]
- for (r in 1:k)
- {
- gam1[i, r] <- pi[r] * exp(-0.5
- * sum((Y[i, ] %*% rho[, , r] - X[i, ] %*% phi[, , r])^2)) * detRho[r]
- }
+ # Update gam[,]; use log to avoid numerical problems
+ logGam <- sapply(1:k, function(r) {
+ log(pi[r]) + log(detRho[r]) - 0.5 *
+ sum((Y[i, ] %*% rho[, , r] - X[i, ] %*% phi[, , r])^2)
+ })
+
+ logGam <- logGam - max(logGam) #adjust without changing proportions
+ gam[i, ] <- exp(logGam)
+ norm_fact <- sum(gam[i, ])
+ gam[i, ] <- gam[i, ] / norm_fact
+ sumLogLLH <- sumLogLLH + log(norm_fact) - log((2 * base::pi)^(m/2))
}
- gam <- gam1 / rowSums(gam1)
- sumLogLLH <- sum(log(rowSums(gam)) - log((2 * base::pi)^(m/2)))
+
sumPen <- sum(pi^gamma * b)
last_llh <- llh
- llh <- -sumLogLLH/n + lambda * sumPen
+ llh <- -sumLogLLH/n #+ lambda * sumPen
dist <- ifelse(ite == 1, llh, (llh - last_llh)/(1 + abs(llh)))
Dist1 <- max((abs(phi - Phi))/(1 + abs(phi)))
Dist2 <- max((abs(rho - Rho))/(1 + abs(rho)))
Dist3 <- max((abs(pi - Pi))/(1 + abs(Pi)))
dist2 <- max(Dist1, Dist2, Dist3)
- if (ite >= mini && (dist >= eps || dist2 >= sqrt(eps)))
+ if (ite >= mini && (dist >= eps || dist2 >= sqrt(eps)))
break
}
- list(phi = phi, rho = rho, pi = pi, llh = llh, S = S)
+ affec = apply(gam, 1, which.max)
+ list(phi = phi, rho = rho, pi = pi, llh = llh, S = S, affec=affec)
}