From: Benjamin Auder Date: Mon, 3 Apr 2017 11:07:37 +0000 (+0200) Subject: work on constructionModeles + main (2 levels or //isation) X-Git-Url: https://git.auder.net/images/assets/doc/html/vendor/DESCRIPTION?a=commitdiff_plain;h=2279a641f2bee1db586e7ab1e13726d111d5daaf;p=valse.git work on constructionModeles + main (2 levels or //isation) --- diff --git a/pkg/R/constructionModelesLassoMLE.R b/pkg/R/constructionModelesLassoMLE.R index d2bb9a5..6c37751 100644 --- a/pkg/R/constructionModelesLassoMLE.R +++ b/pkg/R/constructionModelesLassoMLE.R @@ -1,90 +1,86 @@ -constructionModelesLassoMLE = function(phiInit,rhoInit,piInit,gamInit,mini,maxi,gamma, - X,Y,seuil,tau,selected, parallel = FALSE) +#' constructionModelesLassoMLE +#' +#' TODO: description +#' +#' @param ... +#' +#' @return ... +#' +#' export +constructionModelesLassoMLE = function(phiInit, rhoInit, piInit, gamInit, mini, maxi, + gamma, X, Y, seuil, tau, selected, ncores=3, verbose=FALSE) { - if (parallel) { - #TODO: parameter ncores (chaque tâche peut aussi demander du parallélisme...) - cl = parallel::makeCluster( parallel::detectCores() / 4 ) - parallel::clusterExport(cl=cl, - varlist=c("phiInit","rhoInit","gamInit","mini","maxi","X","Y","seuil","tau"), - envir=environment()) - #Pour chaque lambda de la grille, on calcule les coefficients - out = parLapply( seq_along(glambda), function(lambda) - { - n = dim(X)[1] - p = dim(phiInit)[1] - m = dim(phiInit)[2] - k = dim(phiInit)[3] - - #TODO: phiInit[selected] et X[selected] sont bien sûr faux; par quoi remplacer ? - #lambda == 0 c'est normal ? -> ED : oui, ici on calcule le maximum de vraisembance, donc on ne pénalise plus - res = EMGLLF(phiInit[selected],rhoInit,piInit,gamInit,mini,maxi,gamma,0.,X[selected],Y,tau) - - #comment évaluer la dimension à partir du résultat et de [not]selected ? - #dimension = ... - - #on veut calculer la vraisemblance avec toutes nos estimations - densite = vector("double",n) - for (r in 1:k) - { - delta = Y%*%rho[,,r] - (X[selected]%*%res$phi[selected,,r]) - densite = densite + pi[r] * - det(rho[,,r])/(sqrt(2*base::pi))^m * exp(-tcrossprod(delta)/2.0) - } - llh = c( sum(log(densite[,lambda])), (dimension+m+1)*k-1 ) - list("phi"=res$phi, "rho"=res$rho, "pi"=res$pi, "llh" = llh) - }) - parallel::stopCluster(cl) - out - } - else { - #Pour chaque lambda de la grille, on calcule les coefficients - n = dim(X)[1] - p = dim(phiInit)[1] - m = dim(phiInit)[2] - k = dim(phiInit)[3] - L = length(selected) - phi = list() + if (ncores > 1) + { + cl = parallel::makeCluster(ncores) + parallel::clusterExport( cl, envir=environment(), + varlist=c("phiInit","rhoInit","gamInit","mini","maxi","gamma","X","Y","seuil", + "tau","selected","ncores","verbose") ) + } + + # Individual model computation + computeAtLambda <- function(lambda) + { + if (ncores > 1) + require("valse") #// nodes start with an ampty environment + + if (verbose) + print(paste("Computations for lambda=",lambda)) + + n = dim(X)[1] + p = dim(phiInit)[1] + m = dim(phiInit)[2] + k = dim(phiInit)[3] + + sel.lambda = selected[[lambda]] +# col.sel = which(colSums(sel.lambda)!=0) #if boolean matrix + col.sel <- which( sapply(sel.lambda,length) > 0 ) #if list of selected vars + + if (length(col.sel) == 0) + return (NULL) + + # lambda == 0 because we compute the EMV: no penalization here + res = EMGLLF(phiInit[col.sel,,],rhoInit,piInit,gamInit,mini,maxi,gamma,0, + X[,col.sel],Y,tau) + + # Eval dimension from the result + selected + phiLambda2 = res_EM$phi + rhoLambda = res_EM$rho + piLambda = res_EM$pi phiLambda = array(0, dim = c(p,m,k)) - rho = list() - pi = list() - llh = list() - - out = lapply( seq_along(selected), function(lambda) - { - print(lambda) - sel.lambda = selected[[lambda]] - col.sel = which(colSums(sel.lambda)!=0) - if (length(col.sel)>0){ - res_EM = EMGLLF(phiInit[col.sel,,],rhoInit,piInit,gamInit,mini,maxi,gamma,0.,X[,col.sel],Y,tau) - phiLambda2 = res_EM$phi - rhoLambda = res_EM$rho - piLambda = res_EM$pi - for (j in 1:length(col.sel)){ - phiLambda[col.sel[j],,] = phiLambda2[j,,] - } - - dimension = 0 - for (j in 1:p){ - b = setdiff(1:m, sel.lambda[,j]) - if (length(b) > 0){ - phiLambda[j,b,] = 0.0 - } - dimension = dimension + sum(sel.lambda[,j]!=0) - } - - #on veut calculer la vraisemblance avec toutes nos estimations - densite = vector("double",n) - for (r in 1:k) - { - delta = Y%*%rhoLambda[,,r] - (X[, col.sel]%*%phiLambda[col.sel,,r]) - densite = densite + piLambda[r] * - det(rhoLambda[,,r])/(sqrt(2*base::pi))^m * exp(-tcrossprod(delta)/2.0) - } - llhLambda = c( sum(log(densite)), (dimension+m+1)*k-1 ) - list("phi"= phiLambda, "rho"= rhoLambda, "pi"= piLambda, "llh" = llhLambda) - } - } - ) - return(out) - } + for (j in seq_along(col.sel)) + phiLambda[col.sel[j],,] = phiLambda2[j,,] + + dimension = 0 + for (j in 1:p) + { + b = setdiff(1:m, sel.lambda[,j]) + if (length(b) > 0) + phiLambda[j,b,] = 0.0 + dimension = dimension + sum(sel.lambda[,j]!=0) + } + + # on veut calculer la vraisemblance avec toutes nos estimations + densite = vector("double",n) + for (r in 1:k) + { + delta = Y%*%rhoLambda[,,r] - (X[, col.sel]%*%phiLambda[col.sel,,r]) + densite = densite + piLambda[r] * + det(rhoLambda[,,r])/(sqrt(2*base::pi))^m * exp(-tcrossprod(delta)/2.0) + } + llhLambda = c( sum(log(densite)), (dimension+m+1)*k-1 ) + list("phi"= phiLambda, "rho"= rhoLambda, "pi"= piLambda, "llh" = llhLambda) + } + + #Pour chaque lambda de la grille, on calcule les coefficients + out = + if (ncores > 1) + parLapply(cl, seq_along(glambda), computeAtLambda) + else + lapply(seq_along(glambda), computeAtLambda) + + if (ncores > 1) + parallel::stopCluster(cl) + + out } diff --git a/pkg/R/constructionModelesLassoRank.R b/pkg/R/constructionModelesLassoRank.R index 9254473..c219d75 100644 --- a/pkg/R/constructionModelesLassoRank.R +++ b/pkg/R/constructionModelesLassoRank.R @@ -1,4 +1,14 @@ -constructionModelesLassoRank = function(pi,rho,mini,maxi,X,Y,tau,A1,rangmin,rangmax) +#' constructionModelesLassoRank +#' +#' TODO: description +#' +#' @param ... +#' +#' @return ... +#' +#' export +constructionModelesLassoRank = function(pi, rho, mini, maxi, X, Y, tau, A1, rangmin, + rangmax, ncores, verbose=FALSE) { #get matrix sizes n = dim(X)[1] @@ -27,7 +37,9 @@ constructionModelesLassoRank = function(pi,rho,mini,maxi,X,Y,tau,A1,rangmin,rang # output parameters phi = array(0, dim=c(p,m,k,L*Size)) llh = matrix(0, L*Size, 2) #log-likelihood - for(lambdaIndex in 1:L) + + # TODO: // loop + for(lambdaIndex in 1:L) { # on ne garde que les colonnes actives # 'active' sera l'ensemble des variables informatives diff --git a/pkg/R/main.R b/pkg/R/main.R index 7b78a15..8ce5117 100644 --- a/pkg/R/main.R +++ b/pkg/R/main.R @@ -20,9 +20,9 @@ #' @examples #' #TODO: a few examples #' @export -valse = function(X,Y,procedure = 'LassoMLE',selecMod = 'DDSE',gamma = 1,mini = 10, - maxi = 50,eps = 1e-4,kmin = 2,kmax = 2, - rang.min = 1,rang.max = 10, ncores_k=1, ncores_lambda=3, verbose=FALSE) +valse = function(X, Y, procedure='LassoMLE', selecMod='DDSE', gamma=1, mini=10, maxi=50, + eps=1e-4, kmin=2, kmax=2, rang.min=1, rang.max=10, ncores_outer=1, ncores_inner=3, + verbose=FALSE) { p = dim(X)[2] m = dim(Y)[2] @@ -32,18 +32,18 @@ valse = function(X,Y,procedure = 'LassoMLE',selecMod = 'DDSE',gamma = 1,mini = 1 if (verbose) print("main loop: over all k and all lambda") - if (ncores_k > 1) + if (ncores_outer > 1) { - cl = parallel::makeCluster(ncores_k) + cl = parallel::makeCluster(ncores_outer) parallel::clusterExport( cl=cl, envir=environment(), varlist=c("X","Y","procedure", "selecMod","gamma","mini","maxi","eps","kmin","kmax","rang.min","rang.max", - "ncores_k","ncores_lambda","verbose","p","m","k","tableauRecap") ) + "ncores_outer","ncores_inner","verbose","p","m","k","tableauRecap") ) } # Compute model with k components computeModel <- function(k) { - if (ncores_k > 1) + if (ncores_outer > 1) require("valse") #nodes start with an empty environment if (verbose) @@ -65,7 +65,7 @@ valse = function(X,Y,procedure = 'LassoMLE',selecMod = 'DDSE',gamma = 1,mini = 1 #from the grid: A1 corresponding to selected variables, and #A2 corresponding to unselected variables. S = selectVariables(P$phiInit,P$rhoInit,P$piInit,P$gamInit,mini,maxi,gamma, - grid_lambda,X,Y,1e-8,eps,ncores_lambda) + grid_lambda,X,Y,1e-8,eps,ncores_inner) if (procedure == 'LassoMLE') { @@ -74,12 +74,7 @@ valse = function(X,Y,procedure = 'LassoMLE',selecMod = 'DDSE',gamma = 1,mini = 1 #compute parameter estimations, with the Maximum Likelihood #Estimator, restricted on selected variables. model = constructionModelesLassoMLE(phiInit, rhoInit, piInit, gamInit, mini, - maxi, gamma, X, Y, thresh, eps, S$selected) - llh = matrix(ncol = 2) - for (l in seq_along(model[[k]])) - llh = rbind(llh, model[[k]][[l]]$llh) - LLH = llh[-1,1] - D = llh[-1,2] + maxi, gamma, X, Y, thresh, eps, S$selected, ncores_inner, verbose) } else { @@ -88,25 +83,25 @@ valse = function(X,Y,procedure = 'LassoMLE',selecMod = 'DDSE',gamma = 1,mini = 1 #compute parameter estimations, with the Low Rank #Estimator, restricted on selected variables. model = constructionModelesLassoRank(S$Pi, S$Rho, mini, maxi, X, Y, eps, A1, - rank.min, rank.max) + rank.min, rank.max, ncores_inner, verbose) ################################################ ### Regarder la SUITE - phi = runProcedure2()$phi - Phi2 = Phi - if (dim(Phi2)[1] == 0) - Phi[, , 1:k,] <- phi - else - { - Phi <- array(0, dim = c(p, m, kmax, dim(Phi2)[4] + dim(phi)[4])) - Phi[, , 1:(dim(Phi2)[3]), 1:(dim(Phi2)[4])] <<- Phi2 - Phi[, , 1:k,-(1:(dim(Phi2)[4]))] <<- phi - } +# phi = runProcedure2()$phi +# Phi2 = Phi +# if (dim(Phi2)[1] == 0) +# Phi[, , 1:k,] <- phi +# else +# { +# Phi <- array(0, dim = c(p, m, kmax, dim(Phi2)[4] + dim(phi)[4])) +# Phi[, , 1:(dim(Phi2)[3]), 1:(dim(Phi2)[4])] <<- Phi2 +# Phi[, , 1:k,-(1:(dim(Phi2)[4]))] <<- phi +# } } - tableauRecap[[k]] = matrix(c(LLH, D, rep(k, length(model[[k]])), 1:length(model[[k]])), ncol = 4)) + model } - model <- + model_list <- if (ncores_k > 1) parLapply(cl, kmin:kmax, computeModel) else @@ -114,9 +109,19 @@ valse = function(X,Y,procedure = 'LassoMLE',selecMod = 'DDSE',gamma = 1,mini = 1 if (ncores_k > 1) parallel::stopCluster(cl) + # Get summary "tableauRecap" from models + tableauRecap = t( sapply( seq_along(model_list), function(model) { + llh = matrix(ncol = 2) + for (l in seq_along(model)) + llh = rbind(llh, model[[l]]$llh) + LLH = llh[-1,1] + D = llh[-1,2] + c(LLH, D, rep(k, length(model)), 1:length(model)) + } ) ) + if (verbose) print('Model selection') - tableauRecap = do.call( rbind, tableaurecap ) #stack list cells into a matrix + tableauRecap = do.call( rbind, tableauRecap ) #stack list cells into a matrix tableauRecap = tableauRecap[rowSums(tableauRecap[, 2:4])!=0,] tableauRecap = tableauRecap[(tableauRecap[,1])!=Inf,] data = cbind(1:dim(tableauRecap)[1], tableauRecap[,2], tableauRecap[,2], tableauRecap[,1])