From 08f4604c778da8af7e26b52b1d433a6be82c3139 Mon Sep 17 00:00:00 2001 From: Benjamin Auder Date: Wed, 5 Apr 2017 14:00:16 +0200 Subject: [PATCH] 'update' --- CCC.R | 86 +++++++++++++++++ pkg/DESCRIPTION | 1 - pkg/R/A_NAMESPACE.R | 1 - pkg/R/constructionModelesLassoMLE.R | 144 +++++++++++++--------------- pkg/R/filterModels.R | 36 ------- pkg/R/main.R | 36 +++---- 6 files changed, 172 insertions(+), 132 deletions(-) create mode 100644 CCC.R delete mode 100644 pkg/R/filterModels.R diff --git a/CCC.R b/CCC.R new file mode 100644 index 0000000..9a17c08 --- /dev/null +++ b/CCC.R @@ -0,0 +1,86 @@ +#' 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 (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)) + 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, glambda, computeAtLambda) + else + lapply(glambda, computeAtLambda) + + if (ncores > 1) + parallel::stopCluster(cl) + + out +} diff --git a/pkg/DESCRIPTION b/pkg/DESCRIPTION index b13ee14..0a1c30e 100644 --- a/pkg/DESCRIPTION +++ b/pkg/DESCRIPTION @@ -31,7 +31,6 @@ Collate: 'plot.R' 'main.R' 'selectVariables.R' - 'filterModels.R' 'constructionModelesLassoRank.R' 'constructionModelesLassoMLE.R' 'computeGridLambda.R' diff --git a/pkg/R/A_NAMESPACE.R b/pkg/R/A_NAMESPACE.R index dd06c9c..359cf88 100644 --- a/pkg/R/A_NAMESPACE.R +++ b/pkg/R/A_NAMESPACE.R @@ -7,7 +7,6 @@ #' @include computeGridLambda.R #' @include constructionModelesLassoMLE.R #' @include constructionModelesLassoRank.R -#' @include filterModels.R #' @include selectVariables.R #' @include main.R #' @include plot.R diff --git a/pkg/R/constructionModelesLassoMLE.R b/pkg/R/constructionModelesLassoMLE.R index a49529c..e8013a2 100644 --- a/pkg/R/constructionModelesLassoMLE.R +++ b/pkg/R/constructionModelesLassoMLE.R @@ -8,82 +8,72 @@ #' #' export constructionModelesLassoMLE = function(phiInit, rhoInit, piInit, gamInit, mini, maxi, - gamma, X, Y, thresh, tau, S, ncores=3, artefact = 1e3, verbose=FALSE) + gamma, X, Y, thresh, tau, S, ncores=3, artefact = 1e3, verbose=FALSE) { - if (ncores > 1) - { - cl = parallel::makeCluster(ncores) - parallel::clusterExport( cl, envir=environment(), - varlist=c("phiInit","rhoInit","gamInit","mini","maxi","gamma","X","Y","thresh", - "tau","S","ncores","verbose") ) - } - - # Individual model computation - computeAtLambda <- function(lambda) - { - if (ncores > 1) - require("valse") #// nodes start with an empty 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 = S[[lambda]]$selected - # 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)} else { - - # lambda == 0 because we compute the EMV: no penalization here - res_EM = 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$pi - phiLambda = array(0, dim = c(p,m,k)) - for (j in seq_along(col.sel)) - phiLambda[col.sel[j],,] = phiLambda2[j,,] - - dimension = 0 - for (j in 1:p) - { - b = setdiff(1:m, sel.lambda[[j]])## je confonds un peu ligne et colonne : est-ce dans le bon sens ? - ## moi pour la dimension, j'aurai juste mis length(unlist(sel.lambda)) mais je sais pas si c'est rapide - if (length(b) > 0) - phiLambda[j,b,] = 0.0 - dimension = dimension + sum(sel.lambda[[j]]!=0) - } - - # Computation of the loglikelihood - densite = vector("double",n) - for (r in 1:k) - { - delta = (Y%*%rhoLambda[,,r] - (X[, col.sel]%*%phiLambda[col.sel,,r]))/artefact - print(max(delta)) - densite = densite + piLambda[r] * - det(rhoLambda[,,r])/(sqrt(2*base::pi))^m * exp(-tcrossprod(delta)/2.0) - } - llhLambda = c( sum(artefact^2 * log(densite)), (dimension+m+1)*k-1 ) - list("phi"= phiLambda, "rho"= rhoLambda, "pi"= piLambda, "llh" = llhLambda) - } - } - - # For each lambda, computation of the parameters - out = - if (ncores > 1) - parLapply(cl, 1:length(S), computeAtLambda) - else - lapply(1:length(S), computeAtLambda) - - if (ncores > 1) - parallel::stopCluster(cl) - - out + if (ncores > 1) + { + cl = parallel::makeCluster(ncores) + parallel::clusterExport( cl, envir=environment(), + varlist=c("phiInit","rhoInit","gamInit","mini","maxi","gamma","X","Y","thresh", + "tau","S","ncores","verbose") ) + } + + # Individual model computation + computeAtLambda <- function(lambda) + { + if (ncores > 1) + require("valse") #nodes start with an empty environment + + if (verbose) + print(paste("Computations for lambda=",lambda)) + + n = dim(X)[1] + p = dim(phiInit)[1] + m = dim(phiInit)[2] + k = dim(phiInit)[3] + + sel.lambda = S[[lambda]]$selected +# 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$phi + rhoLambda = res$rho + piLambda = res$pi + phiLambda = array(0, dim = c(p,m,k)) + for (j in seq_along(col.sel)) + phiLambda[col.sel[j],,] = phiLambda2[j,,] + dimension = length(unlist(sel.lambda)) + + # Computation of the loglikelihood + densite = vector("double",n) + for (r in 1:k) + { + delta = (Y%*%rhoLambda[,,r] - (X[, col.sel]%*%phiLambda[col.sel,,r]))/artefact + print(max(delta)) + densite = densite + piLambda[r] * + det(rhoLambda[,,r])/(sqrt(2*base::pi))^m * exp(-tcrossprod(delta)/2.0) + } + llhLambda = c( sum(artefact^2 * log(densite)), (dimension+m+1)*k-1 ) + list("phi"= phiLambda, "rho"= rhoLambda, "pi"= piLambda, "llh" = llhLambda) + } + + # For each lambda, computation of the parameters + out = + if (ncores > 1) + parLapply(cl, 1:length(S), computeAtLambda) + else + lapply(1:length(S), computeAtLambda) + + if (ncores > 1) + parallel::stopCluster(cl) + + out } diff --git a/pkg/R/filterModels.R b/pkg/R/filterModels.R deleted file mode 100644 index 2659ed4..0000000 --- a/pkg/R/filterModels.R +++ /dev/null @@ -1,36 +0,0 @@ -#' Among a collection of models, this function constructs a subcollection of models with -#' models having strictly different dimensions, keeping the model which minimizes -#' the likelihood if there were several with the same dimension -#' -#' @param LLF a matrix, the first column corresponds to likelihoods for several models -#' the second column corresponds to the dimensions of the corresponding models. -#' -#' @return a list with indices, a vector of indices selected models, -#' and D1, a vector of corresponding dimensions -#' -#' @export -filterModels = function(LLF) -{ - D = LLF[,2] - D1 = unique(D) - - indices = rep(1, length(D1)) - #select argmax MLE - if (length(D1)>2) - { - for (i in 1:length(D1)) - { - A = c() - for (j in 1:length(D)) - { - if(D[[j]]==D1[[i]]) - a = c(a, LLF[j,1]) - } - b = max(a) - #indices[i] : first indices of the binary vector where u_i ==1 - indices[i] = which.max(LLF == b) - } - } - - return (list(indices=indices,D1=D1)) -} diff --git a/pkg/R/main.R b/pkg/R/main.R index 2cd345d..bff2ec5 100644 --- a/pkg/R/main.R +++ b/pkg/R/main.R @@ -33,10 +33,10 @@ valse = function(X, Y, procedure='LassoMLE', selecMod='DDSE', gamma=1, mini=10, if (ncores_outer > 1) { - cl = parallel::makeCluster(ncores_outer) + cl = parallel::makeCluster(ncores_outer, outfile='') parallel::clusterExport( cl=cl, envir=environment(), varlist=c("X","Y","procedure", "selecMod","gamma","mini","maxi","eps","kmin","kmax","rang.min","rang.max", - "ncores_outer","ncores_inner","verbose","p","m","k","tableauRecap") ) + "ncores_outer","ncores_inner","verbose","p","m") ) } # Compute models with k components @@ -53,7 +53,6 @@ valse = function(X, Y, procedure='LassoMLE', selecMod='DDSE', gamma=1, mini=10, P = initSmallEM(k, X, Y) grid_lambda <- computeGridLambda(P$phiInit, P$rhoInit, P$piInit, P$gamInit, X, Y, gamma, mini, maxi, eps) - # TODO: 100 = magic number if (length(grid_lambda)>size_coll_mod) grid_lambda = grid_lambda[seq(1, length(grid_lambda), length.out = size_coll_mod)] @@ -70,8 +69,8 @@ valse = function(X, Y, procedure='LassoMLE', selecMod='DDSE', gamma=1, mini=10, print('run the procedure Lasso-MLE') #compute parameter estimations, with the Maximum Likelihood #Estimator, restricted on selected variables. - models <- constructionModelesLassoMLE(P$phiInit, P$rhoInit, P$piInit, P$gamInit, mini, - maxi, gamma, X, Y, thresh, eps, S, ncores_inner, artefact = 1e3, verbose) + models <- constructionModelesLassoMLE(P$phiInit, P$rhoInit, P$piInit, P$gamInit, + mini, maxi, gamma, X, Y, thresh, eps, S, ncores_inner, artefact = 1e3, verbose) } else { @@ -82,6 +81,8 @@ valse = function(X, Y, procedure='LassoMLE', selecMod='DDSE', gamma=1, mini=10, models <- constructionModelesLassoRank(S$Pi, S$Rho, mini, maxi, X, Y, eps, A1, rank.min, rank.max, ncores_inner, verbose) } + #attention certains modeles sont NULL après selectVariables + models = models[sapply(models, function(cell) !is.null(cell))] models } @@ -100,18 +101,19 @@ valse = function(X, Y, procedure='LassoMLE', selecMod='DDSE', gamma=1, mini=10, return (models_list) } - # Get summary "tableauRecap" from models ; TODO: jusqu'à ligne 114 à mon avis là c'est faux :/ - tableauRecap = sapply( models_list, function(models) { - llh = do.call(rbind, lapply(models, function(model) model$llh)) - LLH = llh[-1,1] - D = llh[-1,2] - c(LLH, D, rep(k, length(LLH)), 1:length(LLH)) - }) - tableauRecap - if (verbose) - print('Model selection') - tableauRecap = tableauRecap[rowSums(tableauRecap[, 2:4])!=0,] - tableauRecap = tableauRecap[!is.infinite(tableauRecap[,1]),] + # Get summary "tableauRecap" from models + tableauRecap = do.call( rbind, lapply( models_list, function(models) { + #Pour un groupe de modeles (même k, différents lambda): + llh = matrix(ncol = 2) + for (l in seq_along(models)) + llh = rbind(llh, models[[l]]$llh) + LLH = llh[-1,1] + D = llh[-1,2] + k = length(models[[1]]$pi) + cbind(LLH, D, rep(k, length(models)), 1:length(models)) + } ) ) + tableauRecap = tableauRecap[rowSums(tableauRecap[, 2:4])!=0,] + tableauRecap = tableauRecap[(tableauRecap[,1])!=Inf,] data = cbind(1:dim(tableauRecap)[1], tableauRecap[,2], tableauRecap[,2], tableauRecap[,1]) modSel = capushe::capushe(data, n) -- 2.44.0