X-Git-Url: https://git.auder.net/?p=valse.git;a=blobdiff_plain;f=pkg%2FR%2Fmain.R;h=701a2c93e78262950eec17d3013ee97f2a86ac3d;hp=1908021cd740e91f472d78abed85d4b57997e361;hb=0e0fb59a6ea0a975d1a9059153aa27f54458bf95;hpb=f87ff0f5116c0c1c59c5608e46563ff0f79e5d43 diff --git a/pkg/R/main.R b/pkg/R/main.R index 1908021..701a2c9 100644 --- a/pkg/R/main.R +++ b/pkg/R/main.R @@ -1,212 +1,145 @@ -#' @useDynLib valse - -Valse = setRefClass( - Class = "Valse", - - fields = c( - # User defined - - # regression data (size n*p, where n is the number of observations, - # and p is the number of regressors) - X = "matrix", - # response data (size n*m, where n is the number of observations, - # and m is the number of responses) - Y = "matrix", - - # Optionally user defined (some default values) - - # power in the penalty - gamma = "numeric", - # minimum number of iterations for EM algorithm - mini = "integer", - # maximum number of iterations for EM algorithm - maxi = "integer", - # threshold for stopping EM algorithm - eps = "numeric", - # minimum number of components in the mixture - kmin = "integer", - # maximum number of components in the mixture - kmax = "integer", - # ranks for the Lasso-Rank procedure - rank.min = "integer", - rank.max = "integer", - - # Computed through the workflow - - # initialisation for the reparametrized conditional mean parameter - phiInit = "numeric", - # initialisation for the reparametrized variance parameter - rhoInit = "numeric", - # initialisation for the proportions - piInit = "numeric", - # initialisation for the allocations probabilities in each component - tauInit = "numeric", - # values for the regularization parameter grid - gridLambda = "numeric", - # je ne crois pas vraiment qu'il faille les mettre en sortie, d'autant plus qu'on construit - # une matrice A1 et A2 pour chaque k, et elles sont grandes, donc ca coute un peu cher ... - A1 = "integer", - A2 = "integer", - # collection of estimations for the reparametrized conditional mean parameters - Phi = "numeric", - # collection of estimations for the reparametrized variance parameters - Rho = "numeric", - # collection of estimations for the proportions parameters - Pi = "numeric", - - #immutable (TODO:?) - thresh = "numeric" - ), - - methods = list( - ####################### - #initialize main object - ####################### - initialize = function(X,Y,...) - { - "Initialize Valse object" - - callSuper(...) - - X <<- X - Y <<- Y - gamma <<- ifelse (hasArg("gamma"), gamma, 1.) - mini <<- ifelse (hasArg("mini"), mini, as.integer(5)) - maxi <<- ifelse (hasArg("maxi"), maxi, as.integer(10)) - eps <<- ifelse (hasArg("eps"), eps, 1e-6) - kmin <<- ifelse (hasArg("kmin"), kmin, as.integer(2)) - kmax <<- ifelse (hasArg("kmax"), kmax, as.integer(3)) - rank.min <<- ifelse (hasArg("rank.min"), rank.min, as.integer(2)) - rank.max <<- ifelse (hasArg("rank.max"), rank.max, as.integer(3)) - thresh <<- 1e-15 #immutable (TODO:?) - }, - - ################################## - #core workflow: compute all models - ################################## - - initParameters = function(k) +#' valse +#' +#' Main function +#' +#' @param X matrix of covariates (of size n*p) +#' @param Y matrix of responses (of size n*m) +#' @param procedure among 'LassoMLE' or 'LassoRank' +#' @param selecMod method to select a model among 'DDSE', 'DJump', 'BIC' or 'AIC' +#' @param gamma integer for the power in the penaly, by default = 1 +#' @param mini integer, minimum number of iterations in the EM algorithm, by default = 10 +#' @param maxi integer, maximum number of iterations in the EM algorithm, by default = 100 +#' @param eps real, threshold to say the EM algorithm converges, by default = 1e-4 +#' @param kmin integer, minimum number of clusters, by default = 2 +#' @param kmax integer, maximum number of clusters, by default = 10 +#' @param rang.min integer, minimum rank in the low rank procedure, by default = 1 +#' @param rang.max integer, maximum rank in the +#' @param ncores_outer Number of cores for the outer loop on k +#' @param ncores_inner Number of cores for the inner loop on lambda +#' @param size_coll_mod (Maximum) size of a collection of models +#' @param fast TRUE to use compiled C code, FALSE for R code only +#' @param verbose TRUE to show some execution traces +#' +#' @return a list with estimators of parameters +#' +#' @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=4, rang.min=1, rang.max=10, ncores_outer=1, ncores_inner=1, + size_coll_mod=50, fast=TRUE, verbose=FALSE, plot = TRUE) +{ + p = dim(X)[2] + m = dim(Y)[2] + n = dim(X)[1] + + if (verbose) + print("main loop: over all k and all lambda") + + if (ncores_outer > 1) + { + 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") ) + } + + # Compute models with k components + computeModels <- function(k) + { + if (ncores_outer > 1) + require("valse") #nodes start with an empty environment + + if (verbose) + print(paste("Parameters initialization for k =",k)) + #smallEM initializes parameters by k-means and regression model in each component, + #doing this 20 times, and keeping the values maximizing the likelihood after 10 + #iterations of the EM algorithm. + P = initSmallEM(k, X, Y) + grid_lambda <- computeGridLambda(P$phiInit, P$rhoInit, P$piInit, P$gamInit, X, Y, + gamma, mini, maxi, eps, fast) + if (length(grid_lambda)>size_coll_mod) + grid_lambda = grid_lambda[seq(1, length(grid_lambda), length.out = size_coll_mod)] + + if (verbose) + print("Compute relevant parameters") + #select variables according to each regularization parameter + #from the grid: S$selected corresponding to selected variables + S = selectVariables(P$phiInit, P$rhoInit, P$piInit, P$gamInit, mini, maxi, gamma, + grid_lambda, X, Y, 1e-8, eps, ncores_inner, fast) #TODO: 1e-8 as arg?! eps? + + if (procedure == 'LassoMLE') { - "Parameters initialization" - - #smallEM initializes parameters by k-means and regression model in each component, - #doing this 20 times, and keeping the values maximizing the likelihood after 10 - #iterations of the EM algorithm. - init = initSmallEM(k,X,Y) - phiInit <<- init$phi0 - rhoInit <<- init$rho0 - piInit <<- init$pi0 - tauInit <<- init$tau0 - }, - - computeGridLambda = function() + if (verbose) + print('run the procedure Lasso-MLE') + #compute parameter estimations, with the Maximum Likelihood + #Estimator, restricted on selected variables. + models <- constructionModelesLassoMLE(P$phiInit, P$rhoInit, P$piInit, P$gamInit, + mini, maxi, gamma, X, Y, thresh, eps, S, ncores_inner, artefact=1e3, fast, verbose) + } + else { - "computation of the regularization grid" - #(according to explicit formula given by EM algorithm) - - gridLambda <<- gridLambda(phiInit,rhoInit,piInit,tauInit,X,Y,gamma,mini,maxi,eps) - }, - - computeRelevantParameters = function() - { - "Compute relevant parameters" - - #select variables according to each regularization parameter - #from the grid: A1 corresponding to selected variables, and - #A2 corresponding to unselected variables. - params = selectiontotale( - phiInit,rhoInit,piInit,tauInit,mini,maxi,gamma,gridLambda,X,Y,thresh,eps) - A1 <<- params$A1 - A2 <<- params$A2 - Rho <<- params$Rho - Pi <<- params$Pi - }, - - runProcedure1 = function() - { - "Run procedure 1 [EMGLLF]" - - #compute parameter estimations, with the Maximum Likelihood - #Estimator, restricted on selected variables. - return ( constructionModelesLassoMLE( - phiInit,rhoInit,piInit,tauInit,mini,maxi,gamma,gridLambda,X,Y,thresh,eps,A1,A2) ) - }, - - runProcedure2 = function() - { - "Run procedure 2 [EMGrank]" - - #compute parameter estimations, with the Low Rank - #Estimator, restricted on selected variables. - return ( constructionModelesLassoRank(Pi,Rho,mini,maxi,X,Y,eps, - A1,rank.min,rank.max) ) - }, - - run = function() - { - "main loop: over all k and all lambda" - - # Run the whole procedure, 1 with the - #maximum likelihood refitting, and 2 with the Low Rank refitting. - p = dim(phiInit)[1] - m = dim(phiInit)[2] - for (k in kmin:kmax) - { - print(k) - initParameters(k) - computeGridLambda() - computeRelevantParameters() - if (procedure == 1) - { - r1 = runProcedure1() - Phi2 = Phi - Rho2 = Rho - Pi2 = Pi - p = ncol(X) - m = ncol(Y) - if (is.null(dim(Phi2))) #test was: size(Phi2) == 0 - { - Phi[,,1:k] <<- r1$phi - Rho[,,1:k] <<- r1$rho - Pi[1:k,] <<- r1$pi - } else - { - Phi <<- array(0., dim=c(p,m,kmax,dim(Phi2)[4]+dim(r1$phi)[4])) - Phi[,,1:(dim(Phi2)[3]),1:(dim(Phi2)[4])] <<- Phi2 - Phi[,,1:k,dim(Phi2)[4]+1] <<- r1$phi - Rho <<- array(0., dim=c(m,m,kmax,dim(Rho2)[4]+dim(r1$rho)[4])) - Rho[,,1:(dim(Rho2)[3]),1:(dim(Rho2)[4])] <<- Rho2 - Rho[,,1:k,dim(Rho2)[4]+1] <<- r1$rho - Pi <<- array(0., dim=c(kmax,dim(Pi2)[2]+dim(r1$pi)[2])) - Pi[1:nrow(Pi2),1:ncol(Pi2)] <<- Pi2 - Pi[1:k,ncol(Pi2)+1] <<- r1$pi - } - } else - { - phi = runProcedure2()$phi - Phi2 = Phi - if (dim(Phi2)[1] == 0) - { - Phi[,,1:k,] <<- phi - } else - { - Phi <<- array(0., dim=c(p,m,kmax,dim(Phi2)[4]+dim(phi)[4])) - Phi[,,1:(dim(Phi2)[3]),1:(dim(Phi2)[4])] <<- Phi2 - Phi[,,1:k,-(1:(dim(Phi2)[4]))] <<- phi - } - } - } - } - - ################################################## - #TODO: pruning: select only one (or a few best ?!) model - ################################################## - # - # function[model] selectModel( - # #TODO - # #model = odel(...) - # end - # Give at least the slope heuristic and BIC, and AIC ? - - ) -) + if (verbose) + print('run the procedure Lasso-Rank') + #compute parameter estimations, with the Low Rank + #Estimator, restricted on selected variables. + models <- constructionModelesLassoRank(S$Pi, S$Rho, mini, maxi, X, Y, eps, A1, + rank.min, rank.max, ncores_inner, fast, verbose) + } + #attention certains modeles sont NULL après selectVariables + models = models[sapply(models, function(cell) !is.null(cell))] + models + } + + # List (index k) of lists (index lambda) of models + models_list <- + if (ncores_outer > 1) + parLapply(cl, kmin:kmax, computeModels) + else + lapply(kmin:kmax, computeModels) + if (ncores_outer > 1) + parallel::stopCluster(cl) + + if (! requireNamespace("capushe", quietly=TRUE)) + { + warning("'capushe' not available: returning all models") + return (models_list) + } + + # Get summary "tableauRecap" from models + tableauRecap = do.call( rbind, lapply( seq_along(models_list), function(i) { + models <- models_list[[i]] + #Pour un groupe de modeles (même k, différents lambda): + LLH <- sapply( models, function(model) model$llh[1] ) + k = length(models[[1]]$pi) + # TODO: chuis pas sûr du tout des lignes suivantes... + # J'ai l'impression qu'il manque des infos + ## C'est surtout que la pénalité est la mauvaise, la c'est celle du Lasso, nous on veut ici + ##celle de l'heuristique de pentes + #sumPen = sapply( models, function(model) + # sum( model$pi^gamma * sapply(1:k, function(r) sum(abs(model$phi[,,r]))) ) ) + sumPen = sapply(models, function(model) + k*(dim(model$rho)[1]+sum(model$phi[,,1]!=0)+1)-1) + data.frame(model=paste(i,".",seq_along(models),sep=""), + pen=sumPen/n, complexity=sumPen, contrast=LLH) + } ) ) + + modSel = capushe::capushe(tableauRecap, n) + indModSel <- + if (selecMod == 'DDSE') + as.numeric(modSel@DDSE@model) + else if (selecMod == 'Djump') + as.numeric(modSel@Djump@model) + else if (selecMod == 'BIC') + modSel@BIC_capushe$model + else if (selecMod == 'AIC') + modSel@AIC_capushe$model + + mod = as.character(tableauRecap[indModSel,1]) + listMod = as.integer(unlist(strsplit(mod, "[.]"))) + if (plot){ + print(plot_valse()) + } + models_list[[listMod[1]]][[listMod[2]]] + +}