X-Git-Url: https://git.auder.net/?p=valse.git;a=blobdiff_plain;f=pkg%2FR%2Fmain.R;h=13df89fddad635010a1ab0128fd7c23e54df5efe;hp=1908021cd740e91f472d78abed85d4b57997e361;hb=fb3557f39487d9631ffde30f20b70938d2a6ab0c;hpb=f87ff0f5116c0c1c59c5608e46563ff0f79e5d43 diff --git a/pkg/R/main.R b/pkg/R/main.R index 1908021..13df89f 100644 --- a/pkg/R/main.R +++ b/pkg/R/main.R @@ -1,212 +1,167 @@ -#' @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) - { - "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() - { - "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 ? - - ) -) +#' runValse +#' +#' 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 rank.min integer, minimum rank in the low rank procedure, by default = 1 +#' @param rank.max integer, maximum rank in the low rank procedure, by default = 5 +#' @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 thresh real, threshold to say a variable is relevant, by default = 1e-8 +#' @param grid_lambda, a vector with regularization parameters if known, by default numeric(0) +#' @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 +#' @param plot TRUE to plot the selected models after run +#' +#' @return +#' The selected model if enough data are available to estimate it, +#' or a list of models otherwise. +#' +#' @examples +#' n = 50; m = 10; p = 5 +#' beta = array(0, dim=c(p,m,2)) +#' beta[,,1] = 1 +#' beta[,,2] = 2 +#' data = generateXY(n, c(0.4,0.6), rep(0,p), beta, diag(0.5, p), diag(0.5, m)) +#' X = data$X +#' Y = data$Y +#' res = runValse(X, Y) +#' X <- matrix(runif(100), nrow=50) +#' Y <- matrix(runif(100), nrow=50) +#' res = runValse(X, Y) +#' +#' @export +runValse <- function(X, Y, procedure = "LassoMLE", selecMod = "DDSE", gamma = 1, mini = 10, + maxi = 50, eps = 1e-04, kmin = 2, kmax = 3, rank.min = 1, rank.max = 5, ncores_outer = 1, + ncores_inner = 1, thresh = 1e-08, grid_lambda = numeric(0), size_coll_mod = 10, + fast = TRUE, verbose = FALSE, plot = TRUE) +{ + n <- nrow(X) + p <- ncol(X) + m <- ncol(Y) + + 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", "rank.min", "rank.max", "ncores_outer", "ncores_inner", "thresh", + "size_coll_mod", "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, fast) + if (length(grid_lambda) == 0) + { + 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$gamInit, mini, maxi, + gamma, grid_lambda, X, Y, thresh, eps, ncores_inner, fast) + + if (procedure == "LassoMLE") { + if (verbose) + print("run the procedure Lasso-MLE") + # compute parameter estimations, with the Maximum Likelihood Estimator, + # restricted on selected variables. + models <- constructionModelesLassoMLE(P$phiInit, P$rhoInit, P$piInit, + P$gamInit, mini, maxi, gamma, X, Y, eps, S, ncores_inner, fast, verbose) + } else { + if (verbose) + print("run the procedure Lasso-Rank") + # compute parameter estimations, with the Low Rank Estimator, restricted on + # selected variables. + models <- constructionModelesLassoRank(S, k, mini, maxi, X, Y, eps, rank.min, + rank.max, ncores_inner, fast, verbose) + } + # warning! Some models are NULL after running selectVariables + models <- models[sapply(models, function(cell) !is.null(cell))] + models + } + + # List (index k) of lists (index lambda) of models + models_list <- + if (ncores_outer > 1) { + parLapply(cl, kmin:kmax, computeModels) + } else { + lapply(kmin:kmax, computeModels) + } + if (ncores_outer > 1) + parallel::stopCluster(cl) + + if (!requireNamespace("capushe", quietly = TRUE)) + { + warning("'capushe' not available: returning all models") + return(models_list) + } + + # Get summary 'tableauRecap' from models + tableauRecap <- do.call(rbind, lapply(seq_along(models_list), function(i) + { + models <- models_list[[i]] + # For a collection of models (same k, several lambda): + LLH <- sapply(models, function(model) model$llh[1]) + k <- length(models[[1]]$pi) + sumPen <- sapply(models, function(model) k * (dim(model$rho)[1] + sum(model$phi[, + , 1] != 0) + 1) - 1) + data.frame(model = paste(i, ".", seq_along(models), sep = ""), pen = sumPen/n, + complexity = sumPen, contrast = -LLH) + })) + tableauRecap <- tableauRecap[which(tableauRecap[, 4] != Inf), ] + if (verbose) + print(tableauRecap) + + if (nrow(tableauRecap) > 10) { + modSel <- capushe::capushe(tableauRecap, n) + indModSel <- if (selecMod == "DDSE") + { + as.numeric(modSel@DDSE@model) + } else if (selecMod == "Djump") + { + as.numeric(modSel@Djump@model) + } else if (selecMod == "BIC") + { + modSel@BIC_capushe$model + } else if (selecMod == "AIC") + { + modSel@AIC_capushe$model + } + listMod <- as.integer(unlist(strsplit(as.character(indModSel), "[.]"))) + modelSel <- models_list[[listMod[1]]][[listMod[2]]] + modelSel$models <- tableauRecap + + if (plot) + print(plot_valse(X, Y, modelSel)) + return(modelSel) + } + tableauRecap +}