Fix numerical problems in EMGLLF (R version)

[valse.git] / pkg / R / selectVariables.R

diff --git a/pkg/R/selectVariables.R b/pkg/R/selectVariables.R
index ce7d3b3..bfe4042 100644 (file)
--- a/pkg/R/selectVariables.R
+++ b/pkg/R/selectVariables.R
@@ -1,19 +1,20 @@
-#' selectVariables
+#' selectVariables 
 #'
 #' It is a function which construct, for a given lambda, the sets of relevant variables.
 #'
 #' @param phiInit an initial estimator for phi (size: p*m*k)
 #' @param rhoInit an initial estimator for rho (size: m*m*k)
-#' @param piInit       an initial estimator for pi (size : k)
+#' @param piInit\tan initial estimator for pi (size : k)
 #' @param gamInit an initial estimator for gamma
-#' @param mini         minimum number of iterations in EM algorithm
-#' @param maxi         maximum number of iterations in EM algorithm
-#' @param gamma         power in the penalty
+#' @param mini\t\tminimum number of iterations in EM algorithm
+#' @param maxi\t\tmaximum number of iterations in EM algorithm
+#' @param gamma\t power in the penalty
 #' @param glambda grid of regularization parameters
-#' @param X                     matrix of regressors
-#' @param Y                     matrix of responses
-#' @param thres         threshold to consider a coefficient to be equal to 0
-#' @param tau           threshold to say that EM algorithm has converged
+#' @param X\t\t\t matrix of regressors
+#' @param Y\t\t\t matrix of responses
+#' @param thresh real, threshold to say a variable is relevant, by default = 1e-8
+#' @param eps\t\t threshold to say that EM algorithm has converged
+#' @param ncores Number or cores for parallel execution (1 to disable)
 #'
 #' @return a list of outputs, for each lambda in grid: selected,Rho,Pi
 #'
@@ -21,42 +22,53 @@
 #'
 #' @export
 #'
-selectVariables = function(phiInit,rhoInit,piInit,gamInit,mini,maxi,gamma,glambda,
-       X,Y,thresh,tau, ncores=1) #ncores==1 ==> no //
+selectVariables <- function(phiInit, rhoInit, piInit, gamInit, mini, maxi, gamma, 
+  glambda, X, Y, thresh = 1e-08, eps, ncores = 3, fast)
 {
-       if (ncores > 1)
-       {
-               cl = parallel::makeCluster(ncores)
-               parallel::clusterExport(cl=cl,
-                       varlist=c("phiInit","rhoInit","gamInit","mini","maxi","glambda","X","Y","thresh","tau"),
-                       envir=environment())
-       }
+  if (ncores > 1) {
+    cl <- parallel::makeCluster(ncores, outfile = "")
+    parallel::clusterExport(cl = cl, varlist = c("phiInit", "rhoInit", "gamInit", 
+      "mini", "maxi", "glambda", "X", "Y", "thresh", "eps"), envir = environment())
+  }
 
-       # Calcul pour un lambda
-       computeCoefs <-function(lambda)
-       {
-               params = EMGLLF(phiInit,rhoInit,piInit,gamInit,mini,maxi,gamma,lambda,X,Y,tau)
+  # Computation for a fixed lambda
+  computeCoefs <- function(lambda)
+  {
+    params <- EMGLLF(phiInit, rhoInit, piInit, gamInit, mini, maxi, gamma, lambda, 
+      X, Y, eps, fast)
 
-               p = dim(phiInit)[1]
-               m = dim(phiInit)[2]
+    p <- dim(phiInit)[1]
+    m <- dim(phiInit)[2]
 
-               #selectedVariables: list where element j contains vector of selected variables in [1,m]
-               selectedVariables = lapply(1:p, function(j) {
-                       #from boolean matrix mxk of selected variables obtain the corresponding boolean m-vector,
-                       #and finally return the corresponding indices
-                       seq_len(m)[ apply( abs(params$phi[j,,]) > thresh, 1, any ) ]
-               })
+    # selectedVariables: list where element j contains vector of selected variables
+    # in [1,m]
+    selectedVariables <- lapply(1:p, function(j) {
+      # from boolean matrix mxk of selected variables obtain the corresponding boolean
+      # m-vector, and finally return the corresponding indices
+      seq_len(m)[apply(abs(params$phi[j, , ]) > thresh, 1, any)]
+    })
 
-               list("selected"=selectedVariables,"Rho"=params$Rho,"Pi"=params$Pi)
-       }
+    list(selected = selectedVariables, Rho = params$rho, Pi = params$pi)
+  }
 
-       # Pour chaque lambda de la grille, on calcule les coefficients
-       out <-
-               if (ncores > 1)
-                       parLapply(cl, seq_along(glambda, computeCoefs)
-               else
-                       lapply(seq_along(glambda), computeCoefs)
-       if (ncores > 1)
-               parallel::stopCluster(cl)
-       out
+  # For each lambda in the grid, we compute the coefficients
+  out <-
+    if (ncores > 1) {
+      parLapply(cl, glambda, computeCoefs)
+    } else {
+      lapply(glambda, computeCoefs)
+    }
+  if (ncores > 1) 
+    parallel::stopCluster(cl)
+  # Suppress models which are computed twice En fait, ca ca fait la comparaison de
+  # tous les parametres On veut juste supprimer ceux qui ont les memes variables
+  # sélectionnées
+  # sha1_array <- lapply(out, digest::sha1) out[ duplicated(sha1_array) ]
+  selec <- lapply(out, function(model) model$selected)
+  ind_dup <- duplicated(selec)
+  ind_uniq <- which(!ind_dup)
+  out2 <- list()
+  for (l in 1:length(ind_uniq))
+    out2[[l]] <- out[[ind_uniq[l]]]
+  out2
 }