use multinom instead of glmnet

statistikat · Oct 24, 2014 · e0ab2e9 · e0ab2e9
1 parent 1ff93ab
commit e0ab2e9
Show file tree

Hide file tree

Showing 4 changed files with 5 additions and 8 deletions.
diff --git a/DESCRIPTION b/DESCRIPTION
@@ -12,7 +12,7 @@ Imports:
     robustbase,
     stats,
     sp,
-    vcd,MASS,nnet,e1071,methods,Rcpp,glmnet
+    vcd,MASS,nnet,e1071,methods,Rcpp
 Description: This package introduces new tools for the visualization of missing
     and/or imputed values, which can be used for exploring the data and the
     structure of the missing and/or imputed values. Depending on this structure

diff --git a/NAMESPACE b/NAMESPACE
@@ -102,7 +102,6 @@ import(Rcpp)
 import(colorspace)
 import(data.table)
 import(e1071)
-import(glmnet)
 import(grDevices)
 import(grid)
 import(methods)

diff --git a/R/VIM-package.R b/R/VIM-package.R
@@ -10,7 +10,6 @@
 #' @import grid
 #' @import robustbase
 #' @import colorspace
-#' @import glmnet
 #' @importFrom car bcPower
 #' @importFrom car powerTransform
 #' @importFrom vcd mosaic

diff --git a/R/regressionImp.R b/R/regressionImp.R
@@ -92,9 +92,8 @@ regressionImp_work <- function(formula, family, robust, data,imp_var,imp_suffix,
             else
               mod <- glm(form,data=data,family=fam)
           }else{
-            fam <- "multinomial"
             TFna <- TFna2&!is.na(data[,lhsV])
-            mod <- glmnet(model.matrix(form,data[TFna,]),data[TFna,lhsV],family=fam)
+            mod <- multinom(form,data[TFna,])
           }
         }
         if(imp_var){
@@ -112,11 +111,11 @@ regressionImp_work <- function(formula, family, robust, data,imp_var,imp_suffix,
         tmp <- data[TFna3,]
         tmp[,lhsV] <- 1
         if(nLev>2){
-          modcv <- cv.glmnet(model.matrix(form,data[TFna,]),data[TFna,lhsV],family="multinomial")
-          pre <- predict(mod,newx=model.matrix(form,tmp),type="response",s=modcv$lambda.1se)[,,1]
+
           if(mod_cat){
-            pre <- levels(data[,lhsV])[apply(pre,1,which.max)]
+            pre <- predict(mod,new.data=tmp)
           }else{
+            pre <- predict(mod,new.data=tmp,type="probs")
             pre <- levels(data[,lhsV])[apply(pre,1,function(x)sample(1:length(x),1,prob=x))]
           }
         }else if(nLev==2){