From 6eae1aa623e71d786384d92a48abe9d348fbb719 Mon Sep 17 00:00:00 2001
From: wethenwethen <wethenwethen@gmail.com>
Date: Sat, 21 Mar 2015 10:03:49 +0800
Subject: [PATCH 1/2] Update msvmRFE.R

make it to support multi-class data.
date: Mar-21-2015
---
 msvmRFE.R | 55 +++++++++++++++++++++++++++++++++++++++++++++++++++----
 1 file changed, 51 insertions(+), 4 deletions(-)

diff --git a/msvmRFE.R b/msvmRFE.R
index 94c1f56..337fce1 100644
--- a/msvmRFE.R
+++ b/msvmRFE.R
@@ -12,7 +12,7 @@ svmRFE.wrap <- function(test.fold, X, ...) {
     return(list(feature.ids=features.ranked, train.data.ids=row.names(train.data), test.data.ids=row.names(test.data)))
 }
 
-svmRFE <- function(X, k=1, halve.above=5000) {
+svmRFE <- function(X, k=1, halve.above=50,...) {
 # Feature selection with Multiple SVM Recursive Feature Elimination (RFE) algorithm
     n = ncol(X) - 1
     
@@ -49,12 +49,14 @@ svmRFE <- function(X, k=1, halve.above=5000) {
             c    = vbar / vsd
         } else {
             # Only do 1 pass (i.e. regular SVM-RFE)
-            w = getWeights(NULL, X[, c(1, 1+i.surviving)])
+            w = getWeights2(NULL, X[, c(1, 1+i.surviving)])
             c = w * w
         }
 
         # Rank the features
-        ranking = sort(c, index.return=T)$ix
+        rankingCriteria = 0
+        for(i in 1:ncol(c))rankingCriteria[i] = mean(c[,i])
+        ranking = sort(rankingCriteria, index.return=T)$ix
         if(length(i.surviving) == 1) {
             ranking = 1
         }
@@ -97,6 +99,41 @@ getWeights <- function(test.fold, X) {
     t(svmModel$coefs) %*% svmModel$SV
 }
 
+
+svm.weights<-function(model){
+  w=0
+  if(model$nclasses==2){
+    w=t(model$coefs)%*%model$SV
+  }else{    #when we deal with OVO svm classification
+    ## compute start-index
+    start <- c(1, cumsum(model$nSV)+1)
+    start <- start[-length(start)]
+    
+    calcw <- function (i,j) {
+      ## ranges for class i and j:
+      ri <- start[i] : (start[i] + model$nSV[i] - 1)
+      rj <- start[j] : (start[j] + model$nSV[j] - 1)
+      
+      ## coefs for (i,j):
+      coef1 <- model$coefs[ri, j-1]
+      coef2 <- model$coefs[rj, i]
+      ## return w values:
+      w=t(coef1)%*%model$SV[ri,]+t(coef2)%*%model$SV[rj,]
+      return(w)
+    }
+    
+    W=NULL
+    for (i in 1 : (model$nclasses - 1)){
+      for (j in (i + 1) : model$nclasses){
+        wi=calcw(i,j)
+        W=rbind(W,wi)
+      }
+    }
+    w=W
+  }
+  return(w)
+}
+
 WriteFeatures <- function(results, input, save=T, file='features_ranked.txt') {
 # Compile feature rankings across multiple folds
     featureID = sort(apply(sapply(results, function(x) sort(x$feature, index.return=T)$ix), 1, mean), index=T)$ix
@@ -140,4 +177,14 @@ PlotErrors <- function(errors, errors2=NULL, no.info=0.5, ylim=range(c(errors, e
     AddLine(errors)
     if(!is.null(errors2)) AddLine(errors2, 'gray30')
     abline(h=no.info, lty=3)
-}
\ No newline at end of file
+}
+
+getWeights2 <- function(test.fold, X) {
+  # Fit a linear SVM model and obtain feature weights
+  train.data = X
+  if(!is.null(test.fold)) train.data = X[-test.fold, ]
+  
+  svmModel = svm(train.data[, -1], train.data[, 1], cost=10, cachesize=500,
+                 scale=F, type="C-classification", kernel="linear")
+  return(svm.weights(svmModel))
+}

From 228421673a3c02236e5c9aeef2c0d59d561d1a45 Mon Sep 17 00:00:00 2001
From: wethenwethen <wethenwethen@gmail.com>
Date: Fri, 27 Mar 2015 20:33:14 +0800
Subject: [PATCH 2/2] Update msvmRFE.R

to make it support multi-class data set.
---
 msvmRFE.R | 17 +++--------------
 1 file changed, 3 insertions(+), 14 deletions(-)

diff --git a/msvmRFE.R b/msvmRFE.R
index 337fce1..4456790 100644
--- a/msvmRFE.R
+++ b/msvmRFE.R
@@ -12,7 +12,7 @@ svmRFE.wrap <- function(test.fold, X, ...) {
     return(list(feature.ids=features.ranked, train.data.ids=row.names(train.data), test.data.ids=row.names(test.data)))
 }
 
-svmRFE <- function(X, k=1, halve.above=50,...) {
+svmRFE <- function(X, k=1, halve.above=5000,...) {
 # Feature selection with Multiple SVM Recursive Feature Elimination (RFE) algorithm
     n = ncol(X) - 1
     
@@ -49,7 +49,7 @@ svmRFE <- function(X, k=1, halve.above=50,...) {
             c    = vbar / vsd
         } else {
             # Only do 1 pass (i.e. regular SVM-RFE)
-            w = getWeights2(NULL, X[, c(1, 1+i.surviving)])
+            w = getWeights(NULL, X[, c(1, 1+i.surviving)])
             c = w * w
         }
 
@@ -88,17 +88,6 @@ svmRFE <- function(X, k=1, halve.above=50,...) {
     return (ranked.list)
 }
 
-getWeights <- function(test.fold, X) {
-# Fit a linear SVM model and obtain feature weights
-    train.data = X
-    if(!is.null(test.fold)) train.data = X[-test.fold, ]
-    
-    svmModel = svm(train.data[, -1], train.data[, 1], cost=10, cachesize=500,
-    scale=F, type="C-classification", kernel="linear")
-    
-    t(svmModel$coefs) %*% svmModel$SV
-}
-
 
 svm.weights<-function(model){
   w=0
@@ -179,7 +168,7 @@ PlotErrors <- function(errors, errors2=NULL, no.info=0.5, ylim=range(c(errors, e
     abline(h=no.info, lty=3)
 }
 
-getWeights2 <- function(test.fold, X) {
+getWeights <- function(test.fold, X) {
   # Fit a linear SVM model and obtain feature weights
   train.data = X
   if(!is.null(test.fold)) train.data = X[-test.fold, ]