1

Could you comment how to handle the following non-linear data (svm regression):

tt <- c(1.38, 1.41, 1.38, 1.57, 1.65, 1.45, 1.38, 1.38, 1.38, 1.69, 2.18, 1.89, 0.00, 0.00, 1.20, 0.00, 1.23, 1.40, 1.38, 1.38, 1.38, 1.08, 1.40, 1.88, 1.76, 1.70, 1.87, 0.00, 1.90, 1.40, 0.00, 1.46, 1.51, 0.01, 1.90, 1.63, 0.00, 0.00, 0.01, 2.00, 1.40, 0.00, 1.69, 1.68, 1.70, 1.40, 1.40, 1.64, 1.98, 2.00, 1.40, 2.00, 2.00, 1.78,1.56, 1.46, 1.69, 1.40, 1.87, 1.38, 0.00, 1.40, 1.43, 1.40, 1.69, 1.69, 1.88, 0.94, 1.69, 1.71, 1.57,1.38, 1.10, 1.70, 2.00, 1.70, 1.08, 1.70, 0.00, 1.70, 1.80,0.00, 1.58, 1.80, 1.69, 1.77, 0.00, 0.00, 1.38, 0.00, 0.00, 1.38, 0.00, 0.00)
pp <- c(4,  6,  6,  5,  6,  5,  4,  4,  4,  5,  7,  5,  6 , 6 , 4,  4,  5 , 4 , 5 , 5 , 5  ,6 , 5 , 5,  6 , 7 , 5,   6 ,  4 , 4 , 6,  6 , 6 , 8,  5,  6 , 6  , 5 , 8,  7 , 6,  6,  5 , 5,  6,  6,  6,  5,  5,  5,  5,  6,  7,  6,  4,  6,  5,  6,  6,  6,  8,  6,  4,  4,  5,  5,  6,  6,  7,  4,  6,  4,  4,  5,  5,  4,  4,  6, 10,  7,  6, 10,  5,  7,  5,  4,  8,  7,  4,  6,  4,  4, 4,  6)
qq <- c(2, 2, 2, 3, 1, 3, 3, 3, 3, 1, 0, 2, 0, 2, 3, 0, 0, 1, 1, 1, 1, 0, 1, 1, 1, 0, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 0, 3, 0, 1, 3, 1, 1, 1, 0, 1, 1, 2, 2, 2, 1, 1, 1, 2, 3, 0, 1, 3, 0, 1, 0, 2, 3, 3, 1, 1, 1, 0, 0, 2, 3, 3, 2, 1, 3, 0, 3, 3, 2, 1, 1, 2, 2, 0, 3, 2, 1, 0, 3, 4, 2, 3, 3, 1)

I tried like this, for example

library(kernlab)
huh <- data.frame(tt,pp,qq)
index <- 1:nrow(huh)
testindex <- sample(index, trunc(length(index)/3))
testset <-huh[testindex,]
trainset <- huh[-testindex,]

mod <- ksvm(tt ~pp+qq, data =trainset,type = "eps-svr", kernel = "rbfdot",kpar ="automatic", C = 10, prob.model = TRUE)

the result looks like

Support Vector Machine object of class "ksvm" 

SV type: eps-svr  (regression) 
 parameter : epsilon = 0.1  cost C = 10 

Gaussian Radial Basis kernel function. 
 Hyperparameter : sigma =  0.637663227203429 

Number of Support Vectors : 55 

Objective Function Value : -224.1407 
Training error : 0.581297 
Laplace distr. width : 1.320399

I can extract coefficients and bias (w and b) but I can't find the slack variables (soft-margin) that define the loss-function. Can you suggest me another option to fit such type of data?

nograpes
  • 18,623
  • 1
  • 44
  • 67
user3104176
  • 11
  • 4

2 Answers2

0

Check out package e1701, functions svm() and predict(). it supports regression svm and various kernels.

Elad663
  • 783
  • 1
  • 5
  • 13
  • I have tried to install it and it says: `Warning in install.packages : package ‘e1701’ is not available (for R version 3.2.0)` – sop Jun 05 '15 at 15:17
  • 2
    @sop see http://stackoverflow.com/q/25721884/903061 for dealing with that warning. – Gregor Thomas Oct 25 '16 at 18:20
0

I think you should try with biocLite("e1071") to install e1071 package into R. you could do like: source("http://bioconductor.org/biocLite.R") biocLite("e1071")

At least I've got e1071 package installed into R.

Hope this can help you a little.

Yinghua Li
  • 1