Skip to content

Only one class shown in SVM plot?

6 messages · Meffy, Jessica Streicher

#
Hello users!
I'm calculating a simple model using svm(...) from the e1071 package. So far
so good, with a linear kernel I'm getting 5 SVs. When plotting the result I
see very well separated data clouds, but the underlying color is constantly
pink, so as far as I understand no class separation is shown...
I would be happy if anyone could explain me this behaviour because I think I
have a little knot in my brain here...
Thanks in advance!
Greetings, Matthias 



--
View this message in context: http://r.789695.n4.nabble.com/Only-one-class-shown-in-SVM-plot-tp4637782.html
Sent from the R help mailing list archive at Nabble.com.
#
example!!!!

but a wild guess: if your class information is numeric the default is to do eps-regression, not classification. Use factors or specify the type you want to use.

?svm

might help there.
On 25.07.2012, at 15:31, Meffy wrote:

            
#
Ok, here a simple example. The file 
http://r.789695.n4.nabble.com/file/n4637924/test.csv test.csv  has 400 lines
containing 20 columns (1. column is class label, the other 19 are the
features). 
So what I'm doing is
/
data <- read.csv(file="test.csv", head=F, sep=",")

names(data) <- c("Class","V1", "V2", "V3", "V4", "V5", "V6", "V7", "V8",
"V9", "V10", "V11", "V12", "V13", "V14", "V15", "V16", "V17", "V18", "V19")

model <- svm(as.factor(Class)~., data=data, kernel="linear")
/

This gives me the result

/
Parameters:
   SVM-Type:  C-classification 
 SVM-Kernel:  linear 
       cost:  1 
      gamma:  0.05263158 

Number of Support Vectors:  5
/
When plotting this with

/
plot(model, data, V2~V1)
/

I'm getting
http://r.789695.n4.nabble.com/file/n4637924/svm_result.png 
Where am I wrong here??





--
View this message in context: http://r.789695.n4.nabble.com/Only-one-class-shown-in-SVM-plot-tp4637782p4637924.html
Sent from the R help mailing list archive at Nabble.com.
#
Now i'm scratching my head as well, thought it might have to do with scaling at first, so i turned it off, and also tried scaling the data for the plot instead, but to no avail, it just switches the color, but doesn't show the correct contours.

And it is at least predicting the stuff right, so its doesn't seem to be a problem with the model.
On 26.07.2012, at 15:00, Meffy wrote:

            
4 days later