Question: Changing SVM-Kernal in geNetClassifier function
gravatar for Seymoo
12 months ago by
Seymoo0 wrote:

I am using `geNetClassifier` function of same name package to build a classifier for gene-expression data.

`@classifier` slot of returned function provide information on call as below


svm.default(x = t(esetFilteredDataFrame[buildGenesVector, trainSamples, drop = FALSE]), y = sampleLabels[trainSamples], kernel = "linear", probability = TRUE, C = 1)`


   SVM-Type:  C-classification 
 SVM-Kernel:  linear 
       cost:  1 
      gamma:  0.002739726


I am wondering if it possible somewhat to change SVM-kernal parameter in the `geNetClassifier ` ?

Thanks in advance

ADD COMMENTlink modified 11 months ago • written 12 months ago by Seymoo0
gravatar for Dario Strbenac
12 months ago by
Dario Strbenac1.4k
Dario Strbenac1.4k wrote:

No. It's also bad design that the developers implemented the cost value a constant. The optimal value would differ based on the input dataset. The algorithm basically does feature selection and inputs the chosen genes into to an SVM classifier. You can implement a similar analysis with ClassifyR or caret.

ADD COMMENTlink written 12 months ago by Dario Strbenac1.4k

Hi Dario,
Thanks a lot for your answer and the the great tip about the `caret` package.

The main reason I am using this package is that it outpust a list of discriminant genes for each of putative sub-classes in my data. Is that an option in the packages you mentioned above?

It's very tempting with all those method for train models aside from all SVM based methods! just for my info, if it has happened to you to try some of those methods, in your experience which one is a better alternative for class discovery in gene expression data? I know SVM, and RandomForest are popular approaches based on literature but I am not sure that the only plausible options one can use



ADD REPLYlink modified 12 months ago • written 12 months ago by Seymoo0

No, the packages I mentioned don't output genes selected based on a multi-class criteria. You would need to write the function yourself. Of course, there are many classifiers you can use and different methods will give better classification performance for different datasets.

ADD REPLYlink written 11 months ago by Dario Strbenac1.4k
gravatar for Seymoo
11 months ago by
Seymoo0 wrote:


Hi Sara,

Thanks for making this great package. I am wondering if you add cost argument in the formula to be set by users or if you could explain how to modify it the way it is right now ? I am using Caret package to find optimal cost to be used in the function and I would like to adjust cost parameter.

ADD COMMENTlink written 11 months ago by Seymoo0
Please log in to add an answer.


Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 2.2.0
Traffic: 280 users visited in the last hour