Sharing of a trained SVM model

I am using an SVM (SVM_train, Bioinformatics toolbox) to classify data and would like to make my final trained SVM models available so that others (who may not have access to Matlab) can try to replicate or make use of the models.

Is there an easy way to do this from the SVM_struct structure?

I was thinking of providing a spreadsheet detailing the selected features, the support vectors, the hyperplane bias and he support vector weights for each model.

Answer 1

[Ilya]

There is no unified "SVM format" shared by different implementations. You can always dump the svmstruct to an ascii file and hope that it won't be too hard to feed these parameters to another SVM implementation. It won't be possible without some manipulation anyhow.

If I wanted to make my results reproducible, I would provide the exact formula used for computing predictions. Code this formula in a separate function and make sure it gives the exact same answer as svmclassify when you run it on your data. Then provide all parameters needed for this formula. This is the best way to make sure you provide full and correct info.

[Barry Greene]

Update: I have settled on sharing a mat-file containing model parameters