I have a question though regarding the error: "libsvm classes are not in classpath"
Now, i am very well aware of the necessity of addinge the libsvmjar to the classpath for the standalone weka...but how does this work for weka in knime? I tried adding it to the knime classpath in /knime/preferences/java/Buildpath/Classpath variables, yet that didnt work...
Nope, i need the Weka boosters, parametercv selection etc... I think all one needs to do is include the libsvm in the classpath of the weka plugins while exporting.... could you get back to me if you solved the issue? (christian.koch@pharma.ethz.ch) Thanks!
The next minor release will contain libsvm-support for Weka. In the meantime, if it's really urgent, you may copy the libsvm.jar into the lib-folder of the weka_3.6 plug-in and add a reference to this jar in the META-INF/MANIFEST.MF under Bundle-Classpath. This should work, although I'm not 100% sure.
Cool thank you so much! Another question: I'm trying to provide Parameters to the weka classifiers via the FlowVariables. As there is only 'Serialized Classifier', I assume I have to use that flow variable? What form of String do I provide? I tried all kinds of weka lines e.g. MultilayerPerceptron -B -C - E 100 etc, could you provide an example?
I second that question! I'd love to loop over several bin settings using "RegressionByDiscretization", but I didn't get it done, either. Are we just misinterpreting the usage of that flow variable?
Hi E., The Weka nodes currently can't be used together with the KNIME Flow Variables due to its internal model/settings representation. We are looking into a solution and trying to improve those nodes.
> Can you, in the meantime, use the KNIME LibSVM integration instead?
The problem is that KNIME's version of LibSVM doesnt support C and gamma values >100 (while oryginal LibSVM and Weka SVM does). The problem was indicated in this thread, but till now I hadn't found a solution for this.