I would like to test multiple WEKA (and also others) classifiers on my dataset. The way I dealt with this problem was to add nodes with different classifiers (with predictors and scorers), but with increasing numbers of predictors to test, this workflow is getting unusable (lot of connections, then concatenate results etc).
Do you have any idea how to deal with this problem? The ideal solution would be to loop through different classifiers names and parameters and to enter it via flow variables to weka node (eg FilteredClassifier). But it is not possible (I think).
Any idea of how to simplify this problem would be appreciated.
In KNIME it would hard to design, since you would need to programmatically call the Weka library, which you can always do for example with the Java Snippet or External Tools node. However, I would also recommend having a look to the meta learning Weka category to see if you find an interesting meta leanring schema.
Could you please give me some hints about how to pull this off with Java? RWeka would be nice and comfortable, but apparently it doesn't support my favourite option (namely RegressionByDiscretization) - and it seems there's still no way to run a Weka-centric parameter search with workflow varibles and regular Weka nodes. :-(
In addition, I would add, it is sometimes very important to test parameters on a large scale. Currently with weka the only option is to copy the classifier and the predictor then to change the parameters manually. But at the add you have many nodes and it is difficult to know which parameters you set to a specific node.
So maybe, if we could have access to the parameter values in the flow variables, it would help a lot to test the methods.