Ensemble Learning nodes


After updating to KNIME 2.4.1, I still believe there are a few bugs remaining in the Ensemble nodes (unless of course I am missing using them);

- With the Boosting Learner Loop End, I am getting the error message "Prediction Error too big, Finishing ...", however the model stats are Model Weight 2.926, and Model Error 0.097. This doesnt seem to indicate a model weight close to zero, or a large model error. I have used other datasets with model errors of 0.500 and the loop continues, so why is the loop ending with such a low model error, and yet the warning says the error is too big.

- Also, I still believe in the "Delegating Metanode" that the Column Filter node should be setup differently. It should be set to have the prediction column only in the Exclude list, with "Enforce Exclusion" selected. This will have the effect that no matter which dataset is used, only the prediction column will be excluded. The way its currently setup means that any dataset which differs from the "Cluster Membership" dataset means all columns get excluded.


The boosting node will of course not include the last iteration where the error is too big. So you don't really see it. Also the stopping criterion is not exactly based on the error but on the model weight. Therefore the error may be slightly larger or smaller based on the dataset and the previous iterations.

Dear Simon,


I redid basically the delegating node. I hope it's now easier to understand how it should be defined.

You only need to adjust the joiner settings.

Best, Iris