R (interactive) Nodes excruciatingly slow in comparison to deprecated local implementation?

Hi all,

I am using the new R nodes in Knime 2.11 to estimate and predict a model from the uplift package. I have the problem that these nodes take a painfully long time to finish. This is not due their runtime in R, which I check by looking at the R stdout after completion. 

An example: When I predict an uplift model with 600,000 observations and 30 predictors, the time needed to predict from the stdout of the R-PredictorNode after prediction is about 1,000 seconds, which is fine for me. The node itself, however stays at the 99% mark, stating "importing Workspace" for many, many hours. In fact, I always thought, that it had crashed. Only when I accidently let something run overnight I saw that is was actually completed. What could be the reason for this massive delay?

When I use the corresponding nodes from the local (deprecated) section of R, it works as advertised and takes about 1,000 seconds. So, the problem seems to lie in the interactive R nodes. Also, they don't execute in parallel, which means that streams that use several predictor nodes (Training/Test/Validation samples), take literally weeks for me.

 

Has anybody else encountered this problem?

 

Thx in advance,

 

Dennis