I am performing k-fold partitions and generating a number of tree learner models, but I would like to run the predictor on testing data only once, so is there a way to aggregate the models into one?
If you have several models all having a score lets say between 0 and 1 what they give you is a rank. You could assign a rank with the Java Snippet and or you could normalise the score between 0-100 (or 1). And then add up the ranks for every case in your (new) file and divide the rank by the number of models you have. So you would have an average rank which you could then rank again into a 0.0-1.0 score.
… or alternatively: Simply train a final forest model with all instances after you have finished x-validation?
You can combine multiple PMML models into a single one. You can find a description either here in our blogpost https://www.knime.com/blog/pmml-ensembles-and-knime or in the paper we published about it: https://kdd13pmml.files.wordpress.com/2013/07/ensemblepmml.pdf
Best wishes, Iris