Hi Dan-

 You can do k-fold cross validation as follows (I built a super node that makes it tidy - probably is not the most efficient but it works without syntax):

1) Create a new field that is a random number - say N(mu,sigma^2)

2) Pass the data through a binning node that creates deciles (for 10-fold validation or 5 equal groups for 5 fold) based on this new field from step 1.

3) Create 10 derive nodes attached to step 2 (to create in effect 10 separate data streams, NOT one after another)  which says for example if decile =1 then partition ='Testing' else 'Training'. Then after this node add another derive node that creates a field called BY which is equal to '1'. Do this for each of the 10 switching out of course '1' for 2 through 10 (0-9 if I recall how the binning is numbered).

4) Bring the 10 data streams together with an append node.

5) Run your model using as a partition the 'partition' field from step 3 and separate models (split) using the BY field from step 3. This is set in the model node.

 You are in effect creating 9 additional copies of the data set - where in each copy, a10th of the data is for testing and the other 90% is for training. You can add an analysis node after the model to see the accuracy or RMSE or whatever on each of the 10 splits to average them.

I wrote quick from memory but write back if you are not following

Jeff