I'm trying to build a Forest of Trees model using the RP Forest component. The training set is highly imbalanced (26 actives, ~450000 (random assumed) inactives), so I tried equalizing the class size , weighting equally, and bagging 500 trees. I keep getting a "Severe Error caught" error. Reducing the number of trees to 100 doesn't help. In the end, I have to turn off the class size equalization and equal weighting, and drop the number of trees to 50.

Is it possible that for such an unbalanced set, when it tries to build a balanced set for a particular tree that it ends up with a sample that is missing one of the classes (the active one in this case)?

I'm running PP 7.5.2.

Lee