Yes, that's true.
It was pulled back, went through lots of refactoring. The rationale was that the code would go through so much revision that it needed to be private until it stabilized. Pattern now has much better support for parallelizing algorithms for machine learning -- including some changes to Cascading internally. PMML is a formal specification for workflows, and much of that maps directly to pipeline constructs in Cascading. We have a hunch that the new support for ensembles is particularly innovative.
The GitHub repo for Pattern is just about to be pushed public again. Meanwhile, there are a few companies with this code in production, and support for new algorithms has been added. We'll also start to move into model creation, instead of just model scoring.