There is a good implementation in Mahout and it scales well on Hadoop, so its a pretty good out of the box whole data set learning algorithm. No advantage over

There is a good implementation in Mahout and it scales well on Hadoop, so its a pretty good out of the box whole data set learning algorithm. No advantage over taking a decent statistically valid subset and running C4.5 (or a variant with floating point splits) apart from it's less work.