This is a C++ implementation of a boosted decision tree using AdaBoost for boosting.  Though the implementation is bug-free, it does suffer from a critical design drawback: data is stored in a special data structure designed to make repeated sorting efficient (by sorting subsets of data within the container, rather than creating new containers each time a subset is to be sorted).  The problem is that data is that the data is not sorted directly; instead, indices into the data are sorted.  This approach ruins cache coherency, making performance much slower than it should be.