Library for gradient boosting tree
If you're looking for a python version, the latest release of scikit-learn features gradient boosted regression trees for classification and regression (docs).
It is similar to R's gbm package - gbm is faster for (least-squares) regression wheres scikit-learn's implementation is faster at test-time and when your number of features > 1000.
These don't neccessarily meet all your preferences, but there's also:
- Treenet a commercialization and extension of Jerome Friedman's original implementation. Not open source but we've found it to work pretty well
- R gbm package for gradient boosted trees specifically.
I would recommend xgboost (which did not exist by the time the question was asked), which is an open source R/python package.
It is currently among the fastest gradient boosting tree methods that exists, allows for regression/classification, supports sparse matrices...