Business and other problems not amenable to deep learning are often best solved by using well-tuned Gradient-boosted decision trees. These methods are, like deep learning, capable of solving arbitrarily complex problems via nonlinear mappings, but can do so without requiring the large training sets and compute-intensive processing that deep learning sometimes can.
This project shows that such methods are supported on Gradient by demonstrating training of gradient-boosted decision trees (GBT) using the well-known open source machine learning (ML) library H2O.
We also show H2O's automated machine learning (AutoML) capability that can search the model hyperparameter tuning space. This can both save the user time required to so do manually, and produce better results by finding hyperparameter combinations that the user may miss. AutoML used in this way can surpass even expert human data scientists in some situations.