learning-to-learn. Experiments using reinforcement learning to chose model hyperparameters