open:approaching-almost-any-machine-learning-problem