Simulation and Optimization Competitions

Reinforcement learning (RL) is an interesting case among the different branches of machine learning. On the one hand, it is quite demanding from a technical standpoint: various intuitions from supervised learning do not hold, and the associated mathematical apparatus is quite a bit more advanced; on the other hand, it is the easiest one to explain to an outsider or layperson. A simple analogy is teaching your pet (I am very intentionally trying to steer clear of the dogs versus cats debate) to perform tricks: you provide a treat for a trick well done, and refuse it otherwise.

Reinforcement learning was a latecomer to the competition party on Kaggle, but the situation has changed in the last few years with the introduction of simulation competitions. In this chapter, we will describe this new and exciting part of the Kaggle universe. So far – at the time of writing – there have been four Featured competitions and two Playground...