This exercise investigates the usage of function approximators in RL prediction tasks, which allows to estimate the state value when the state space is continuous.
OpenAI's MountainCar is considered as the environment.
- linear function approximation and enhancement by feature engineering
- optimally fitted function approximation using recursive least squares TD
- nonlinear function approximation using artificial neural networks (this serves as a demonstation example for the usage of
Tensorflowin RL)