Hands-On Reinforcement Learning for Games

Exercises

As we progressed through this book, the exercises have morphed from learning exercises to almost research efforts, and that is the case in this chapter. Therefore, the exercises in this chapter are meant for the hardcore RL enthusiast and may not be for everyone:

1. Tune the hyperparameters for one of the sample visual environments in the ML-Agents toolkit.
2. Modify the visual observation standard encoder found in the ML-Agents toolkit to include additional layers or different kernel filter settings.
3. Train an agent with nature_cnn or resnet visual encoder networks and compare their performance with earlier examples using the base visual encoder.
4. Modify the resnet visual encoder to accommodate many more layers or other variations of ...