Sutton & Barto RL Book Notes

Good solutions to all chapters: https://github.com/vojtamolda/reinforcement-learning-an-introduction/tree/main

Chapter 2

The k-armed bandit setting isolates exploration vs exploitation.
Action values are estimated as running means of the received rewards, an error-driven update with step size $\frac{1}{n}$; on non-stationary problems a constant step size is preferred.
Exploration via epsilon greedy, optimistic initial values, or upper confidence bound uncertainty bonuses.

Chapter 11

• deadly triad

References

Richard Sutton
reinforcement learning