Module 23

Reinforcement learning

Advanced module (after graduating): agent, reward, Q-learning

When you finish this module you can say:

I can train an agent with rewards: I write the environment as a class and teach it a policy with Q-learning.

By the end of this module you will be able to train an agent that finds its way to the goal on a 4x4 grid world, with nothing but a reward to learn from.

Lessons

6 lessons · ≈ 66 min

1. Agent, environment, reward10 min
2. Return and discount10 min
3. Exploration and exploitation10 min
4. The Q-table and Q-learning12 min
5. Reward design and traps12 min
6. Project and bridge12 min · project

Practice exercises

Short writing exercises, independent of the lessons: Go to the exercises

1. Which plan is worth more?
2. Measure the exploration rate
3. Arrows from a Q-table