With a little forethought, we can present topics in reinforcement learning with no need of calculus, probability & statistics, or linear algebra. That's important, because it creates an on-ramp to machine learning four years ahead of schedule.
We will use
Google Colab to examine
Q-learning, by solving a maze. The algorithm focuses on the problem of delayed rewards:
how do you move toward a solution that is several steps into the future? We also encounter a model-free solution:
how can our algorithm possibly tackle other problems, like tic-tac-toe, without changing a single line of code?By constraining the dimensions of our first problem (the maze is 4x4), we can focus on how Q-learning works. I will provide a free & open-source lesson oin GitHub that you are welcome to use in your classroom. The lesson assumes one semester of programming (functions, conditionals, arrays), and uses Python.
The free courseware is here:
https://github.com/LoisLab/CSTA_NE_2019
