Details: A game theory assignment to be completed in python.

hw4 (1).pdf1D

There were clarifications sent out to the class:
Attached are some example plots to show you what I am talking about for some of the plots I have requested.

You may also consider generating a table. with different algorithms as columns and rows, and then statistics on how they perform against one another in the entries (e.g., do they converge? are they converging to Nash or a cooperative solution? what is the learned policy? how does convergence depend on initial conditions if relevant?)

The whole goal is to get you to explore the different possible learning algorithms and understand how they fair against themselves and one another. So this assignment is a very simplified environment in which you can hopefully explore how different learning algorithms long run behavior is dependent on the decision rule and information it has available.
For example you may find that two UCB players go to cooperative solutions in PD and a UCB vs e-greedy player go to a strategy that is better for one player than another. Essentially what can happen is that players beliefs may end up getting skewed by the samples they are seeing from their competitor.

And attached is the paper that sort of motivated this assignment.

Note: Just do multiple tournaments for the case where you add noise to the cost matrices or change the intiialization/parameters—e.g., epsilon in e-greedy.