Part $1$:

Write code for a multi$-$arm bandit algorithm that has the following characteristics:

A: number of arms

P: Distribution of rewards $[0,1].$ Use the beta distribution so you can tune the rewards distribution based on two parameters. Choose your own parameter settings and graph the distributions in one plot.

r$\_$i: reward $(0$ or $1)$ taken from probability distribution P$\_$i

T: number of rounds played $($gambles$)$

Part $2$:

Suppose you have $4$ arms $($A$=4).$ Implement a random, a greedy, an epsilon$-$first greedy, and epsilon greedy, and a upper confidence band $($UCB$1)$ approach to selecting the best arm to play. Ensure the strategies only use the rewards when determining

Part $3$:

Evaluate the performance of the $5$ strategies by $1)$ plotting the regret of each round $[$i$.$e$.,$ plot Regret$($round#$)$ versus round$]$ and $2)$ plotting the expected regret averaged over $50$ rounds $[$i$.$e$.,$ plot average Regret$($round#$)$ versus round$].$ Regret is the difference between actual reward and reward if you played optimally.