An epsilon$-$greedy strategy for the stochastic multi$-$armed bandits set up exploits the current

best arm with probability $(1)$ and explores with a small probability $.$ Consider a

problem instance with $10$ arms where the reward for the i$-$th $($i $=1,...,10)$ arm is Beta

distributed with parameters $\backslash$alpha i $=5,\backslash$beta i $=5$ i$.$ Implement the epsilon$-$greedy algorithm

and compare it with the performance of the UCB and the EXP$-3$ algorithm. Plot the regret

bounds and comment on your observations. $($Bonus: Can you formally show a regret

guarantee for the epsilon$-$greedy algorithm?$)$