Instead of using the $\backslash$epsi $-$greedy$$ algorithm, we will now do some interesting exploration with softmax. We first introduce a new type of policy: A stochastic policy $\backslash$pi $($a$|$s$)$ represents the probability of action a being prescribed, conditioned on the current state. In other words, the policy is a now a distribution over possible actions, rather than a function that outputs a deterministic action.