break ties in Gambler's Problem

[hansweytjens]

May I propose following line to substitute for line 54:

policy[state] = actions[np.argmax(np.round(action_returns[1:],5))]

The [1:] avoid choosing the '0' action which doesn't change state nor exptected returns. Since numpy.argmax chooses the first option in case of ties, rounding the near-ties assures the one associated with the smallest action (or bet) is selected. The output of the app now resembles Figure 4.3. in the Sutton/Bartho's book.

PS just signed up for Github, may not be using it optimally yet

[ShangtongZhang]

This looks awesome, I just made a commit. Thanks!