roycoding / slots Goto Github PK

slots.py:377: RuntimeWarning: invalid value encountered in true_divide sum(self.pulls) * np.max(np.nan_to_num(self.wins / self.pulls))

Options:

• Refactor type hints for 3.5 compatibility
• Remove 3.5 support

I want to be sure that the current implementation of the Bayesian bandits works if the payouts are arbitrary sizes, rather than just the default 0 or 1.

We need tests!

• I am planning to use pytest.
• Asymptotic tests would be nice for the MAB strategies.

Refactor to historical and future payouts for both fixed payout values (both binary and numerical) and variable payouts.

• For live trials, handle binary or numerical payouts.
• For off-line trials, handle fixed binary or numerical payouts and variable numerical payouts.

This probably requires a relatively significant API change.

Running

mab = slots.MAB()
mab.run()

results in an error.

This seems to be due to how the defaults are handled for a "live" trial.

live=False might be a better default?

Remove explicit support for Python 2.7 and 3.4