dougct / predictability Goto Github PK

When using your code to find max_predictability, I was getting a lot of results of 0, so I scrutinized your code.

In pred_lims.max_predictability(), you traverse p from 1e-4 to 1 with steplength of 1e-4. At lines 34 and 35, your pi_max is h+func(p)-S, and then you're trying to compare pi_max to 0.001. I can understand that you are trying to compare S = - H(p) + (1 - p)log(N - 1) on both sides of the equation, but a major problem here is that pi_max can be negative in some cases, (e.g., p=0.0001, S=1.44, N=3), so an abs() needs to be added here, i.e. if abs(pi max) <= 0.001:

In addition, such a manner to calculate max_predictability is time-consuming, maybe try scipy.optimize.fsolve.

:)