Negative PMF

[jiboncom]

Hello,
I am trying to use PoiBin for a Maximun Likelihood estimation, but when the PMF is extremely small (e.g. e-15) it sometimes returns negative numbers. These generate an exception in most solvers

[tsakim]

Hi @jiboncom,

sorry for the late reply. For such small numbers it may indeed be a precision issue. Would you have an example where you have encountered this problem?
Numbers around e-15 could be close to the machine limit of the floating numbers. You can check the smallest representable positive number using numpy.finfo like so:

>>> import numpy as np
>>> np.finfo(float).eps
2.220446049250313e-16

Could you provide an example?

Best,
Mika

[PercyLau]

Hi, Mika

Could you please try this code?

import numpy as np from poibin import PoiBin from scipy.stats import binom a = np.ones(100) b = PoiBin(a) print(b.pmf(0)) print(b.cdf(10))

On my desktop ( i7-4700k, Ubuntu 16.04.05), the terminal prints
-4.98906471691e-15 -8.93280946174e-16

Best,
Percy