Ignore divide by 0 warning in NumPy

Question

I have a function for statistic issues:

import numpy as np
from scipy.special import gamma as Gamma

def Foo(xdata):
    ...
    return x1 * (
                 ( #R is a numpy vector
                  ( ((R - x2)/beta) ** (x3 -1) ) * 
                  ( np.exp( - ((R - x2) / x4) ) ) /
                  ( x4 * Gamma(x3))
                 ).real
                )

Sometimes I get from the shell the following warning:

RuntimeWarning: divide by zero encountered in...

I use the numpy isinf function to correct the results of the function in other files, so I do not need this warning.

Is there a way to ignore the message? In other words, I do not want the shell to print this message.

I do not want to disable all python warnings, just this one.

You could just catch it and ignore it if you don't need to recover from it. — Carcigenicate, Apr 29 '15 at 17:30
you can always use `try... except ZeroDivisionError` (or whatever error is being thrown) — letsc, Apr 29 '15 at 17:30
possible duplicate of [How to disable python warnings](http://stackoverflow.com/questions/14463277/how-to-disable-python-warnings) — marsh, Apr 29 '15 at 17:34
@letsc no error is being thrown. the OP is getting a warning. — abcd, Apr 29 '15 at 17:35
marsh I don't want to disable all the warning. Just this if it's possible. — overcomer, Apr 29 '15 at 17:38

score 188 · Accepted Answer · edited May 25 '23 at 10:42

You can disable the warning with numpy.seterr. Put this before the possible division by zero:

np.seterr(divide='ignore')

That'll disable zero division warnings globally. If you just want to disable them for a little bit, you can use numpy.errstate in a with clause:

with np.errstate(divide='ignore'):
    # some code here

For a zero by zero division (undetermined, results in a NaN), the error behaviour has changed with numpy version 1.8.0: this is now considered "invalid", while previously it was "divide".

Thus, if there is a chance you your numerator could be zero as well, use

np.seterr(divide='ignore', invalid='ignore')

or

with np.errstate(divide='ignore', invalid='ignore'):
    # some code here

See the "Compatibility" section in the release notes, last paragraph before the "New Features" section:

Comparing NaN floating point numbers now raises the invalid runtime warning. If a NaN is expected the warning can be ignored using np.errstate.

4

I love the last solution – Miladiouss Apr 14 '20 at 23:22

score -3 · Answer 2 · answered Apr 29 '15 at 18:22

-3

You could also use numpy.divide for division. That way you don't have to explicitly disable warnings.

In [725]: np.divide(2, 0)
Out[725]: 0

answered Apr 29 '15 at 18:22

fixxxer

15,568
15
58
76

6

This is actually not true. At least, for NumPy version 1.15.4 – Jalo Jan 16 '19 at 11:59
3

No longer true. This now returns inf – yahya Apr 29 '21 at 00:56

Ignore divide by 0 warning in NumPy

2 Answers2

Linked

Related