Better way to calculate difference between numpy unsigned ints in Python

Question

I am calculating the absolute difference between two numpy.uint8s in python with d = abs(a - b), which results in an overflow error if b is bigger than a. What's the best way to avoid this?

Er... no it doesn't. Python automatically promotes integers to longs when necessary. Also, there's no unsigned integer type in Python. Please post code that demonstrates the problem you're actually having, and a traceback. — kindall, Sep 26 '12 at 23:55
I get a and b from an external library (pygtk) and their type is numpy.uint8. The error is RuntimeWarning: overflow encountered in ubyte_scalars. Because of this it's not possible to give a short working example. — blues, Sep 26 '12 at 23:58

score 3 · Accepted Answer · edited May 23 '17 at 11:45

3

As your comment indicates, they aren't ints; they're numpy.uint8s. Just convert them to ints:

>>> a, b = map(numpy.uint8, (50, 60))
>>> a - b
__main__:1: RuntimeWarning: overflow encountered in ubyte_scalars
246
>>> a, b = map(int, (a, b))
>>> a - b
-10

Since you are concerned about speed, here are a couple of tests (borrowing Sven's form, with thanks):

>>> %timeit abs(int(a) - int(b))
1000000 loops, best of 3: 410 ns per loop
>>> %timeit a - b if a > b else b - a
1000000 loops, best of 3: 470 ns per loop

So yes, it's faster, but unless we're talking about doing it hundreds of millions of times, it won't matter a bit.

edited May 23 '17 at 11:45

Community

1
1

answered Sep 27 '12 at 00:02

senderle

145,869
36
209
233

Is this faster than the test Ionut suggested? Because I need to do this very often. – blues Sep 27 '12 at 00:07
@blues, if you have a numpy array of them (which looks a lot like a list with duck typing) you can get *much* faster probably, but if its just single values numpy can't help you with speed... – seberg Sep 27 '12 at 17:23

score 2 · Answer 2 · answered Sep 27 '12 at 00:05

2

The easiest way is to manually convert your numbers to Python ints first:

d = abs(int(a) - int(b))

Python ints can't overflow (unless the memory is full).

answered Sep 27 '12 at 00:05

Sven Marnach

574,206
118
941
841

score 0 · Answer 3 · answered Mar 24 '15 at 16:52

For me it does not throw overflow error, just results in false values if b is bigger than a. To stay in uint8 boundaries use this function:

def abs_dev (a, b) :
    sub1 = a - b
    sub2 = b - a
    mask = a < b
    sub1[mask] = sub2[mask]
    return sub1

Better way to calculate difference between numpy unsigned ints in Python

3 Answers3