Sum of Boolean Operators on Numpy Bool Arrays (Bug?)

Question

I've come across a surprising situation when using numpy's arrays. The following code

(True==True)+(True==True)

returns 2, as one would expect. While

import numpy
Array=numpy.zeros((2,2),dtype=bool)
(Array[0][0]==Array[0][0])+(Array[1][0]==Array[1][0])

returns True. This leads to:

(Array[0][0]==Array[0][0])+(Array[1][0]==Array[1][0])-1

returning 0, while

(Array[0][0]==Array[0][0])-1+(Array[1][0]==Array[1][0])

returns 1, making the sum not commutative!

Is this intended? If so, why?

jonrsharpe · Accepted Answer · 2014-12-04 17:06:55Z

4

It would appear that numpy.bool_ behaves slightly differently to vanilla Python bool:

>>> int(True+True) == int(True) + int(True)
True
>>> int(numpy.bool_(1)+numpy.bool_(1)) == int(numpy.bool_(1)) + int(numpy.bool_(1))
False

This is because:

>>> True+True
2
>>> numpy.bool_(1)+numpy.bool_(1)
True
>>> int(numpy.bool_(1)+numpy.bool_(1))
1

Basically, the addition operation for numpy.bool_ is logical, rather than numerical; to get the same behaviour with bool:

>>> int(True and True)
1

This is fine if you only use it for truthiness, as intended, but if you try to use it in an integer context without being explicit about that, you end up surprised. As soon as you're explicit, expected behaviour is restored:

>>> int(numpy.bool_(1)) + int(numpy.bool_(1))
2

edited Dec 4, 2014 at 17:06

answered Dec 4, 2014 at 16:20

jonrsharpe

123k31 gold badges278 silver badges488 bronze badges

Sign up to request clarification or add additional context in comments.

3 Comments

Tilman Over a year ago

Do you know if there is a reason why numpys bool was designed to behave different than vanillas bool? Or wasn't the sum of bools intended in vanillas bool either? Maybe no one had given it a thought ^^

jonrsharpe Over a year ago

I don't, no. There is some discussion about removing these operators here, if you're interested.

Tilman Over a year ago

Thanks for your answer, this motivated me to write cleaner and prettier Python code. Even more so after seeing all the other issues I hadn't noticed (and probably would never notice).

user1532647 · Accepted Answer · 2014-12-04 16:21:16Z

1

I think the problem is th autocasting.

in this case:

(Array[0][0]==Array[0][0])+(Array[1][0]==Array[1][0])-1
Python do:
(Array[0][0]==Array[0][0])+(Array[1][0]==Array[1][0]) = True
True -1 =cast= 1 -1 = 0

In the second case the cast is do it before:

(Array[0][0]==Array[0][0])-1+(Array[1][0]==Array[1][0])
True - 1 + True 
(True - 1 =cast= 0)
0 + True =cast again=  0+ 1 = 1

So, it's not a bug. It's a autocasting in diferent parts.

answered Dec 4, 2014 at 16:21

user1532647

461 silver badge6 bronze badges

Collectives™ on Stack Overflow

Sum of Boolean Operators on Numpy Bool Arrays (Bug?)

2 Answers 2

3 Comments

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

3 Comments

Comments

Your Answer

Sign up or log in

Post as a guest

Related