Why does numpy.dtype(numpy.float64) evaluate to False

Question

Can someone explain the logic behind the output of the following script?

import numpy
if(numpy.dtype(numpy.float64):
    print "Expected"
else:
    print "Surprise!!!!"

Especially considering:

import numpy
if(object):
    print "Expected!"
else:
    print "Surprise"

Thanks :)

possible duplicate of Why is numpy.dtype('float64') special? — Stephen Diehl
– Stephen Diehl, Commented Mar 22, 2014 at 9:24
@StephenDiehl that question is about why dtype('float64') == None; this one is about why bool(dtype('float64')) == False. — lvc
– lvc, Commented Mar 22, 2014 at 9:29
I do not believe it is. "Why is numpy.dtype('float64') special?" is answered, but the answer does not apply in this case. So related: absolutely; but not a duplicate. — Troels Blum
– Troels Blum, Commented Mar 22, 2014 at 9:30

John Zwinck · Accepted Answer · 2014-03-22 09:31:26Z

3

np.dtype does not define __nonzero__, but it does define __len__. As per the documentation, this means when you use it in a boolean context, it will evaluate to True if __len__ returns non-zero. But it always returns zero, regardless of what type you pass in:

>>> bool(np.dtype(int))
False
>>> bool(np.dtype(float))
False
>>> bool(np.dtype(np.int8))
False

On the other hand, a compound data type does return nonzero, thus True:

>>> bool(np.dtype([('foo', int)]))
True

You might then ask why the "length" of a simple dtype is zero, when the length of a compound one with a single element is one. I imagine that's something about dimensionality: an array with a simple dtype and one-dimensional size is itself one-dimensional, but an array with a compound dtype and one-dimensional size may be thought of as two-dimensional, regardless of how many elements are in the compound dtype.

answered Mar 22, 2014 at 9:31

John Zwinck

252k44 gold badges346 silver badges459 bronze badges

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

4 Comments

user2357112 Over a year ago

Strange, especially since 0-dimensional arrays properly raise TypeError for len. I suspect these parts of the dtype API just didn't get much attention.

Troels Blum Over a year ago

Thanks! That explains it. I would consider that a bug in numpy. easily fixed by defining nonzero in numpy.dtype?

lvc Over a year ago

Your supposition about compound dtypes seems to be slightly off: len(np.dtype([('foo', int), ('bar', int)])) is 2. __getitem__ also exists, and behaves as you would expect. @TroelsBlum this isn't necessarily a bug. It seems dtypes follow the sequence protocol, with simple dtypes being an empty sequence - and empty sequences are conventionally False.

John Zwinck Over a year ago

@lvc: I know a two-element compound dtype has len==2. That makes sense. It also has dimension 1, conceptually. My point is that a simple dtype has dimension 0, so a 1D array of floats is 1D, but a 1D array of dtype([('foo', float)]) can be considered 2D.

Collectives™ on Stack Overflow

Why does numpy.dtype(numpy.float64) evaluate to False

1 Answer 1

4 Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

4 Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related