CODEWITHSUNDEEP
pythonnumpy
ayeayeron
ayeayeron

Reputation: 63

odd numpy behaviour, what's happening here?

I was having trouble using numpy.average, it hit an error every time until I converted the inputs to numpy.float64s. I had a look at the source and it's because of the following behaviour, what's the reason for the difference?

>>> f, f32 = numpy.float(1.0), numpy.float32(1.0)

>>> (f == 1.0).any()

Traceback (most recent call last):
  File "<pyshell#80>", line 1, in <module>
    (f == 1.0).any()
AttributeError: 'bool' object has no attribute 'any'
>>> (f32 == 1.0).any()
True

Upvotes: 1

Views: 158

Answers (3)

Dleep
Dleep

Reputation: 1065

It's because a numpy.float32 object doesn't return a bool object upon comparison with __eq__, but a numpy.bool_ one.

>>> type((f == 1.0))
<type 'bool'>
>>> type((f32 == 1.0))
<type 'numpy.bool_'>

Upvotes: 1

hpaulj
hpaulj

Reputation: 231325

You don't normally need to use expressions like numpy.float(1.0).

The most common, basic operation in numpy is to make an array.

f = np.array([1.0, 1.23])
mf = f.mean()  # or
mf = np.mean(f)

or with integers

f = np.array([1, 3, 5, 3, 5])
f==3
(f==3).any()

Note I am using integers with == test. == test is not a good idea when using floats.

I could go further and specify the dtype:

 f = np.arange(5, dtype=int)
 f = np.ones((3,4), dtype=float)

Individual items of these arrays will be of type np.float, np.int, etc. There's rarely a need to create one directly np.float(1.23). Specify the dtype when creating an array, but otherwise let numpy take care of those details.

Upvotes: 0

eph
eph

Reputation: 2028

>>> type(numpy.float32(1.0))
<type 'numpy.float32'>
>>> type(numpy.float(1.0))
<type 'float'>
>>> type(numpy.float32(1.0) == 1.0)
<type 'numpy.bool_'>
>>> type(numpy.float(1.0) == 1.0)
<type 'bool'>

The numpy.float32 and numpy.bool_ are used to not only store scalar but also vector:

>>> numpy.float32([1.0, 2.0])
array([ 1.,  2.], dtype=float32)
>>> numpy.float32([1.0, 2.0]) == 1.0
array([ True, False], dtype=bool)

So there is an any() function in numpy.bool_ to check whether any of the items is True.

Upvotes: 2

Related Questions