Python: Making numpy default to float32

Question

Is there any clean way of setting numpy to use float32 values instead of float64 globally?

Also an interesting discussion on scipy list: old.nabble.com/switching-to-float32-ts24203533.html#a24203533 — joris
– joris, Commented Apr 19, 2011 at 20:41
Saw the nabble conversation but missed the float128 (thanks sven), and it appears they're both saying the same thing, 'No, not really'. On the Nabble discussion there was then mention of adding it to the cookbook which would be nice. I myself just did a few hacks similar to the float128 question sven mentioned (and answered). Pity — Bolster
– Bolster, Commented Apr 20, 2011 at 0:03

talonmies · Accepted Answer · 2011-04-20 12:05:38Z

13

Not that I am aware of. You either need to specify the dtype explicitly when you call the constructor for any array, or cast an array to float32 (use the ndarray.astype method) before passing it to your GPU code (I take it this is what the question pertains to?). If it is the GPU case you are really worried about, I favor the latter - it can become very annoying to try and keep everything in single precision without an extremely thorough understanding of the numpy broadcasting rules and very carefully designed code.

Another alternative might be to create your own methods which overload the standard numpy constructors (so numpy.zeros, numpy.ones, numpy.empty). That should go pretty close to keeping everything in float32.

edited Apr 20, 2011 at 12:05

answered Apr 20, 2011 at 12:00

talonmies

72.8k35 gold badges204 silver badges296 bronze badges

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

2 Comments

Bolster Over a year ago

That's what I'm doing for the GPU case, but I was concerned about direct comparison between CPU bound equivalent calculation precision for testing, as part of a 'weird-bug-hunt'

Chiel Over a year ago

Is this comment still valid concerning the broadcasting rules? If arrays are declared with dtype float32, is there then any risk in implicit conversions or broadcasting?

Mihai Capotă · Accepted Answer · 2018-06-22 22:26:09Z

7

This question showed up on the NumPy issue tracker. The answer is:

There isn't, sorry. And I'm afraid we're unlikely to add such a thing[.]

answered Jun 22, 2018 at 22:26

Mihai Capotă

2,4902 gold badges34 silver badges26 bronze badges

Comments

0-_-0 · Accepted Answer · 2018-11-01 18:22:11Z

3

For each function you can overload by:

def array(*args, **kwargs):
    kwargs.setdefault("dtype", np.float32)
    return np.array(*args, **kwargs)

As posted by njsmith on github

answered Nov 1, 2018 at 18:22

0-_-0

1,56319 silver badges18 bronze badges

1 Comment

Tom McLean Over a year ago

dtype can be a positional argument as well, so this can break sometimes

Collectives™ on Stack Overflow

Python: Making numpy default to float32

3 Answers 3

2 Comments

Comments

1 Comment

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

3 Answers 3

2 Comments

Comments

1 Comment

Your Answer

Sign up or log in

Post as a guest

Linked

Related