On Sat, Oct 6, 2012 at 12:17 PM, Ralf Gommers <ralf.gommers@gmail.com>wrote:

On Fri, Oct 5, 2012 at 5:17 PM, Dan Goodman <dg.gmane@thesamovar.net>wrote:

...

Hi,

numpy.set_printoptions(precision=...) doesn't affect single floats, even if they are numpy floats rather than Python floats. Is this a bug or is there some reason for this behaviour? I ask because I have a class that derives from numpy.float64 and adds some extra information, and I'd like to be able to control the precision. I could fix it to use the precision set by numpy.set_printoptions, but then it would be inconsistent with how numpy itself handles precision. Thoughts?

Do you mean scalars or arrays? For me set_printoptions only affects arrays and not scalars. Both float32 and float64 arrays work as advertised:

In [28]: np.set_printoptions(precision=4)

In [29]: np.array([np.float32(1.234567891011011101111012345679)]) Out[29]: array([ 1.2346], dtype=float32)

In [30]: np.array([np.float64(1.234567891011011101111012345679)]) Out[30]: array([ 1.2346])

In [31]: np.set_printoptions(precision=8)

In [32]: np.array([np.float32(1.234567891011011101111012345679)]) Out[32]: array([ 1.23456788], dtype=float32)

In [33]: np.array([np.float64(1.234567891011011101111012345679)]) Out[33]: array([ 1.23456789])

But for scalars it doesn't work:

In [34]: np.float32(1.234567891011011101111012345679) Out[34]: 1.2345679

In [35]: np.float64(1.234567891011011101111012345679) Out[35]: 1.2345678910110112

In [36]: np.set_printoptions(precision=4)

In [37]: np.float32(1.234567891011011101111012345679) Out[37]: 1.2345679

In [38]: np.float64(1.234567891011011101111012345679) Out[38]: 1.2345678910110112

Ralf

It also does not affect zero-dimensional (i.e. scalar) arrays (e.g. array(1.2345)): In [1]: x = array(1./3) In [2]: x Out[2]: array(0.3333333333333333) In [3]: set_printoptions(precision=3) In [4]: x Out[4]: array(0.3333333333333333) In [5]: type(x) Out[5]: numpy.ndarray `y` is a 1-d array, so this works as expected: In [6]: y = array([1./3]) In [7]: y Out[7]: array([ 0.333]) Warren

_______________________________________________ NumPy-Discussion mailing list NumPy-Discussion@scipy.org http://mail.scipy.org/mailman/listinfo/numpy-discussion

On Sun, Oct 7, 2012 at 6:36 PM, Dan Goodman <dg.gmane@thesamovar.net> wrote:

On 06/10/2012 18:17, Ralf Gommers wrote:

...

Do you mean scalars or arrays? For me set_printoptions only affects arrays and not scalars. Both float32 and float64 arrays work as advertised:

Yep, I mean scalars (although as Warren noted, it also affects e.g. array(1.23456789)).

My impression from previous discussions on the mailing list is that this is by design: printing of scalars and scalar arrays is handed off to python which doesn't have a set_printoptions. printing of arrays is under the control of numpy I don't see a way to influence the __str__ or __repr__ of a python float. Josef

Dan

_______________________________________________ NumPy-Discussion mailing list NumPy-Discussion@scipy.org http://mail.scipy.org/mailman/listinfo/numpy-discussion