[SciPy-dev] numpy.random.randint

Robert Kern robert.kern at gmail.com
Fri Mar 10 12:45:26 CST 2006

Alan G Isaac wrote:
>>Johannes Löhnert wrote: 
>>>I just noticed that in function numpy.random.randint, the 
>>>shape=... keyword was replaced by size=... 
>>>I find this somewhat counterintuitive, as it denotes 
>>>the shape of the desired array (as tuple), 
>>>although it also takes integer values. Also it is an 
>>>unnecessary break with backwards compatibility (to RandomArray). 
>>>I would suggest to allow both "shape" & "size", with 
>>>"shape" having precedence. 
> On Fri, 10 Mar 2006, Robert Kern apparently wrote: 
>>It was a choice for compatibility with the scipy PRNGs all of which use "size". 
>>"shape" was not a good argument name because many of the probability 
>>distributions have a "shape" parameter. 
>>I don't think it's wise to make more ways to do exactly 
>>the same thing. 
> But this is really horrible:
>     >>> import numpy as N
>     >>> x = N.random.standard_normal(size=(3,2))
>     >>> x.size
>     6
>     >>> x.shape
>     (3, 2)
>     >>> x = N.random.standard_normal(shape=(3,2))
>     Traceback (most recent call last):
>       File "<stdin>", line 1, in ?
>     TypeError: 'shape' is an invalid keyword argument for this function
> Since numpy is less than version 1, this really should be 
> changed for consistency, and allowing both for a while with 
> a deprecation warning on 'size' is reasonable.

The keyword in the functions does not necessarily match with the attribute names
on the objects. I don't find this particularly shocking. Or bad. But that's just
my tastes.

So I'm -1 on changing numpy.random and all of scipy.stats.

> I do not think any of the numpy distributions have a competing 'shape'
> parameter, but if so that just makes the situation *worse*.

I also avoided the "shape" terminology for the "shape" parameters, too, to avoid
conflation. However, a lot of the probability distributions do have a "shape"
parameter. lognormal, gamma, beta, pareto, and on and on and on. Anything that
isn't a location or a scale parameter is called a shape parameter in probability

Robert Kern
robert.kern at gmail.com

"I have come to believe that the whole world is an enigma, a harmless enigma
 that is made terrible by our own mad attempt to interpret it as though it had
 an underlying truth."
  -- Umberto Eco

More information about the Scipy-dev mailing list