[Numeric] why is Float32 incorrect for ufuncs?

[David M. Cooke]

At some point, Curzio Basso <curzio.basso at unibas.ch> wrote:

> Hello everyone,
>
> I am a beginner with Numerical Python, and there is a thing I do not
> understand in the behaviour of the ufuncs. Maybe someone can enlighten
> me.
>
> Why is the following not working?
>
>  >>> import Numeric
>  >>> a=Numeric.ones((2,2),Numeric.Float32)
>  >>> a
> array([[ 1.,  1.],
>         [ 1.,  1.]],'f')
>  >>> a/=2
> Traceback (most recent call last):
>    File "<stdin>", line 1, in ?
> TypeError: return array has incorrect type
>
> If I use Numeric.Float64 as typecode everything works fine, but I do
> not understand why is this needed.
>
> thanks for your help. curzio.

Because a/2 returns an array with typecode of Numeric.Float64, which
can not be copied into an array of typecode Numeric.Float32. That's
becasue Numeric does type upcasting. Note that a Python float is
equivalent to Numeric.Float64.

This is a bit of a wart in Numeric. You can work around it by calling
ones with savespace=1, or calling a.savespace(1). Or, use numarray,
where the default behaviour is not to do upcasting:

>>> import numarray
>>> a = numarray.ones( (2,2), type=numarray.Float32)
>>> print a
array([[ 1.,  1.],
       [ 1.,  1.]], type=Float32)
>>> a /= 2
>>> print a
array([[ 0.5,  0.5],
       [ 0.5,  0.5]], type=Float32)

-- 
|>|\/|<
/--------------------------------------------------------------------------\
|David M. Cooke
|cookedm(at)physics(dot)mcmaster(dot)ca