[Numpy-discussion] Detect subclass of ndarray

[Charles R Harris]

On 3/24/07, Travis Oliphant <oliphant at ee.byu.edu> wrote:
>
> Alan G Isaac wrote:
> > On Sat, 24 Mar 2007, Travis Oliphant apparently wrote:
> >
> >> My opinion is that a 1-d array in matrix-multiplication
> >> should always be interpreted as a row vector.  Is this not
> >> what is currently done?   If not, then it is a bug in my
> >> mind.
> >>
> >
> >
> >>>> N.__version__
> >>>>
> > '1.0'
> >
> >>>> I
> >>>>
> > matrix([[ 1.,  0.],
> >         [ 0.,  1.]])
> >
> >>>> I*N.ones(2)
> >>>>
> > matrix([[ 1.,  1.]])
> >
> > If N.ones(2) were treated as a row vector,
> > matrix multiplication is not possible.
> > So the question is what should happen.
> > I would like an exception raised.
> > The current behavior is lmost certainly not desirable,
> > although it has the virute of matching ``dot``.
> >
> >
> I think that's why it is the current behavior --- by default rather than
> by design.    But, dot has different use cases.
>
> I'd be fine with an error raised on matrix multiplication (as long as
> dot is not changed).   In other words, I'd like to see 1-d arrays always
> interpreted the same way (as row vectors) when used in matrix
> multiplication.


The relevant bit of code is

    def __mul__(self, other):
        if isinstance(other, N.ndarray) or N.isscalar(other) or \
               not hasattr(other, '__rmul__'):
            return N.dot(self, other)
        else:
            return NotImplemented

How about I just replace

N.dot(self, other)

by

N.dot(self, asmatrix(other))

or maybe something like

try :
    rhs = asmatrix(other)
    val = dot(self, rhs)
except :
    val = NotImplemented
return val

It is not clear to me why the check for __rmul__ is included in the original
code, as I believe it causes problems when the rhs is a list, i.e.

mat(eye(2))*[1,1]

currently fails.

Chuck
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/numpy-discussion/attachments/20070324/48f3ff93/attachment.html>