[Numpy-discussion] matrix indexing question

Colin J. Williams cjw@sympatico...
Tue Mar 27 13:39:22 CDT 2007


Alan G Isaac wrote:
> On Mon, 26 Mar 2007, "Colin J. Williams" apparently wrote: 
>> One would expect the iteration over A to return row 
>> vectors, represented by (1, n) matrices. 
> 
> This is again simple assertion.
> **Why** would "one" expect this?
> Some people clearly do not.
> 
> One person commented that this unexpected behavior was 
> a source of error in their code.
> 
> Another person commented that they did not even guess that 
> such a thing would be possible.
> 
> Experience with Python should lead to the ability to 
> anticipate the outcome.  Apparently this is not the case.
> That suggests a design problem.
> 
> What about **Python** would lead us to expect this behavior??
> 
> In *contrast*, everyone agrees that for a matrix M,
> we should get a matrix from M[0,:].
> This is expected and desirable.

Perhaps our differences lies in two things:

1. the fact that the text books typically take the column vector as the 
default.  For a Python version, based on C it makes more sense to treat 
the rows as vectors, as data is stored contiguously by row.

2. the convention has been proposed that the vector is more conveniently 
implemented as a matrix, where one dimension is one. The vector could be 
treated as a subclass of the matrix but this adds complexity with little 
clear benefit.  PyMatrix has matrix methods isVector, isCVector and 
isRVector.

I can see some merit in conforming to text book usage and would be glad 
to consider changes when I complete the port to numpy, in a few months.

Colin W.
> 
> Cheers,
> Alan Isaac



More information about the Numpy-discussion mailing list