[Numpy-discussion] matrix indexing question

Alan Isaac aisaac@american....
Tue Mar 27 00:05:06 CDT 2007


On Mon, 26 Mar 2007, Travis Oliphant wrote: 
> It actually has been offered.   You just don't accept it.  
> Matrices are containers of matrices.  
> If M is an (mxn) matrix then M[0] is a (1xn) matrix.  
> Viewing this 1xn matrix as a 1-d array loses it's row-vectorness.  
> This seems perfectly logical and acceptable to me.  I'm waiting for a 
> better explanation as to why this is not acceptable.  
> Arguments that rest on what is and what isn't "Pythonic" are not very 
> persuasive as this is very often in the eye of the beholder. 

Again, I only raised a *question* about whether
there might be a design problem here.  My goal
was only to have this explored, and I've tried
to explain why.

The evidence I offer:
- it is surprising (at least to some) that iterating
  over a matrix yields matrices
- I believe it is unexpected (prior to instruction) and that 
  there is a natural more expected behavior
- if that is right, deviation from the expected should have 
  a good justification
- this behavior has tripped up at least a couple people and
  I expect that to happen to many others over time (because 
  I believe the behavior is unexpected)
- when I desire to iterate over a matrix I always want the 
  arrays. (Of course there is a way to have them; that's
  not the point).  I'm interested to see a use case where
  the rows are desired as matrices

As you indicate, none of this constitutes an "argument".
And since nobody appears to agree, I should shut up.
This will be my last post on this subject.

Cheers,
Alan Isaac






More information about the Numpy-discussion mailing list