Re: Ignorance

[Robert T. Short]

On 03/06/2012 09:52 AM, Jordi Gutiérrez Hermoso wrote:
> On 6 March 2012 12:46, Robert T. Short<address@hidden>  wrote:
> > On 03/05/2012 07:34 PM, John W. Eaton wrote:
> > > On  5-Mar-2012, Robert T. Short wrote:
> > >
> > > | I don't know the dimensions or the direction a-priori.  The obvious
> > > | answer is 1:N followed by a repmat but there are so many cute little
> > > | octave tricks that I was hoping there was something really clever.
> > >
> > > Given
> > >
> > >    x = 1:4;
> > >    y = ones (3, 1);
> > >    z = ones (3, 4);
> > >
> > > here are the ways I can think of:
> > >
> > >    Repmat:
> > >
> > >      repmat (x, [3,1])
> > >
> > >    Kron:
> > >
> > >      kron (x, y)
> > >
> > >    Indexing:
> > >
> > >      x(y,:)
> > >
> > >    Outer product:
> > >
> > >      y * x
> > >
> > >    Broadcasting:
> > >
> > >      z .* x
> > >
> > > Are there any others?
> > >
> > > jwe
> > Excellent.  The kind of thing I was looking for was the indexing solution.
> >   Using kron, outer product, sums, broadcasting involve computation so I
> > would rather not use them for something like this.  Thanks for the tips.
> tic-tock them... You may be surprised at which one is fastest.
>
> Also, I'm expecting that this is an intermediate step for a latter
> computation, which is why I suggested broadcasting, which does the
> latter computation without explicitly doing something as useless as
> copying data around.
>
> - Jordi G. H.
I will do some performance comparisons for my own edification - I based 
my statements on some comments in the manual about indexing vs. things 
like a*ones(x,y).  The indexing thing worked out slick as a whistle 
though and really was the right solution to my problem.

I did understand what you meant about using broadcasting.  However, I am 
working on a little problem that does not lend itself to array 
computations, similar to (and based on) the Bessel function thing I have 
been so wound up about.  Basically my problem really requires that I 
compute each element of an array individually, so building 
like-dimensioned arrays of arguments seems a pretty sensible way to go.  
I will submit the results on the forge list when I get done and you can 
pick me apart then. :-[

I really do like the broadcasting thing.  The problem is that octave is 
moving way ahead of MATLAB and it is becoming increasingly more 
difficult to provide compatibility unless I dumb down my scripts.  Mind 
you, that is a good problem to have.

Bob