Re: POC for automatic bsxfun

[Michael D Godfrey]

On 08/17/2011 12:12 AM, Jordi Gutiérrez Hermoso wrote:

The attached patch makes bsxfun automatic for most binary operators.
The power operator seems to be the only one that is treated
differently, so it's not covered by this patch.

I'm not pushing this patch since it's potentially quite contentious,
but I enjoy that e.g. it's now quite easy to get the matrix of all
pairwise differences of a vector:

     x = rand(5,1);
     d = x - x';

which I think is a definite improvement over

     d = bsxfun(@minus, x, x');

not to mention less obscure and more likely to be discovered with
experimentation. If this idea sticks, I would like to write additions
to the manual to explain this behaviour, how it differs from Matlab,
when to use it, and when to avoid it.

What do you guys think?

- Jordi G. H.

This capability is very useful.  The only question is: is it "too dangerous."
I vote for doing it, but  full documentation is surely needed.  And, some
diagnostic  modes should be considered.  Or even, if people insist, a way
of if effect disabling it.  At least a "warning-on" mode.

But, do not just decide not to do it.  The general idea of "replication of dimensions"
has been used in other contexts and is very effective.

Michael