In MATLAB we are used to the difference between * and .*, why not something like that for bsx? I can’t think of what would be a good symbol to use… Maybe the question mark? Lexically, it might suggest “hopefully my dims will expand”…

a +? b

Of course it won’t work in all cases, but at least it makes the simple arithmetic (and maybe logical?) ones more readable.

I further tried

    x = 1:10;
    y = 1:20;
    z = bsxfun(@mytimes, x, y');

This time, there is no error message and I got to my breakpoint in the mytimes function.

Thank you,
Ben

I also tried using a self-defined function called mytimes.

Q = bsxfun(@mtimes, V, MM);

where the mytimes function is

Hi Loren,

I also tried using a self-defined function called mytimes.

function out = mytimes(v,m)

out = v * m;

I set breakpoint in mytimes function, but again got error message without call mytimes function.

Could you provide a example of using cells?

Thank you very much!
Ben

bsxfun does support cells, but the operator mtimes does not. The function being called by bsxfun would have to know how to work with cell arrays.

–Loren

I am trying to run the following code where V and MM are two cells.

Q = bsxfun(@mtimes, V, MM);

In Matlab document, it is said “The inputs must be of the following types: numeric, logical, char, struct, cell.” But I got error message:

Error using bsxfun
Operands must be numeric arrays.

Does bsxfun not support cell or I made something wrong? Thanks a lot for your help in advance.

Best,
Ben

I see what you mean, with bad code producing answers instead of error messages. Adding a long Nx1 array with a 1xN array would give a huge NxN matrix, instead of a simple “what are you doing?” message. Changing something as basic as this should definitely not be taken lightly! But at least it wouldn’t change the meaning of existing code.

I was pointed at a language called Yorick some days ago. It has some very crazy, but interesting, ideas on indexing: http://yorick.sourceforge.net/manual/yorick_37.php#SEC37
Note especially the “*” and “..” rubber indexing. Again something I would have used extensively if it had been available in MATLAB! :)

Best,
Cris.

I want to echo Steve’s comments. I am all for it as well. Just need to find the right path through.

–Loren

We are worried about compatibility and making debugging very difficult for users since more expressions will get answers when users didn’t mean to write what they wrote.

–Loren

I was thinking, what is the reason for implementing the |bsxfun| function, and not making the singleton expansion the default behavior for all operators? MATLAB has always done singleton expansion in the case of a scalar value (i.e. all dimensions are singleton): |A+4| is the same as |A+repmat(4,size(A))|. Is there any conceivable problem that could arise if MATLAB were to always expand any singleton dimensions? I would love to be able to do |A-mean(A)|! I’ve been thinking about this for a while now, and I can’t come up with a scenario where this would backfire.

Cheers,
Cris.