JOSHUA HOLBROOK flickr : github : twitter

The Importance of Cheap Containers (posted 13 Mar 2011)

I’ve been hammering out my thesis for the last few weeks, and while writing up the numerical studies I did, I came across this gem of code:

%tabulator
%turns lame structures into some csv action

%given params:
%  answers
%  angles
%  kxy
%  kz

% Bad style, but I'm dealing with a cluttered namespace because I'm not functionalizing these.
% This is because parameter passing is a pita. So, leave me alone.
[Kxy, Kz] = meshgrid(kxy, kz);

fprintf('angle, kxy, kz, kmeas \n');
for t=1:length(angles),
    for pt=1:size(Kxy,1)*size(Kxy,2),
	fprintf([ num2str(angles(t)) ', ' num2str(Kxy(pt)) ', ' num2str(Kz(pt)) ', ' num2str(answers{t}{pt}{1}) '\n']);
    end
end

The interesting part here is the comment about bad style.

See, there are basically two ways to handle a procedure in MATLAB. The first is to write a function that returns something:

% ftoc.m
function C = ftoc(F)
    C = (5/9)*F-32;
end

which must be used like this:

> ftoc(212)
100

Or, you can write a procedure to manipulate the existing global namespace in your MATLAB session:

% ctof.m

F = (C + 32)*(9/5);

which must be used like this:

> C = 100

    C = 100

> ctof
> F

    F = 212

>

I think this latter use should be considered Terrible Style in MATLAB. Hell, it should be considered less-than-desirable everywhere. And yet, people do this sort of thing all the time in MATLAB.

Why? Well, I suspect that it has to do with MATLAB’s data structures and syntax. For example, in that particular procedure I wrote, I needed k_xy, k_z, angles and answers, the latter of which was a nested cell array. In fact, the very reason for this procedure’s existence had to do with the inherent awkwardness that came with trying to do things with answers directly.

Now, I don’t want to say that this would make everything better, but suppose that, instead of that mess, we could’ve written something more like:

tabulator({ answers: answers,
            kxy: k_xy,
            kz: k_z,
            angles: angles });

Or, even better, what if I simply bundled everything together once, and used the bundle for all my function calls?

> analyzer(bundle);
    *results*
> tabulator(bundle);
    *results*

and etcetera? Now, MATLAB does have structs. However, in my experience they’re a bit annoying to work with. They’re certainly not as easy to use as javascript’s objects or python’s dicts.

I would like to put forth the preposition that, in order to keep clean code for scientific computing uses, especially in the REPL as is done with MATLAB, that it is critical to have some sort of quick, easy, cheap bundling tool, like javascript’s object or python’s dictionary.

I’d expound on this some more, but I’d better get back to work. My graduation is at stake!

Friends:

ALL THE BLOGPOSTS STANDING IN THE LINE FOR THE BATHROOM: