Thanks. I’ve added your suggestion to the enhancement list for consideration.

–Loren

2) Well, if it were documented, I suppose some would find it. I know I looked there before posting here. But I am sure you have usability experts who can answer that question for you.

Congratulations on your new position! Certainly mean and var could be combined? I don’t think it would do away with the need for mean, but could be more efficient IF people realized how to best use the new variance to get both. The question I have for all of these thoughts is

1) is that really a bottleneck in the code
2) will uses figure it out and find what they need

Thanks for the suggestion.
–Loren

In my work I regularly calculate the mean and variance of several hundreds of matrices, each with over a milion elements. Now, this usually runs faster than one might suspect and limited amouts of RAM are generally the big problem.

However, when calculating the variance, one gets the mean as a by-product. Would it be possible for var (or a variant of it) to output the mean as well as the variance? Would it be useful? It would definately be economical.

Sincerely
Daniel Armyr
Newly appointed “Head Matlab Expert” at his department.

A more practical example would be inlining calls to functions coded in MATLAB - making it as efficient (say) to call a function within a loop as to execute the same program with the function hand coded inside the loop.

ax(1) = subplot(2,1,1);
plot(x1,y1);
ax(2) = subplot(2,1,2);
plot(x2, y2);
set(ax, ‘ylim’, minmax([y1 y2]));

It’s perhaps a trivial use, but I use it pretty often.

Much food for thought here and I thank you for it. I will mention just a couple of points.

I was trying to see how common the combined use of min and max are. I still don’t know if I have a feel for that. I was also trying to understand how big a portion of your overall algorithms were dominated by the calls to min and max vs. spending time doing lots of other stuff as well. My guess is that min and max, while perhaps taking some time, would not be dominant in most of your application profile reports. But again, I don’t know.

We do often have to decide in a design what to do for large for small data sizes because as Tim mentioned, algorithms can be dominated by computer architecture, etc.

As for readability, I am not sure that minAndMax, or whatever name might be used, we be much less readible than calling min and then max. Sometimes the function names do help out. Still I understand the desire for readability and applaud it totally.

Thanks again for chiming in!
–Loren

http://www.cise.ufl.edu/research/sparse/CSparse/CSparse/Source/cs_randperm.c

It uses a minor variant of the shuffle method laz cites on Wikipedia. Mine is not ideal because it uses rand()%(n-k), and that’s not so good because the lower bits from rand() are as “random” as the upper bits. I didn’t write it because I needed statistical purity, though. I use it in dmperm (the one in MATLAB) except that dmperm in MATLAB doesn’t have an option given to the user to use its randomized version.

Is the performance of randperm critical to your application?