Justin Winokur

I originally wrote the search with nested for loops (I know, I should be shot), but then to "optimize" it, I used ismember and find to make it better. Unfortunately, it slowed it down rather significantly. I was excited to see this code, but it turn out that it does it the same way. Too bad.

This is a great program. I use it almost every day. It works well 99% of the time. Sometimes, if there is a really complicated plot, it does not export PDF as a vector file. In those cases, I use Matlab's print command to eps and then let Mac OS X convert the eps to pdf. Even eps2pdf command line doesn't do it.

Do you know what makes it fail to export vector images sometimes?

Either way, great program.

A relative tolerance can be useful, but as usual for small values (e.g. 1e-20) further problems will occur.

I think it is a more important problem, that your function does not perform what its name advertises: Not the *closest* row is replied, but the first one inside the tolerance. Although a message is displayed, I'd prefer to use another name or algorithm, e.g.:

[d, ind] = min(sum(bsxfun(@minus, full_array, row) .^ 2, 2));
if d > tol, ind = -1; end

Replying the empty matrix could be more intuitive than -1, when no matching row was found.

"(min(distance)-tol)<=eps" is a bad comparison: Imagine, that the distances and tolerance has a magnitude of 1e80. Then comparing with EPS is not sufficient - eps(tol) would be smarter.
Instead of calculating the square root of the distances, it would be much more efficient to compare the squared distances with the squared tolerance.
This would reply the row with the smallest distance, not the first one below the tolerance:
[value, found_index] = min(distance - tol)