MATLAB Answers


How to vectorize strfind

Asked by Paolo Binetti on 22 Dec 2016
Latest activity Edited by Paolo Binetti on 23 Dec 2016
Is it possible to use strfind in a vectorized way? Suppose I want to get find not just one pattern inside a string, but several of them at the same time, so that the output would not be a vector of indexes, but a matrix: is it possible?
Concretely, take a string 'TRSDGHNENJRRDSENTRFDGDGT'. I want to find 'TR', 'DG', and 'EN'. The output of the function would be a matrix 3 x length(string) where line one are zeros and ones at indexes relative to 'TR', line two for 'DG' and line three for 'EN'. Possible?
The purpose is to avoid a for loop which even with pre-allocation is time-consuming. But I actually don't even know if this vectorization I am thinking of would be quicker.

  1 Comment

Image Analyst
on 23 Dec 2016
Define quick for you. Exactly how long is your for loop solution taking? I can do around five hundred million iterations of a for loop in less than half a second. How many billions of iterations are you doing in your loop, and how long is it actually taking?

Sign in to comment.

1 Answer

Answer by Walter Roberson
on 22 Dec 2016

In R2016b,
targets = ['TR'; 'DG'; 'EN'];
output = [targets(:,1) == S(1:end-1) & targets(:,2) == S(2:end), false(size(targets,1),1)];
In earlier versions you would need to use bsxfun()


Thank you but I do not have these functions. On the other hand, I found that I can do it by converting the char array containing all the patterns to be searched into a cell array, and feed this to strfind. But the cell array is too memory-intensive, and the vectorized strfind is too slow, when applied to my problem (not the example).
Walter Roberson
on 23 Dec 2016
The bsxfun version would be
output = [bsxfun(@eq, targets(:,1), S(1:end-1)) & bsxfun(@eq, targets(:,2), S(2:end)), false(size(targets,1),1)];
If you do not have bsxfun then you must be using a version before R2007a, and if you are then it is important for us to know that so that we do not keep suggesting features you cannot use.

Sign in to comment.