How to run find function faster?

Dear All,

I have an array A with 6000 elements in integers, for example, "10", "12", "5" and others. Actually array A saves the number of non-zero elements in each rows of a sparse matrix B. I want to do some calculations on those selected rows in B with certian number of non-zero elements. I want to find out the indecis of selected values "10", "12" and "5" in A.

I did the following:

[~,ind1] = find(A==10);

[~,ind2] = find(A==12);

[~,ind3] = find(A==5).

I found it took about 1 second which is too slow. I am wondering if there is a faster way to do this

Thanks a lot.

Benson

[Stephen23]

"I am wondering if there is a faster way to do this"

It rather depends on what you are doing with those indices, which you did not tell us.

[Benson Gou]

Hi, Stephen,

Thanks for your reply. Actually array A saves the number of non-zero elements in each rows of a sparse matrix. I want to do some calculations on those selected rows with certian number of non-zero elements.

Thanks.

Benson

[Stephen23]

As dpb already wrote, you should first try using (faster and more efficient) logical indexing, before using FIND.

[James Tursa]

The overall goal isn't to make find( ) faster ... the overall goal is to make your code run faster. For that, we need to know what your overall code is doing. E.g., maybe there is a way to get your result without explicitly using find( ) at all. Please give us details about how you are using the results from find( ) downstream in your code.

[Paul]

Is anyone else surprised that these three calls to find takes "about 1 second" for an 6000-element array of integers?

[dpb]

Not without more details; "know nuthink!" of just how was timed or what was the platform.

Also, I've not compared, but that the A array is sparse could well be a factor as well...

Answer 1

[dpb]

Generally one doesn't need find explicitly at all; just use the logical addressing vector returned by the logical test.

Alternatives for something like the above would quite possibly be findgroups and splitapply depending upon just what is to be done next.

Or, of course, there are all the binning functions such as discretize that might come into play.

Without the end use here, it's hard to say specifically.

One note, however, is that creating sequentially-numbered variables such as above is generally a sign that the code factorization is not optimal in MATLAB -- arrays or cell arrays would be far better in all likelihood. Once you've created such a set of variables, you've made using them difficult for anything other than just duplicating the same code over and over...

[Benson Gou]

Hi, dpb,

Thanks for your reply. But I still do not know how to do it after several times tries.

Best regards,

Benson

[dpb]

Well, we've got to see what you're trying, specifically, to be able to answer specifically.

No code, not even an error message isn't much for even the crystal ball to work with...