Reduce computational time when calling cellfun

I have:

for i=1:length(cellarray)
    id1=cellfun(@(x)any(ismember('1',x)),cellarray{i})
    id2=cellfun(@(x)any(ismember('2',x)),cellarray{i});
    id3=cellfun(@(x)any(ismember('3',x)),cellarray{i});
    ...  
end

which I would like to rewrite as something like:

for i=1:length(cellarray)
    value=cellfun(@(x)x(1),cellarray{i})
    id1=any(ismember('1',value));
    id2=any(ismember('2',value));
    id3=any(ismember('3',value));
    ...
end

in order to try to reduce computational time. Here,

cellarray={'AA','KK','AKs','QQ','AKo',...};

The problem is, that the first approach correctly evaluates each cell and return the results into a vector id1, id2, ... whereas the second approach considers all elements in the cell array simultaneously. Changing "any(ismember(...))" to "ismember(...)" also only evaluates the entire array.

How do I make sure that each cell is evaluated? Say the cell array has 5 elements, then id1 should have 5 elements containing 0's or 1's.

[Jan]

Are the elements of cellarray are strings? No, sorry, they are cells of cell strings, correct?

Please do not let us guess, but provide some sample data in valid Matlab syntax. And please insert them by editing the question, not as comment or answer.

Comments to the code:

1. strfind is much more efficient than ismember(CHAR).
2. Creating a bunch of variables called id1, id2, ... is a bad programming practize. id{1}, id{2}, ... would be more convenient.
3. id1=any(ismember('1',x)),value) contains 2 left and 3 right parenthesis. Please fix this.

Answer 1

[Jan]

for i = 1:length(cellarray)
  S   = cellarray{i};
  id1 = any(strfind(S, '1'));
  id2 = any(strfind(S, '2'));
  id3 = any(strfind(S, '3'));
  ...  
end

If the cell array is very large and you look for a few characters only, this approach could be more efficient, when you want to process the complete cell without a loop:

id1 = ~strcmp(cellarray, strrep(cellarray, '1', '*'));
...

[Ole Hansen]

Thanks for the comments. I would prefer to use ismember to start with, since it easily identifies individual characters within a string. With ismember, however, I have trouble implementing the code as mentioned in the original post.

[Jan]

ISMEMBER is slow, because it has a large overhead, e.g. to sort the characters in the string. If the strings have less than about 20 characters, this wastes time. Because you have asked for a faster method, I suggested a faster method.

Let me repeat, that it would be helpful to create an helpful answer efficiently, when you post all relevant details. What kind of output do you need? What is the purpose of cellfun(@(x)x(1),cellarray{i})?

Why does cellfun(@(x)any(ismember('1',x)),cellarray{i}) work at all, if cellarray is a cell string? Then cellarray{i} is a string, and not a apropriate input for cellfun().

Summary: At first I did not know the inputs. Now, after you have posted the input, I do not understand how your program could work with it. The wanted output is still not defiend exactly. So how do you expect us to accelerate your function?

Please, Ole, post a running example which we can run using copy&paste. This is essential for speed improvements.

[Ole Hansen]

I used strfind instead of ismember and found a reduction in computational speed by around 25% I assume this reduction is specifically for my script.

I cannot answer all of your questions at hand; there are some things I do not understand as of now - I will return shortly.