Find matching rows of cell arrays containing strings

Hi,

I am working with a pair of cell arrays, containing strings. The cell arrays are of the same size, and they contain the same rows, but in different order. E.g.:

A = {'ABC' '123'; 'A' '100'; 'C' '0'};
B = {'A' '100'; 'C' '0'; 'ABC' '123'};

It will always be the case that the rows are simply ordered differently. What I am trying to accomplish is to find this ordering. That is, row 1 in A becomes row 3 in B, row 2 in A becomes row 1 in B, and row 3 in A becomes row 2 in B. So I need some sort of output like:

output = [3 2 1]

or something containing the same kind of information. I just need to know how the rows are permuted. I've tried using cellfun with ismember and so on, but I can't seem to make that handle rows, as opposed to cell by cell. I've also tried converting A and B into arrays numbers (because it's my impression that it's easier to deal with numbers than strings), but then I run into the problem that not all the cells have the same size, so there's a concatenating error.

In short, I've tried a bunch of different cellfun's with strfind, ismember, etc., but I haven't found a method that works for rows specifically.

Any help is greatly appreciated!

Answer 1

[Azzi Abdelmalek]

B=circshift(A,[-1 0])

Edit

A = {'ABC' '123'; 'A' '100'; 'C' '0'};
B = {'A' '100'; 'C' '0'; 'ABC' '123'}
[n,m]=size(A);
[~,idx]=ismember(A(:),B(:));
[ii,jj]=ind2sub([n m],idx);
output=ii(1:n)';

[Peder]

Hi Azzi,

The particular example was pretty simple. In general A and B won't be cyclic permutations of each other.

[Azzi Abdelmalek]

B([3 1 2],:)=A

[Peder]

Yeah, but the question isn't about finding B from A; it's about finding that vector, [3 1 2] (this can generally be more complicated), using only A and B.

[Azzi Abdelmalek]

n=size(A,1);
A1=arrayfun(@(x) strjoin(A(x,:),'_'),(1:n)','un',0)
m=size(B,1)
B1=arrayfun(@(x) strjoin(B(x,:),'_'),(1:m)','un',0)
[~,idx]=ismember(A1,B1)

[Peder]

Hi Azzi,

That's interesting, strjoin was the method I used earlier (I guess I should have mentioned that), but the problem with that is that it quite slow. That is, it spends a lot of time (at least on my computer) just joining strings together. Although it works, I was hoping there was a faster solution.

Many thanks for the help with this!

[Peder]

Alternatively, if I took a single row from A (and then looping through the rows) and compared it with B, would that work? That is, say size(A)=[1 4], size(B)=[100 4], could I somehow get a 100 by 4 array of ligicals, indicating whether each of the four elements of A matches the elements of each row in B?

Maybe that was unclear, but I would ideally like to avoid strjoin, as it's been slowing down my code so far.

[Azzi Abdelmalek]

Try this, it's much faster

A = {'ABC' '123'; 'A' '100'; 'C' '0'};
B = {'A' '100'; 'C' '0'; 'ABC' '123'}
[n,m]=size(A);
[~,idx]=ismember(A(:),B(:));
[ii,jj]=ind2sub([n m],idx);
output=ii(1:n)';

[Azzi Abdelmalek]

Test the speed

A = {'ABC' '123'; 'A' '100'; 'C' '0'};
B=A([3 1 2],:)
nn=5000
A=repmat(A,nn,1);
B=repmat(B,nn,1);
tic
n=size(A,1);
A1=arrayfun(@(x) strjoin(A(x,:),'_'),(1:n)','un',0);
m=size(B,1);
B1=arrayfun(@(x) strjoin(B(x,:),'_'),(1:m)','un',0);
[~,output1]=ismember(A1,B1);
t1=toc

tic
[n,m]=size(A);
[~,idx]=ismember(A(:),B(:));
[ii,jj]=ind2sub([n m],idx);
output=ii(1:n);
t2=toc
isequal(output1,output)

Result

Elapsed time is 9.558223 seconds.
Elapsed time is 0.034396 seconds.

[Peder]

Hi Azzi,

That's really impressive! I'll try that out in my code and see how it goes! Based on your tests, it looks extremely promising! Many thanks for the help! :)