row indices of an array which is a subset of another array

Eg. b is a matrix of size(17,2). (BIG MATRIX)

b(:,1)= [ 7;6;5;4;4;3;2;1;20;19;18;17;16;20;21;22;23]

b(:,2) = [30;29;28;27;8;9;10;11;8;12;13;14;15;27;26;25;24]

Now, I need the row indices from this array b which has matching entries in a vector a. The size of vector a is smaller than b. But if the indices are repeated, I want the neighboring row entry.

What I mean by neighboring row entry?

lets say a is : [4; 3; 2; 1]. The entries of a will either be in b(:,1) or b(:,2). there would surely not be any intersection. When I search for a in b(:,1) or (b(:,2)), i will get the row indices [5; 6; 7; 8]. (These are the row indices.) When I search for a = [20;21;22;23] in b(:,1), I want it to return the row indices as [14 15 16 17]..

When a is : [7;6;5;4], I want the row indices [1 2 3 4].

"ismember" allows me to return the first or the last row entries. My requirement here is a bit different.

Answer 1

[Jan]

Maybe your mean:

b = cat(2, [ 7;6;5;4;4;3;2;1;20;19;18;17;16;20;21;22;23], ...
           [30;29;28;27;8;9;10;11;8;12;13;14;15;27;26;25;24]);

a = [20; 21; 22; 23];

[~, Result] = ismember(a, b(:, 1), 'legacy');
if ~any(Result)
  [~, Result] = ismember(a, b(:, 2), 'legacy');
end

But I detest the 'legacy' flag. Even the documentation is vague about its meaning:

 preserves the behavior of the ismember function from R2012b and prior
 releases using any of the input arguments in previous syntaxes.

Without knowing what the prior (to what?!) releases did, this description is useless. There is no way to let unique search for 'stable' and 'last' at the same time.

Therefore I'd prefer:

b = cat(2, [ 7;6;5;4;4;3;2;1;20;19;18;17;16;20;21;22;23], ...
           [30;29;28;27;8;9;10;11;8;12;13;14;15;27;26;25;24]);
[B1, Index1] = uniqueStableLast(b);
[B2, Index2] = uniqueStableLast(b);

[~, loc] = ismember(a, B1);
if any(loc)
   Result = Index1(loc);
else
  [~, loc] = ismember(a, B2);
   Result  = Index2(loc);
end

With:

function [B, AI, BI] = uniqueStableLast(A)
nA = numel(A);
if nA
   % Stable with choosing the last element:
   [As, SV] = sort(A(:), 'descend');
   As       = As(nA:-1:1);
   SV       = SV(nA:-1:1);

     % Stable with the first element would be:
     % [As, SV] = sort(A(:));

     if nargout == 1
       UV(SV) = [false; diff(As) ~= 0];
       B      = A(UV);
     else  % Indices requested:
        UV      = [false; diff(As)] ~= 0];
        UVs(SV) = UV;
        AI      = find(UVs);
        B       = A(UVs);

        if nargout == 3 
          % Complex creation of BI such that BB(BI) == A:
          v      = zeros(1, nA);
          v(AI)  = 1:length(AI);    % Sequence related to BB
          vs     = v(SV);           % Sorted like A
          vf     = vs(vs ~= 0);     % Just the filled entries
          BI(SV) = vf(cumsum(UV));  % Inflate multiple elements
        end
     end
  else
     B  = [];
     AI = [];
     BI = [];
  end

By the way: This is even faster than Matlab's unique('stable').

[Chaitanya Sanghavi]

Thanks a lot for your effort. !! I am very much concerned for the performance which is nice since you say it works faster than matlab's unique("stable").