使用matlab索引矩阵中的行

Question

Suppose I have an empty m-by-n-by-p dimensional cell called "cellPoints", and I also have a D-by-3 dimensional array called "cellIdx" where each row i contains the subscripts in "cellPoints". Now I want to compute "cellPoints" so that cellPoints{x, y, z} contains an array of row numbers in "cellIdx".

A naive implementation could be

for i = 1:size(cellIdx, 1)
    cellPoints{cellIdx(i, 1), cellIdx(i, 2), cellIdx(i, 3)} = ...
    [cellPoints{cellIdx(i, 1), cellIdx(i, 2), cellIdx(i, 3)};i];
end

As an example, suppose

cellPoints = cell(10, 10, 10);% user defined, cannot change
cellIdx = [1, 3, 2;
           3, 2, 1;
           1, 3, 2;
           1, 4, 2]

Then

cellPoints{1, 3, 2} = [1;3];
cellPoints{3, 2, 1} = [2];
cellPoints{1, 4, 2} = [4];

and other indices of cellPoints should be empty

Since cellIdx is a large matrix and this is clearly inefficient, are there any other better implementations?

I've tried using unique(cellIdx, 'rows') to find unique rows in cellIdx, and then writing a for-loop to compute cellPoints, but it's even slower than above.

Answer 1

See if this is faster:

cellPoints = cell(10,10,10); %// initiallize to proper size
[~, jj, kk] = unique(cellIdx, 'rows', 'stable')
sz = size(cellPoints);
sz = [1 sz(1:end-1)];
csz = cumprod(sz).'; %'// will be used to build linear index
ind = 1+(cellIdx(jj,:)-1)*csz; %// linear index to fill cellPoints
cellPoints(ind) = accumarray(kk, 1:numel(kk), [], @(x) {sort(x)});

Or remove sort from the last line if order within each cell is not important.