Min of each column of a sparse matrix?
3 views (last 30 days)
Paras Gupta on 28 Jun 2022
The ‘min’ function generally works for full matrices (or dense matrices) only. In the case of sparse matrices, the ‘min’ function would consider the missing zero elements thereby giving the result as 0.
The following code illustrates how a vector with the minimum of each column can be computed in the case of sparse matrices.
% MATRIX is the given sparse matrix
% First, we use the find the row and column indices for non-zero elements
[ii,jj] = find(MATRIX);
% MIN is the vector with min of each column
% We use the accumarray function to apply the @min function on the selected
% index groups jj
MIN = accumarray(jj,nonzeros(MATRIX),,@min)
You can refer to the documentation on sparse matrices, find, nonzeros, and accumarray for more information to understand how the above code works. A different method as proposed in the following answer can also be referred to compute the desired result. https://in.mathworks.com/matlabcentral/answers/35309-max-min-of-sparse-matrices
Hope this helps!
Jan on 28 Jun 2022
Edited: Jan on 28 Jun 2022
For large inputs a loop is faster than accumarray:
X = sprand(2000, 20000, 0.2);
for k = 1:10
[ii,jj] = find(X);
MIN = accumarray(jj, nonzeros(X), , @min);
for k = 1:10
MIN2 = myMin(X);
function M = myMin(X);
[i1, i2, v] = find(X);
M = Inf(1, size(X, 2));
for k = 1:numel(v)
if M(i2(k)) > v(k)
M(i2(k)) = v(k);