Originally posted by: ilovetoyota

If I need to construct very big matrix with zero and non-zero coefficient,

it is very hard to count NUMNZ, is there any easy way to construct lp, such that I can just add row by row, without caring how many non-zero coefficient?

Thanks
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

You can do CPXnewcols() to create empty columns (variables) and then
use CPXaddrows() to add the constraints row-by-row.
You will have to compute the number of non-zeroes per row, though.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: ilovetoyota

Is there any way to bypass the step of counting the number of zeros or non-zeros per row etc?

Or is there any published code on the internet that can help me to read in a matrix directly? How do most of you handle the input of a matrix?

Thanks a lot!
========================================
You can do CPXnewcols() to create empty columns (variables) and then
use CPXaddrows() to add the constraints row-by-row.
You will have to compute the number of non-zeroes per row, though.
#CPLEXOptimizers
#DecisionOptimization

Originally posted by: SystemAdmin

I don't see why counting the number of non-zeroes in a row is such a big issue for you.
Assume you have the number of columns in variable 'cols' and the row in 'row':

int cols = ...;
double *row = ...;
 
int i, nzs = 0;
double *compressed = malloc(sizeof(*compressed) * cols);
 
if ( !compressed ) {
   /* Out of memory. */
   ...
}
 
for (i = 0; i < cols; ++i) {
   if ( fabs(row[i]) > 0.0 =
      compressed[nzs++] = row[i];
}

After that you have the non-zeroes in 'compressed' and the number of non-zeroes in 'nzs'.
I guess most people store their matrices in compressed form and so don't have the issue
of reading a dense matrix.

You can also try to pass the dense row to CPXaddrows(). I think the zeroes in the row
won't do any harm to CPLEX. It is just a big waste of space unless the matrix is dense.
#CPLEXOptimizers
#DecisionOptimization