Slow constraints addition in Python vs. C++
AnsweredHello,
I'm getting troubles with my model. Here is my code :
T = model.addMVar((numnonzero,), vtype=GRB.BINARY)
for s1 in range(numnonzero):
for s2 in range(s1+1,numnonzero):
row1 = nonzero[s1][0]
col1 = nonzero[s1][1]
row2 = nonzero[s2][0]
col2 = nonzero[s2][1]
if (row1 == row2 or col1 == col2 or (X[row1][col2] > 0.001 and X[row2][col1] > 0.001)):
model.addConstr(T[s1]+T[s2]<=1)
where nonzero is the indice set of the non-zero values of X and numnonzero is the length of this set. For some 20-by-20 matrices, it takes more than 20 seconds only to build this model (even if the optimization itself is quite faster). For example, I have the same model in C++ and the time to construct the model is almost 0, here is the code :
int indices[2]
double coefficients[2] = {1.0,1.0};
for (std::size_t s1 = 0; s1 < numnonzero; ++s1):
{
for (std::size_t s2 = s1 + 1; s2 < numnonzero; ++s2)
{
std::size_t row1 = nonzeros[s1].row;
std::size_t col1 = nonzeros[s1].col;
std::size_t row2 = nonzeros[s2].row;
std::size_t col2 = nonzeros[s2].col;
if (row1 == row2 || col1 == col2 || (isNonzero(row1,col2) && isNonzero(row2,col1)))
{
indices[0]=s1;
indices[1]=s2;
GRBaddconstr(model,2,&indices[0],&coefficients[0], '<', 1.0, nullptr);
}
}
}
So my question is: why does it take so long in Python while I almost can get it instantly in my C++ model, and how to manage this ?
Have a good day,
Timothy
-
-
Gurobi Staff
Hi Timothy,
It looks like you are mixing matrix-friendly and term-based modeling in Python.
Did you try to use Model.addVars() instead of Model.addMVar() ?Best regards,
Marika0 -
-
Hello Marika, thanks for this answer!
I first wrote a "matrix-based modeling" but the time to build the model was even worse, so I tried like this. However, just using Model.addVars() makes the generation 50 times faster, thank you!
0 -
Please sign in to leave a comment.
Comments
2 comments