SciPy sparse matrix as input format for the jacobian for a linear constraint

[ruben-haalebos]

Hi all,

I'm a little confused about the use of the SciPy sparse matrix format when specifying the jacobian for a linear constraint. The documentation states that "We have created a linked list sparse matrix using scipy.sparse. Any SciPy sparse matrix format can be accepted. ". However, when I provide a Scipy sparse matrix I get a warning message that is issued by the ConvertToCOO function

"UserWarning: Using scipy.sparse matrices with pyOptSparse is VERY STRONGLY discouraged. Please use the simplified pyOptSparse format which allows for fixed sparsity structure and explicit zeros in the matrix. There is no way to guarantee a fixed sparsity structure with scipy matrices which is what the underlying optimizers require. Using scipy.sparse matrices may cause unexpected errors."

When I first convert my SciPy sparse matrix to pyOptSparse format, for example by using the convertToCSR function, I have no warning message but I'm not sure it's the intended use.

Many thanks,
Ruben

[ewu63]

What you're describing is correct, pyOptSparse's sparse matrix format is preferred, and you can use e.g. convertToCSR to convert a SciPy matrix.

[ruben-haalebos]

Thank you for your answer !