Feeding scipy.sparse() sparse matrices into CVXOPT

Question

[I'm following the answer here]

I am trying to feed sparse matrices in CVXOPT. Consider the following minimal example:

import numpy
import cvxopt
import scipy.sparse

K = 10
n = 36

g_0 = numpy.random.randn(n, K)
d_0 = numpy.zeros(n) + 1.0
g_2 = scipy.sparse.dia_matrix(([d_0], [0]), shape=(n, n))
g_3 = scipy.sparse.dia_matrix(([-d_0], [0]), shape=(n, n))
g_1 = scipy.sparse.coo_matrix(g_0)
g_4 = scipy.sparse.hstack([g_1, g_2, g_3])

A   = cvxopt.spmatrix(g_4.data.tolist(), g_4.col.tolist(), g_4.row.tolist(), size = g_4.shape)

I get:

TypeError: dimension too small

Is this a bug or (more likely) am I misunderstanding this answer?

score 1 · Accepted Answer · answered Oct 24 '16 at 10:56

You just switched the row-column-order to column-row-order within your arguments during the matrix-creation call.

This is in conflict with the argument of size g_4.shape. Look at cvxopt's docs. Size first treats, I (2nd arg), then J (3rd arg).

A   = cvxopt.spmatrix(g_4.data.tolist(), g_4.col.tolist(), g_4.row.tolist(), size = g_4.shape)  # wrong
A   = cvxopt.spmatrix(g_4.data.tolist(), g_4.row.tolist(), g_4.col.tolist(), size = g_4.shape)  # correct

Feeding scipy.sparse() sparse matrices into CVXOPT

1 Answers1