def testStructureVectorMul(self):
blocks = [np.array([[1], [2]]), np.array([[3], [4]]), np.array([[1], [2], [3]])]
indices = np.array([0, 2, 2], dtype=int)
indptr = np.array([0, 2, 2, 3], dtype=int)
bsizei = [2, 2, 3]
bsizej = [1, 1, 1]
M = pus.vbsr_matrix(blocks, indices, indptr, bsizei, bsizej)
v = M * np.array([1, 3, 2])
self.assertTrue(np.array_equal(v, M.tocsr().todense() * np.mat([1, 3, 2]).transpose()))
def setUp(self):
# VS1 is an example variable size block matrix
blocks = [np.array([[1, 2], [3, 4]]), np.array([[1], [2]]), np.array([[1], [2], [3]])]
indices = np.array([0, 1, 1], dtype=int)
indptr = np.array([0, 2, 3], dtype=int)
bsizei = [2, 3]
bsizej = [2, 1]
self.VS1 = pus.vbsr_matrix(blocks, indices, indptr, bsizei, bsizej)
self.VS1D = self.VS1.tocsr().todense()
def testMul(self):
from scipy.sparse import eye, csr_matrix
block = np.array([[1.0, 2], [3, 4]])
indices = np.array([0, 1], dtype=int)
indptr = np.array([0, 1, 2], dtype=int)
bsizei = [2, 2]
bsizej = [2, 2]
# Construct a 2x2 block matrix of 2x2 blocks
b = pus.vbsr_matrix([block, block], indices, indptr, bsizei, bsizej)
# Multiply by the identity at the block level
bi = b * eye(2, 2)
self.assertTrue((b.tocsr().todense() == bi.tocsr().todense()).all())
# Try the same things with VS1
VS1i = self.VS1 * eye(2, 2)
self.assertTrue((self.VS1D == VS1i.tocsr().todense()).all())
# Repeat the identity tests with left multiplication
ib = b.__rmul__(eye(2, 2))
self.assertTrue((b.tocsr().todense() == ib.tocsr().todense()).all())
iVS1 = self.VS1.__rmul__(eye(2, 2))
self.assertTrue((self.VS1D == iVS1.tocsr().todense()).all())
# Now try summing some rows
x = csr_matrix(np.mat([[1], [1]]))
bx = b * x
self.assertTrue((bx.tocsr().todense() == np.bmat([[block], [block]])).all())
# Check that this won't work on the matrix with variable (and incompatible) block sizes
self.assertRaises(ValueError, lambda: self.VS1 * x)
# Sum some columns:
xb = b.__rmul__(x.transpose())
self.assertTrue((xb.tocsr().todense() == np.bmat([[block, block]])).all())