def test_2323_contraction(no, nv, frac):
w = np.ascontiguousarray(np.random.rand(nv, nv, nv, nv))
t_shape = (no, no, nv, nv)
nnz = int(np.prod(t_shape) * frac)
t2 = create_sparse_ndarray(t_shape, nnz)
m = int(t2.size * frac)
# Compress by getting the largest m elements
t2_compressed = SparseTensor4d(t2, t2.shape, m, "largest")
# t2new += lib.einsum('abcd,ijcd->ijab', Wvvvv, tau)
t2_new_np = np.einsum("abcd,ijcd->ijab", w, t2, order="C")
t2_new_fri = np.zeros(t2.shape, order="C")
contract_DTSpT(w, t2_compressed, t2_new_fri, "2323")
npt.assert_almost_equal(t2_new_fri, t2_new_np)
def test_0312_contraction(no, nv, frac):
t_shape = (no, no, nv, nv)
nnz = int(np.prod(t_shape) * frac)
t2 = create_sparse_ndarray(t_shape, nnz)
ovov = np.ascontiguousarray(np.random.rand(no, nv, no, nv))
m = int(t2.size * frac)
# Compress by getting the largest m elements
t2_compressed = SparseTensor4d(t2, t2.shape, m, "largest")
Wakci_np = -0.5 * np.einsum("lckd,ilda->akci", ovov, t2, order="C")
Wakci = np.zeros(Wakci_np.shape, order="C")
contract_DTSpT(ovov, t2_compressed, Wakci, "0312")
npt.assert_almost_equal(Wakci, Wakci_np)
def test_0101_contraction(no, nv, frac):
w = np.ascontiguousarray(np.random.rand(no, no, no, no))
t_shape = (no, no, nv, nv)
nnz = int(np.prod(t_shape) * frac)
t2 = create_sparse_ndarray(t_shape, nnz)
m = int(t2.size * frac)
# Compress by getting the largest m elements
t2_compressed = SparseTensor4d(t2, t2.shape, m, "largest")
t2_new_np = np.einsum("klij,klab->ijab", w, t2, order="C")
t2_new_fri = np.zeros(t2.shape, order="C")
contract_DTSpT(w, t2_compressed, t2_new_fri, "0101")
npt.assert_almost_equal(t2_new_fri, t2_new_np)
def test_1323_contraction(no, nv, frac):
t_shape = (no, no, nv, nv)
nnz = int(np.prod(t_shape) * frac)
t2 = create_sparse_ndarray(t_shape, nnz)
ovov = np.ascontiguousarray(np.random.rand(no, nv, no, nv))
m = int(t2.size * frac)
# Compress by getting the largest m elements
t2_compressed = SparseTensor4d(t2, t2.shape, m, "largest")
Wklij_np = np.einsum("kcld,ijcd->klij", ovov, t2, order="C")
Wklij = np.zeros(Wklij_np.shape, order="C")
contract_DTSpT(ovov, t2_compressed, Wklij, "1323")
npt.assert_almost_equal(Wklij, Wklij_np)
def test_sparse_init(no, nv, frac):
print()
# Setup random array
a = np.ascontiguousarray(np.random.rand(no, no, nv, nv))
m = int(a.size * frac)
# Keep the tests small
if m > 200:
print(f"M={m}")
raise ValueError(
"M >= 1000, choose a smaller matrix or a smaller fraction.")
# Compress by getting the largest m elements
a_compressed = SparseTensor4d(a, a.shape, m, "largest")
# a_compressed.print()
# Check that all elements in a_compressed have the right value/index
for mi in range(m):
idx, value = a_compressed.get_element(mi)
npt.assert_equal(value, a[tuple(idx)])