def test_sparse_csc(uniform_sel):
sparse_csc = gen_sparse_matrix(M, D, np.float32, 0.01).transpose_csc()
centers = uniform_sel.select(sparse_csc, None, 5)
assert centers.size() == (5, M), "UniformSel did not output correct size"
assert centers.is_csc is True, "UniformSel did not preserve sparsity correctly"
assert centers.dtype == sparse_csc.dtype
assert centers.device == sparse_csc.device
def test_sparse_csr(uniform_sel):
sparse_csr = gen_sparse_matrix(M, D, np.float32, 0.01)
centers = uniform_sel.select(sparse_csr, None, 100)
assert centers.size() == (100, D), "UniformSel did not output correct size"
assert centers.is_csr is True, "UniformSel did not preserve sparsity correctly"
assert centers.dtype == sparse_csr.dtype
assert centers.device == sparse_csr.device
def s_B(self):
B = gen_sparse_matrix(m,
self.sparse_dim,
np.float64,
density=self.sparse_density,
seed=14)
Bd = torch.from_numpy(B.to_scipy().todense())
return B, Bd
def s_A(self):
A = gen_sparse_matrix(n,
self.sparse_dim,
np.float64,
density=self.sparse_density,
seed=14)
Ad = torch.from_numpy(A.to_scipy().todense())
return A, Ad
def test_sparse_csc():
uniform_sel = UniformSelector(np.random.default_rng(0), num_centers=5)
sparse_csc = gen_sparse_matrix(M, D, np.float32, 0.01).transpose_csc()
centers: falkon.sparse.SparseTensor = uniform_sel.select(sparse_csc, None)
assert centers.size() == (5, M), "UniformSel did not output correct size"
assert centers.is_csc is True, "UniformSel did not preserve sparsity correctly"
assert centers.dtype == sparse_csc.dtype
assert centers.device == sparse_csc.device
centers, idx = uniform_sel.select_indices(sparse_csc, None)
assert centers.device == idx.device
assert idx.dtype == torch.long
assert len(idx) == centers.shape[0]
def test_sparse_csr(uniform_sel):
sparse_csr = gen_sparse_matrix(M, D, np.float32, 0.01)
centers: falkon.sparse.SparseTensor = uniform_sel.select(sparse_csr, None)
assert centers.size() == (num_centers,
D), "UniformSel did not output correct size"
assert centers.is_csr is True, "UniformSel did not preserve sparsity correctly"
assert centers.dtype == sparse_csr.dtype
assert centers.device == sparse_csr.device
centers, idx = uniform_sel.select_indices(sparse_csr, None)
assert centers.device == idx.device
assert idx.dtype == torch.long
assert len(idx) == centers.shape[0]
def s_B():
m, d, density = 550, 20000, 1e-4
B = gen_sparse_matrix(m, d, np.float64, density=density, seed=14)
Bd = torch.from_numpy(B.to_scipy().todense())
return B, Bd
def s_A():
n, d, density = 500, 20000, 1e-4
A = gen_sparse_matrix(n, d, np.float64, density=density, seed=14)
Ad = torch.from_numpy(A.to_scipy().todense())
return A, Ad
def s_A():
A = gen_sparse_matrix(s_n, s_d, np.float64, density=density, seed=14)
Ad = torch.from_numpy(A.to_scipy().todense())
return A, Ad
def s_B():
B = gen_sparse_matrix(s_m, s_d, np.float64, density=density, seed=14)
Bd = torch.from_numpy(B.to_scipy().todense())
return B, Bd
