def test_svd_decomposition():
tensor = np.ones([2, 3, 4, 5, 6])
np_res = numpy_backend.NumPyBackend().svd_decomposition(tensor, 3)
sh_res = shell_backend.ShellBackend().svd_decomposition(tensor, 3)
for x, y in zip(np_res, sh_res):
assert x.shape == y.shape
def test_concat_shape():
shapes = [(5, 2), (3, ), (4, 6)]
result = shell_backend.ShellBackend().concat_shape(shapes)
assert result == (5, 2, 3, 4, 6)
def test_prod():
result = shell_backend.ShellBackend().prod(np.ones([3, 5, 2]))
assert result == 30
def assertBackendsAgree(f, args):
np_result = getattr(numpy_backend.NumPyBackend(), f)(**args)
sh_result = getattr(shell_backend.ShellBackend(), f)(**args)
assert np_result.shape == sh_result.shape
def test_sparse_shape():
backend = shell_backend.ShellBackend()
tensor = backend.randn((2, 3, 4), seed=10)
np.testing.assert_allclose(backend.sparse_shape(tensor), tensor.shape)
def test_shape_tuple():
tensor = np.ones([3, 5, 2])
np_result = numpy_backend.NumPyBackend().shape_tuple(tensor)
sh_result = shell_backend.ShellBackend().shape_tuple(tensor)
assert np_result == sh_result
def test_broadcast_left_multiplication():
backend = shell_backend.ShellBackend()
tensor1 = backend.randn((3, ))
tensor2 = backend.randn((3, 4, 2))
out = backend.broadcast_left_multiplication(tensor1, tensor2)
np.testing.assert_allclose(out.shape, [3, 4, 2])
def test_broadcast_left_multiplication_raises():
backend = shell_backend.ShellBackend()
tensor1 = backend.randn((3, 3))
tensor2 = backend.randn((3, 4, 2))
with pytest.raises(ValueError):
backend.broadcast_left_multiplication(tensor1, tensor2)
def test_matrix_inv_raises():
backend = shell_backend.ShellBackend()
matrix = backend.randn((4, 4, 4), seed=10)
with pytest.raises(ValueError):
backend.inv(matrix)
def test_matrix_inv():
backend = shell_backend.ShellBackend()
matrix = backend.randn((4, 4), seed=10)
inverse = backend.inv(matrix)
assert inverse.shape == matrix.shape
def index_update():
backend = shell_backend.ShellBackend()
tensor_1 = np.ones([2, 3, 4])
tensor_2 = backend.index_update(tensor_1, tensor_1 > 0.1, 0)
assert tensor_1.shape == tensor_2.shape
def test_eigh():
matrix = np.ones([3, 3])
vals, vecs = shell_backend.ShellBackend().eigh(matrix)
assert vals.shape == (3, )
assert vecs.shape == (3, 3)
def test_einsum(): expression = "ab,bc->ac" tensor1, tensor2 = np.ones([5, 3]), np.ones([3, 6]) np_result = numpy_backend.NumPyBackend().einsum(expression, tensor1, tensor2) sh_result = shell_backend.ShellBackend().einsum(expression, tensor1, tensor2) assert np_result.shape == sh_result.shape