def test_generate_matrix(self):
dtype = self.dtype
x_shape, s_shape = self.x_s_shapes
sv = numpy.random.uniform(0.5, 1.5, s_shape).astype(dtype().real.dtype)
x = testing.generate_matrix(x_shape, dtype=dtype, singular_values=sv)
assert x.shape == x_shape
s = numpy.linalg.svd(
x.astype(numpy.complex128),
full_matrices=False,
compute_uv=False,
)
sv = matrix._broadcast_to(sv, s.shape)
sv_sorted = numpy.sort(sv, axis=-1)[..., ::-1]
rtol = 1e-3 if dtype == numpy.float16 else 1e-7
numpy.testing.assert_allclose(s, sv_sorted, rtol=rtol)
def test_shape_mismatch(self):
with self.assertRaises(ValueError):
testing.generate_matrix((2, 2), singular_values=numpy.ones(3))
def test_negative_singular_values(self):
with self.assertRaises(ValueError):
testing.generate_matrix((2, 2), singular_values=[1, -1])
def test_invalid_dtype_singular_values(self):
with self.assertRaises(TypeError):
testing.generate_matrix((2, 2), singular_values=1 + 0j)
def test_invalid_dtype(self):
with self.assertRaises(TypeError):
testing.generate_matrix((2, 2),
dtype=numpy.int32,
singular_values=1)
def test_invalid_shape(self):
with self.assertRaises(ValueError):
testing.generate_matrix((2, ), singular_values=1)
def test_no_singular_values(self):
with self.assertRaises(TypeError):
testing.generate_matrix((2, 2))
