experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Change linear_operator_test_util.random_postive_definite_matrix to return nicer conditioned matrix by default (a Wishart matrix). Enable tests in oss again.

Browse Source 
PiperOrigin-RevId: 294817440
Change-Id: Ib9a976e9c78fba41863639d7c158e2a3b25523e8
This commit is contained in:
Ian Langmore 2020-02-12 19:41:48 -08:00 committed by TensorFlower Gardener
parent cec24afe2d
commit 4d5741f1d4
2 changed files with 26 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
tensorflow/python/kernel_tests/linalg/BUILD
View File

@ -260,7 +260,6 @@ cuda_py_test(
srcs = ["linear_operator_full_matrix_test.py"], srcs = ["linear_operator_full_matrix_test.py"],
shard_count = 5, shard_count = 5,
tags = [ tags = [
"no_oss", # b/149346156
"noasan", "noasan",
"optonly", "optonly",
], ],

34
tensorflow/python/ops/linalg/linear_operator_test_util.py
View File

@ -825,15 +825,23 @@ class NonSquareLinearOperatorDerivedClassTest(LinearOperatorDerivedClassTest):
return 2 return 2
def random_positive_definite_matrix(shape, dtype, force_well_conditioned=False): def random_positive_definite_matrix(shape,
"""[batch] positive definite matrix. dtype,
oversampling_ratio=4,
force_well_conditioned=False):
"""[batch] positive definite Wisart matrix.
A Wishart(N, S) matrix is the S sample covariance matrix of an N-variate
(standard) Normal random variable.
Args: Args:
shape: `TensorShape` or Python list. Shape of the returned matrix. shape: `TensorShape` or Python list. Shape of the returned matrix.
dtype: `TensorFlow` `dtype` or Python dtype. dtype: `TensorFlow` `dtype` or Python dtype.
force_well_conditioned: Python bool. If `True`, returned matrix has oversampling_ratio: S / N in the above. If S < N, the matrix will be
eigenvalues with modulus in `(1, 4)`. Otherwise, eigenvalues are singular (unless `force_well_conditioned is True`).
chi-squared random variables. force_well_conditioned: Python bool. If `True`, add `1` to the diagonal
of the Wishart matrix, then divide by 2, ensuring most eigenvalues are
close to 1.
Returns: Returns:
`Tensor` with desired shape and dtype. `Tensor` with desired shape and dtype.
@ -843,11 +851,21 @@ def random_positive_definite_matrix(shape, dtype, force_well_conditioned=False):
shape = tensor_shape.TensorShape(shape) shape = tensor_shape.TensorShape(shape)
# Matrix must be square. # Matrix must be square.
shape.dims[-1].assert_is_compatible_with(shape.dims[-2]) shape.dims[-1].assert_is_compatible_with(shape.dims[-2])
shape = shape.as_list()
n = shape[-2]
s = oversampling_ratio * shape[-1]
wigner_shape = shape[:-2] + [n, s]
with ops.name_scope("random_positive_definite_matrix"): with ops.name_scope("random_positive_definite_matrix"):
tril = random_tril_matrix( wigner = random_normal(
shape, dtype, force_well_conditioned=force_well_conditioned) wigner_shape,
return math_ops.matmul(tril, tril, adjoint_b=True) dtype=dtype,
stddev=math_ops.cast(1 / np.sqrt(s), dtype.real_dtype))
wishart = math_ops.matmul(wigner, wigner, adjoint_b=True)
if force_well_conditioned:
wishart += linalg_ops.eye(n, dtype=dtype)
wishart /= math_ops.cast(2, dtype)
return wishart
def random_tril_matrix(shape, def random_tril_matrix(shape,