| Commit message (Collapse) | Author | Age |
|
|
|
| |
PiperOrigin-RevId: 214824023
|
|
|
|
|
|
|
|
| |
self.test_session() has been deprecated in 9962eb5e84b15e309410071b06c2ed2d6148ed44 as its name confuses readers of the test. Moving to cached_session() instead which is more explicit about:
* the fact that the session may be reused.
* the session is not closed even when doing a "with self.test_session()" statement.
PiperOrigin-RevId: 212336464
|
|
|
|
|
|
|
|
| |
* Adding pre_train_ops to ConstrainedMinimizationProblem. These are tf.Operations that should be executed before each training step.
* Updating ConstrainedOptimizers to have the train_ops they create depend on the pre_train_ops from the ConstrainedMinimizationProblem.
PiperOrigin-RevId: 208871419
|
| |
|
| |
|
|\
| |
| | |
Fix warning in constrained_optimization test
|
| | |
|
| |
| |
| |
| | |
Signed-off-by: Yong Tang <yong.tang.github@outlook.com>
|
|/
|
|
|
|
|
|
|
| |
In constrained_optimization test, keep_dims was
used for reduce_sum. Since keep_dims has been deprecated
it generates unnecessary warning. This fix updates
keep_dims -> keepdims to disable the warning.
Signed-off-by: Yong Tang <yong.tang.github@outlook.com>
|
|
interfaces: ConstrainedMinimizationProblem, which specifies a constrained optimization problem, and ConstrainedOptimizer, which is slightly different from a tf.train.Optimizer, mostly due to the fact that it is meant to optimize ConstrainedMinimizationProblems. In addition to these two interfaces, three ConstrainedOptimizer implementations are included, as well as helper functions which, given a set of candidate solutions, heuristically find the best candidate (to the constrained problem), or the best distribution over candidates.
For more details, please see our arXiv paper: "https://arxiv.org/abs/1804.06500".
PiperOrigin-RevId: 193999550
|