Add distribution support for incrementing the global step.
authorA. Unique TensorFlower <gardener@tensorflow.org>
Fri, 30 Mar 2018 01:11:09 +0000 (18:11 -0700)
committerTensorFlower Gardener <gardener@tensorflow.org>
Fri, 30 Mar 2018 01:13:48 +0000 (18:13 -0700)
commita8a95bf470877a9f70dfc7e587b30071ebc4ca5d
treeaa465d8192f9bf3dced510d2c10a73fcb8df7650
parentec70d93c3035b6d1b031d0ff22b1493e8b2a67c1
Add distribution support for incrementing the global step.
Don't require Dataset as input to eval and predict even when using
a DistributionStrategy.

PiperOrigin-RevId: 191017191
tensorflow/python/estimator/canned/baseline_test.py
tensorflow/python/estimator/canned/boosted_trees.py
tensorflow/python/estimator/canned/dnn_linear_combined.py
tensorflow/python/estimator/canned/dnn_testing_utils.py
tensorflow/python/estimator/canned/linear_testing_utils.py
tensorflow/python/estimator/estimator.py
tensorflow/python/training/distribute.py