It was previously using instance type checking, and heuristic is not fully accurate. Update the logic to parse the dataset graph_def, which should be deterministic about the information we need. This also allows keras to remove the private API symbol usage of tf.data. PiperOrigin-RevId: 332087112 Change-Id: I6db0cfbe3ac00ace2d736b6a2b0a15cf04981185 |
||
---|---|---|
.. | ||
debug_errors.py | ||
debug_fibonacci.py | ||
debug_keras.py | ||
debug_mnist_v1.py | ||
debug_tflearn_iris.py | ||
examples_v1_test.sh |