TensorFlow 1 version View source on GitHub

Returns the global Policy.

The global policy is the default policy used for layers, if no policy is passed to the layer constructor. If no policy has been set with keras.mixed_precision.experimental.set_policy, this will return a policy constructed from tf.keras.backend.floatx() (floatx defaults to float32).

If TensorFlow 2 behavior has been disabled with tf.compat.v1.disable_v2_behavior(), this will instead return a special "_infer" policy which infers the dtype from the dtype of the first input the first time the layer is called. This behavior matches the behavior that existed in TensorFlow 1.

See tf.keras.mixed_precision.experimental.Policy for more information on policies.

The global Policy.