tf.compat.v1.enable_v2

In TensorFlow 2.0, iterating over a TensorShape instance returns values.

tf.compat.v1.enable_v2_tensorshape()

This enables the new behavior.

Concretely, tensor_shape[i] returned a Dimension instance in V1, but it V2 it returns either an integer, or None.

Examples:

#######################
# If you had this in V1:
value = tensor_shape[i].value

# Do this in V2 instead:
value = tensor_shape[i]

#######################
# If you had this in V1:
for dim in tensor_shape:
  value = dim.value
  print(value)

# Do this in V2 instead:
for value in tensor_shape:
  print(value)

#######################
# If you had this in V1:
dim = tensor_shape[i]
dim.assert_is_compatible_with(other_shape)  # or using any other shape method

# Do this in V2 instead:
if tensor_shape.rank is None:
  dim = Dimension(None)
else:
  dim = tensor_shape.dims[i]
dim.assert_is_compatible_with(other_shape)  # or using any other shape method

# The V2 suggestion above is more explicit, which will save you from
# the following trap (present in V1):
# you might do in-place modifications to `dim` and expect them to be reflected
# in `tensor_shape[i]`, but they would not be.

TensorFlow

tf

tf.audio

tf.autograph

tf.bitwise

tf.compat

tf.config

tf.data

tf.debugging

tf.distribute

tf.dtypes

tf.errors

tf.estimator

tf.experimental

tf.feature_column

tf.graph_util

tf.image

tf.initializers

tf.io

tf.keras

tf.linalg

tf.lite

tf.lookup

tf.losses

tf.math

tf.metrics

tf.nest

tf.nn

tf.optimizers

tf.quantization

tf.queue

tf.ragged

tf.random

tf.raw_ops

tf.saved_model

tf.sets

tf.signal

tf.sparse

tf.strings

tf.summary

tf.sysconfig

tf.test

tf.tpu

tf.train

tf.version

tf.xla

tf.compat / v1 / v1.enable_v2_behavior

Examples: