Module: tf.tpu.experimental

TensorFlow 2 version

Public API for tf.tpu.experimental namespace.

Classes

class AdagradParameters: Optimization parameters for Adagrad with TPU embeddings.

class AdamParameters: Optimization parameters for Adam with TPU embeddings.

class DeviceAssignment: Mapping from logical cores in a computation to the physical TPU topology.

class StochasticGradientDescentParameters: Optimization parameters for stochastic gradient descent for TPU embeddings.

Functions

embedding_column(...): TPU version of tf.compat.v1.feature_column.embedding_column.

initialize_tpu_system(...): Initialize the TPU devices.

shared_embedding_columns(...): TPU version of tf.compat.v1.feature_column.shared_embedding_columns.