tf.keras.experimental.SequenceFeatures

View source on GitHub

A layer for sequence input.

Inherits From: Layer, Module

tf.keras.experimental.SequenceFeatures(
    feature_columns, trainable=True, name=None, **kwargs
)

All feature_columns must be sequence dense columns with the same sequence_length. The output of this method can be fed into sequence networks, such as RNN.

The output of this method is a 3D Tensor of shape [batch_size, T, D]. T is the maximum sequence length for this batch, which could differ from batch to batch.

If multiple feature_columns are given with Di num_elements each, their outputs are concatenated. So, the final Tensor has shape [batch_size, T, D0 + D1 + ... + Dn].

Example:

import tensorflow as tf

# Behavior of some cells or feature columns may depend on whether we are in
# training or inference mode, e.g. applying dropout.
training = True
rating = tf.feature_column.sequence_numeric_column('rating')
watches = tf.feature_column.sequence_categorical_column_with_identity(
    'watches', num_buckets=1000)
watches_embedding = tf.feature_column.embedding_column(watches,
                                            dimension=10)
columns = [rating, watches_embedding]

features = {
 'rating': tf.sparse.from_dense([[1.0,1.1, 0, 0, 0],
                                             [2.0,2.1,2.2, 2.3, 2.5]]),
 'watches': tf.sparse.from_dense([[2, 85, 0, 0, 0],[33,78, 2, 73, 1]])
}

sequence_input_layer = tf.keras.experimental.SequenceFeatures(columns)
sequence_input, sequence_length = sequence_input_layer(
   features, training=training)
sequence_length_mask = tf.sequence_mask(sequence_length)
hidden_size = 32
rnn_cell = tf.keras.layers.SimpleRNNCell(hidden_size)
rnn_layer = tf.keras.layers.RNN(rnn_cell)
outputs, state = rnn_layer(sequence_input, mask=sequence_length_mask)