tf.keras.utils.pad_sequences()

Action:

Normalize length

This function converts a list (of length num_samples) of sequences (lists of integers) to a shape of (num_samples, num_timesteps) 2D Numpy array.num_timesteps is the maxlen parameter (if provided), or the longest sequence in the listlength.

shorter than num_timestepsThe sequences are filled with values ​​until they are num_timesteps.

Longer than num_timestepsThe sequence will be truncated to fit the desired length.

The position where padding or truncation occurs is determined by parameter padding and truncation, respectively.Prefilling or removing values ​​from the beginning of the sequence is the default.

parameter

tf.keras.utils.pad_sequences(sequences, # sequence lengthmaxlen=None, # optional Int, maximum length of all sequences.If not provided, the sequence will be padded to the length of the longest single sequence.dtype='int32', # optional, defaults to "int32".Type of output sequence.To pad a sequence with variable-length strings, you can use object.padding='pre', # string, "pre" or "post" (optional, defaults to "pre"): padding before or after each sequence.truncating='pre', # string, "pre" or "post" (optional, defaults to "pre"): Remove values ​​from sequences greater than maxlen, whether at the beginning or end of the sequence.value=0.0 # float or string, fill value.(Optional, defaults to 0.))

Return value

Numpy array with shape (len(sequences), maxlen)

Example

import tensorflow as tf # import tensorflowsequence = [[1], [2, 3], [4, 5, 6]] # input sequencetf.keras.preprocessing.sequence.pad_sequences(sequence) # length normalization

array([[0, 0, 1],[0, 2, 3],[4, 5, 6]])

import tensorflow as tf # import tensorflowsequence = [[1], [2, 3], [4, 5, 6]] # input sequencetf.keras.preprocessing.sequence.pad_sequences(sequence, padding='post') # length normalization

array([[1, 0, 0],[2, 3, 0],[4, 5, 6]])

Main reference: tf.keras.utils.pad_sequences | TensorFlow Core v2.9.1 (google.cn)