Sequence padding

In this recipe, we will learn how Keras can be used for sequence padding. Padding is useful when sequences are sent in batches to the LSTM network. 

Getting ready

Import the function:

from keras.preprocessing.sequence import pad_sequences

pad_sequences is a function defined as follows:

pad_sequences(sequences, maxlen=None, dtype='int32', padding='pre', truncating='pre', value=0.0)

How to do it...

Let's look at the various padding options.

Pre-padding with default 0.0 padding

First, let's look at how to use pad_sequences with default pre-padding:

from keras.preprocessing.sequence import pad_sequences
 # define sequences
 sequences = [
 [1, 2, 3, 4],
 [5, 6, 7],
 [5]
 ]
 # pad sequence
 padded = pad_sequences(sequences)
 print(padded)

An output of the preceding print statement will show all the sequences padded to length 4. 

Post-padding

To pad 0.0 on at the end of shorter arrays, use padding='post', as shown in the following code snippet:

padded_post = pad_sequences(sequences,padding=...