When the lengths of each vector are not the same, it is a method to make them the same length by adding 0 or cutting.
For example ...
>>> from keras.preprocessing import sequence
>>> import numpy as np
>>> data = [np.array([[1,2,3],[4,5,6]]),
... np.array([[1,2,3],[4,5,6],[7,8,9],[10,11,12],[13,14,15]])]
>>> data
[array([[1, 2, 3],
[4, 5, 6]]), array([[ 1, 2, 3],
[ 4, 5, 6],
[ 7, 8, 9],
[10, 11, 12],
[13, 14, 15]])]
>>> #Align the length to 4.
>>> data = sequence.pad_sequences(data, maxlen=4,padding="post", truncating="post")
>>> data
array([[[ 1, 2, 3],
[ 4, 5, 6],
[ 0, 0, 0],
[ 0, 0, 0]],
[[ 1, 2, 3],
[ 4, 5, 6],
[ 7, 8, 9],
[10, 11, 12]]], dtype=int32)
If you do not specify dtype, the value of int32 is returned by default.
Then, ** if the original data has floating point, it will be forcibly converted to int32 **.
For example, 0.1 becomes 0.
↓ When dtype is not specified
>>> from keras.preprocessing import sequence
>>> import numpy as np
>>> #data mixed with float
>>> data = [np.array([[0.1,0.2,0.3],[0.4,0.5,0.6]]),
... np.array([[1,2,3],[4,5,6],[7,8,9],[10,11,12],[13,14,15]])]
>>> data
[array([[0.1, 0.2, 0.3],
[0.4, 0.5, 0.6]]), array([[ 1, 2, 3],
[ 4, 5, 6],
[ 7, 8, 9],
[10, 11, 12],
[13, 14, 15]])]
>>> #Align the length to 4.
>>> data = sequence.pad_sequences(data, maxlen=4,padding="post", truncating="post")
>>> #The value that was float is automatically cast to int32 and becomes 0
>>> data
array([[[ 0, 0, 0],
[ 0, 0, 0],
[ 0, 0, 0],
[ 0, 0, 0]],
[[ 1, 2, 3],
[ 4, 5, 6],
[ 7, 8, 9],
[10, 11, 12]]], dtype=int32)
When using pad_sequences, specify dtype.
sequence.pad_sequences(data, maxlen=4, padding="post",
truncating="post", dtype=float32)
Recommended Posts