pad_sequence#

ivy.pad_sequence(sequences, *, batch_first=False, padding_value=0)[source]#

Pad a list of variable-length sequences to the same length.

Parameters:

sequences (Iterable[Union[Array, NativeArray]]) – A list of input sequences (arrays) to pad and stack. Each sequence should have shape (seq_len, feature_size), where seq_len can vary between sequences.
batch_first (bool, default: False) – If True, the output tensor will have shape (batch_size, max_len, feature_size), where batch_size is the number of sequences and max_len is the length of the longest sequence. If False, the output tensor will have shape (max_len, batch_size, feature_size). Default is False.
padding_value (Union[Iterable[Tuple[Number]], Number], default: 0) – The value to use for padding shorter sequences to the maximum length. This value is broadcasted to all padded areas of the sequences. Default is 0.

Return type:

Array

Returns:

ret – Padded tensor with shape determined by batch_first. The length of each sequence is padded to match the length of the longest sequence in sequences.

Examples

With ivy.Array input:

>>> seq1 = ivy.array([[1, 2], [3, 4]])
>>> seq2 = ivy.array([[5, 6]])
>>> y = ivy.pad_sequence([seq1, seq2], batch_first=True, padding_value=-1)
>>> print(y)
ivy.array([[[ 1,  2],
            [ 3,  4]],
           [[ 5,  6],
            [-1, -1]]])

>>> y = ivy.pad_sequence([seq1, seq2], batch_first=False, padding_value=0)
>>> print(y)
ivy.array([[[1, 2],
            [5, 6]],
           [[3, 4],
            [0, 0]]])

With ivy.NativeArray input:

>>> seq1 = ivy.native_array([[1, 2], [3, 4]])
>>> seq2 = ivy.native_array([[5, 6]])
>>> y = ivy.pad_sequence([seq1, seq2], batch_first=True)
>>> print(y)
ivy.array([[[1, 2],
            [3, 4]],
           [[5, 6],
            [0, 0]]])