giagrad.Tensor.pad#

Tensor.pad(*padding, mode: str = 'constant', **kwargs)[source]#

Pads tensor.

Padding size specified by *padding maps every argument starting from the rightmost axis. If *padding is a single int N it will be interpreted as if “before” and “after” padding for the last axis is symmetric, i.e. (N_before, N_after). If a tuple of two integers is supplied, it will be interpreted as (N_before, N_after) padding.

Padding mode has the same options as numpy.pad.

See also

numpy.pad()

Parameters:

*padding¶ (int or (int, int)) – Number of values padded to the edges of the rightmost axes.
mode¶ (str, default: 'constant') – Padding mode defined by numpy.pad.
**kwargs¶ – Optional arguments passed to numpy.pad.

Examples

>>> t = Tensor.empty(2, 2, 3, dtype=int).uniform(-5, 5)
>>> t
tensor: [[[ 0 -1  0]
          [ 0  2  0]]
...
         [[-1 -2  0]
          [-3 -1  3]]]

A single int padds the last axis with before and after symmetrically:

>>> t.pad(2)
tensor: [[[ 0  0  0 -1  0  0  0]
          [ 0  0  0  2  0  0  0]]
...
         [[ 0  0 -1 -2  0  0  0]
          [ 0  0 -3 -1  3  0  0]]] fn: ConstantPad
>>> t.pad((1, 0), 2, (1, 3))
tensor: [[[ 0  0  0  0  0  0  0]
          [ 0  0  0  0  0  0  0]
          [ 0  0  0  0  0  0  0]
          [ 0  0  0  0  0  0  0]
          [ 0  0  0  0  0  0  0]
          [ 0  0  0  0  0  0  0]]
...
         [[ 0  0  0  0  0  0  0]
          [ 0  0  0  0  0  0  0]
          [ 0  0 -1  0  0  0  0]
          [ 0  0  2  0  0  0  0]
          [ 0  0  0  0  0  0  0]
          [ 0  0  0  0  0  0  0]]
...
         [[ 0  0  0  0  0  0  0]
          [ 0  0  0  0  0  0  0]
          [ 0 -1 -2  0  0  0  0]
          [ 0 -3 -1  3  0  0  0]
          [ 0  0  0  0  0  0  0]
          [ 0  0  0  0  0  0  0]]] fn: ConstantPad