Periodic-boundary padding for tensors

[Jverduzc]

Click to expand!

Issue Type

Feature Request

Source

binary

Tensorflow Version

2.10

Custom Code

No

OS Platform and Distribution

No response

Mobile device

No response

Python version

3.9

Bazel version

No response

GCC/Compiler version

No response

CUDA/cuDNN version

No response

GPU model and memory

No response

Current Behaviour?

This issue is to request a feature for padding with "wrap", narrowing down suggestions from Issue #956 (https://github.com/tensorflow/tensorflow/issues/956).

This mode of padding is included as an option in numpy.pad()
https://docs.scipy.org/doc/numpy/reference/generated/numpy.pad.html

It would be very beneficial for applications that require periodic boundary conditions, in particular for physics-based calculations.

Standalone code to reproduce the issue

# From the original issue:

import numpy
matrix = numpy.arange(25).reshape(5,5)
paddings = [[1,1],[1,1]]
# periodic boundary conditions
padded_matrix = numpy.pad(matrix, pad_width=paddings, mode='wrap')

Relevant log output

No response

[cantonios]

Just to confirm, we're talking specifically about tf.pad?

[Jverduzc]

Yes, I want to use this operation in combination with the 'valid' padding. It would be great to have it implemented as an option in Conv2D or Conv3D if possible.

[cantonios]

Those are two separate things. Adding an option to tf.pad is relatively straightforward. Adding to all convolutions is a whole other beast which requires changing our APIs - our current convolution API doesn't have separate padding and padding_mode options, it assumes we always pad with zeros and only controls the size.

[Jverduzc]

I figured. This issue was implementing it with tf.pad() to do the padding operation separately from the convolution.

[Jverduzc]

Following up on this issue of adding tf.pad() option for 'wrap' as available in np.pad()

[cantonios]

Sounds fine to me if you'd like to add it as a new feature.

[madt2709]

Hey, I'd like to pick this up if no one is working on this.

[Jverduzc]

That would be awesome!

[github-actions]

This issue is stale because it has been open for 180 days with no activity. It will be closed if no further activity occurs. Thank you.

[mattbahr]

I'd like to try picking up on this where @madt2709 left off if that's alright.