Average :: (Visibilties) -> Visibilities

[Joshuaalbert]

We average visibilities over time and frequency into smaller data volumes.

How:

• In comes a set of visibilities of a particular structure.
• In order to average in time in a static way we need a pre-determined pattern of reduction.
• This can be a segmented_sum, see https://www.tensorflow.org/api_docs/python/tf/math/segment_sum and https://jax.readthedocs.io/en/latest/_autosummary/jax.ops.segment_sum.html
• We assign each input an output index, segment_ids. For static, num_segments must be provided which is just the length of the output visibilities.
• In this case unique_indices=False because it's an averaging op.
• indices_are_sorted=True should be ensured for performance.
• bucket_size should be used for performance stability.
• To get the segment_ids we determine for each input visibility, the output visibility it's going to. I.e. what solution interval it's going to.
• Will need to handle non-flat segment sum with a multi-index ravel, followed by reshape.
• So it's determine solution multi-index -> compute flattened index -> segment_average on flattened data -> reshape output.