debroadcast

named_arrays.debroadcast(array, axes=None)

Remove redundant axes introduced by broadcasting.

This is the approximate inverse of broadcast_to(): it collapses every axis along which array is constant (all slices equal), returning the smallest array which broadcasts back to the original. In particular, na.debroadcast(na.broadcast_to(a, shape)) recovers an array equal to a whenever a is not itself constant along any of its own axes.

An axis is only removed if every element of array is equal along that axis, so any axis containing a NaN (which compares unequal to itself) is never removed.

Parameters:

• array (ArrayT) – The array to remove redundant axes from.

• axes (None | str | Sequence[str]) – The axes to consider removing. If None (the default), every axis of array is considered. Axes not present in array are ignored.

Return type:

ArrayT

See also

broadcast_to()

Broadcast an array to a given shape.

Examples

Broadcast a 1D array to two dimensions, then recover the original.

import named_arrays as na

a = na.random.uniform(0, 1, dict(x=3))

b = na.broadcast_to(a, dict(x=3, y=4))

# ``b`` is constant along ``y``, so that axis is removed
c = na.debroadcast(b)

c.shape

{'x': 3}

# the recovered array equals the original
bool((c == a).all())

True

A subset of axes can be considered by passing axes explicitly.

na.debroadcast(b, axes="x").shape

{'x': 3, 'y': 4}