method
ufunc.reduceat(a, indices, axis=0, dtype=None, out=None)
Performs a (local) reduce with specified slices over a single axis.
For i in range(len(indices))
, reduceat
computes ufunc.reduce(a[indices[i]:indices[i+1]])
, which becomes the ith generalized “row” parallel to axis
in the final result (i.e., in a 2D array, for example, if axis = 0
, it becomes the ith row, but if axis = 1
, it becomes the ith column). There are three exceptions to this:
i = len(indices)  1
(so for the last index), indices[i+1] = a.shape[axis]
.indices[i] >= indices[i + 1]
, the ith generalized “row” is simply a[indices[i]]
.indices[i] >= len(a)
or indices[i] < 0
, an error is raised.The shape of the output depends on the size of indices
, and may be larger than a
(this happens if len(indices) > a.shape[axis]
).
Parameters: 


Returns: 

A descriptive example:
If a
is 1D, the function ufunc.accumulate(a)
is the same as ufunc.reduceat(a, indices)[::2]
where indices
is range(len(array)  1)
with a zero placed in every other element: indices = zeros(2 * len(a)  1)
, indices[1::2] = range(1, len(a))
.
Don’t be fooled by this attribute’s name: reduceat(a)
is not necessarily smaller than a
.
To take the running sum of four successive values:
>>> np.add.reduceat(np.arange(8),[0,4, 1,5, 2,6, 3,7])[::2] array([ 6, 10, 14, 18])
A 2D example:
>>> x = np.linspace(0, 15, 16).reshape(4,4) >>> x array([[ 0., 1., 2., 3.], [ 4., 5., 6., 7.], [ 8., 9., 10., 11.], [12., 13., 14., 15.]])
# reduce such that the result has the following five rows: # [row1 + row2 + row3] # [row4] # [row2] # [row3] # [row1 + row2 + row3 + row4]
>>> np.add.reduceat(x, [0, 3, 1, 2, 0]) array([[12., 15., 18., 21.], [12., 13., 14., 15.], [ 4., 5., 6., 7.], [ 8., 9., 10., 11.], [24., 28., 32., 36.]])
# reduce such that result has the following two columns: # [col1 * col2 * col3, col4]
>>> np.multiply.reduceat(x, [0, 3], 1) array([[ 0., 3.], [ 120., 7.], [ 720., 11.], [2184., 15.]])
© 2005–2019 NumPy Developers
Licensed under the 3clause BSD License.
https://docs.scipy.org/doc/numpy1.17.0/reference/generated/numpy.ufunc.reduceat.html