scipy.sparse.coo_array.

tensordot#

coo_array.tensordot(other, axes=2)[source]#

Return the tensordot product with another array along the given axes.

The tensordot differs from dot and matmul in that any axis can be chosen for each of the first and second array and the sum of the products is computed just like for matrix multiplication, only not just for the rows of the first times the columns of the second. It takes the dot product of the collection of vectors along the specified axes. Here we can even take the sum of the products along two or even more axes if desired. So, tensordot is a dot product computation applied to arrays of any dimension >= 1. It is like matmul but over arbitrary axes for each matrix.

Given two tensors, a and b, and the desired axes specified as a 2-tuple/list/array containing two sequences of axis numbers, (a_axes, b_axes), sum the products of a’s and b’s elements (components) over the axes specified by a_axes and b_axes. The axes input can be a single non-negative integer, N; if it is, then the last N dimensions of a and the first N dimensions of b are summed over.

Parameters:

a, barray_like

Tensors to “dot”.

axesint or (2,) array_like

integer_like If an int N, sum over the last N axes of a and the first N axes of b in order. The sizes of the corresponding axes must match.
(2,) array_like A 2-tuple of sequences of axes to be summed over, the first applying to a, the second to b. The sequences must be the same length. The shape of the corresponding axes must match between a and b.

Returns:

outputcoo_array: The tensor dot product of this array with other. It will be dense/sparse if other is dense/sparse.

See also

dot

Examples

>>> import numpy as np
>>> import scipy.sparse
>>> A = scipy.sparse.coo_array([[[2, 3], [0, 0]], [[0, 1], [0, 5]]])
>>> A.shape
(2, 2, 2)

Integer axes N are shorthand for (range(-N, 0), range(0, N)):

>>> A.tensordot(A, axes=1).toarray()
array([[[[ 4,  9],
         [ 0, 15]],

        [[ 0,  0],
         [ 0,  0]]],


       [[[ 0,  1],
         [ 0,  5]],

        [[ 0,  5],
         [ 0, 25]]]])
>>> A.tensordot(A, axes=2).toarray()
array([[ 4,  6],
       [ 0, 25]])
>>> A.tensordot(A, axes=3)
array(39)

Using tuple for axes:

>>> a = scipy.sparse.coo_array(np.arange(60).reshape(3,4,5))
>>> b = np.arange(24).reshape(4,3,2)
>>> c = a.tensordot(b, axes=([1,0],[0,1]))
>>> c.shape
(5, 2)
>>> c
array([[4400, 4730],
       [4532, 4874],
       [4664, 5018],
       [4796, 5162],
       [4928, 5306]])