总结python中的n数矩阵

Question

I have a matrix of 40*2000 a vector of of dimension 1500. I used numpy.outer to compute outer product of vector with each column of matrix as:

 np.outer(vector, matrix) 

It showed memory error so I splitted matrix as matrix[:,:10] and computed separatly. It resulted into a matrix M of 1500*(40*10). I need sum each of the matrix here i'e first 40 columns with another next 40 column and so on..... If i used loop, it is going to be slow.

     M[:,:40]+ M[:,40:80] + .....

Could someone help me to do this operation efficiently?

Answer 1

A bit of reshaping is all you need to be able to sum over the axis:

import numpy

M = numpy.arange(100).reshape(5, 20)

M[:, :4]
#>>> array([[ 0,  1,  2,  3],
#>>>        [20, 21, 22, 23],
#>>>        [40, 41, 42, 43],
#>>>        [60, 61, 62, 63],
#>>>        [80, 81, 82, 83]])

M[:, 4:8]
#>>> array([[ 4,  5,  6,  7],
#>>>        [24, 25, 26, 27],
#>>>        [44, 45, 46, 47],
#>>>        [64, 65, 66, 67],
#>>>        [84, 85, 86, 87]])

...

M[:, 16:20]
#>>> array([[16, 17, 18, 19],
#>>>        [36, 37, 38, 39],
#>>>        [56, 57, 58, 59],
#>>>        [76, 77, 78, 79],
#>>>        [96, 97, 98, 99]])

M.reshape(M.shape[0], -1, 4).sum(axis=1)
#>>> array([[ 40,  45,  50,  55],
#>>>        [140, 145, 150, 155],
#>>>        [240, 245, 250, 255],
#>>>        [340, 345, 350, 355],
#>>>        [440, 445, 450, 455]])