arrays - Difference between a -= b and a = a - b in Python -

i have applied this solution averaging every n rows of matrix. although solution works in general had problems when applied 7x1 array. have noticed problem when using -= operator. make small example:

import numpy np  = np.array([1,2,3]) b = np.copy(a)  a[1:] -= a[:-1] b[1:] = b[1:] - b[:-1]  print print b 

which outputs:

[1 1 2] [1 1 1] 

so, in case of array a -= b produces different result a = - b. thought until these 2 ways same. difference?

how come method mentioning summing every n rows in matrix working e.g. 7x4 matrix not 7x1 array?

note: using in-place operations on numpy arrays share memory in no longer problem in version 1.13.0 onward (see details here). 2 operation produce same result. answer applies earlier versions of numpy.

---

mutating arrays while they're being used in computations can lead unexpected results!

in example in question, subtraction -= modifies second element of a , uses modified second element in operation on third element of a.

here happens a[1:] -= a[:-1] step step:

• a array data [1, 2, 3].

• we have 2 views onto data: a[1:] [2, 3], , a[:-1] [1, 2].

• the in-place subtraction -= begins. first element of a[:-1], 1, subtracted first element of a[1:]. has modified a [1, 1, 3]. have a[1:] view of data [1, 3], , a[:-1] view of data [1, 1] (the second element of array a has been changed).

• a[:-1] [1, 1] , numpy must subtract second element which 1 (not 2 anymore!) second element of a[1:]. makes a[1:] view of values [1, 2].

• a array values [1, 1, 2].

b[1:] = b[1:] - b[:-1] not have problem because b[1:] - b[:-1] creates new array first , assigns values in array b[1:]. not modify b during subtraction, views b[1:] , b[:-1] not change.

---

the general advice avoid modifying 1 view inplace if overlap. includes operators -=, *=, etc. , using out parameter in universal functions (like np.subtract , np.multiply) write 1 of arrays.