Problem with variable assignment in python

Question

I'm running the following code in python:

w= np.random.rand(3) 
w_old=np.zeros((3,))

while (np.linalg.norm(w - w_old)) / np.linalg.norm(w) > 1e-5:

    w_old=w

    print w
    print w_old

    w[0]-=eta*dE1i
    w[1]-=eta*dE2i
    w[2]-=eta*dE3i

    print w
    print w_old

The results for the prints are :

[ 0.22877423  0.59402658  0.16657174]

[ 0.22877423  0.59402658  0.16657174]

and

[ 0.21625852  0.5573612   0.123111  ]

[ 0.21625852  0.5573612   0.123111  ]

I'm wondering why the value of w_old has been changed? Shouldn't it be updated after going back to the beginning of the while loop? How can I fix this?

Every change to w also affects w_old. The equals assignment isn't doing what you think it does — OneCricketeer
– OneCricketeer, Commented Aug 18, 2017 at 0:18

MSeifert · Accepted Answer · 2017-08-18 00:21:57Z

2

Just using

w_old = w

doesn't copy w, using = just tells python that you want another name for whatever is stored in w. So every in-place change to w will also change w_old. There is a really nice blog post about it in case you want more details Ned Batchelder: "Facts and myths about Python names and values"

You can explicitly copy a numpy array, for example, using the copy method:

w_old = w.copy()

answered Aug 18, 2017 at 0:21

MSeifert

154k41 gold badges356 silver badges377 bronze badges

Sign up to request clarification or add additional context in comments.

Comments

chari Muvilla · Accepted Answer · 2017-08-18 00:34:59Z

1

Just to clear things up, variables are independent so:

a=5
b=a

will work just fine if you change either a or b. But lists and dictionaries cannot be copied the same way, so:

a=[1,2]
b=a
a[0]=100
print(b[0])

will give you 100 because now a and b are the same thing with two names. So as suggested before use

w_old = w.copy()

instead so w_old is a new list and not just another name for the old one.

answered Aug 18, 2017 at 0:34

chari Muvilla

697 bronze badges

1 Comment

MSeifert Over a year ago

Even they are just integers: assignment won't copy them. The difference is just that: There is no way you can change an integer in-place so you never "notice" that it's the same object (except when you use is).

Blade · Accepted Answer · 2019-07-24 14:07:50Z

Just to conclude the answers above, according to python documentation

Assignment statements in Python do not copy objects, they create bindings between a target and an object.

So, the solution is

For collections that are mutable or contain mutable items, a copy is sometimes needed so one can change one copy without changing the other.

So using copy() we can rewrite the code as

import numpy as np

w= np.random.rand(3)
w_old=np.zeros((3,))

while (np.linalg.norm(w - w_old)) / np.linalg.norm(w) > 1e-5:

    w_old=w.copy()

    print(w)
    print(w_old)

    w[0]-=0.01
    w[1]-=0.01
    w[2]-=0.01

    print(w)
    print(w_old)

and get

[0.79666571 0.82305671 0.41167625]
[0.79666571 0.82305671 0.41167625]

and

[0.78666571 0.81305671 0.40167625]
[0.79666571 0.82305671 0.41167625]

Collectives™ on Stack Overflow

Problem with variable assignment in python

3 Answers 3

Comments

1 Comment

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

3 Answers 3

Comments

1 Comment

Comments

Your Answer

Sign up or log in

Post as a guest

Related