Python appending class attribute with numpy array type

Question

I want to call a class attribute and append it to a list. Here is a simple script:

class class_1():
    def __init__(self):
        self.x = np.array([0, 0])

    def update(self):
        self.x += 1
        return self.x

cl_1 = class_1()
a = []
for i in range(3):
    y = cl_1.update()
    print(y)
    a.append(y)
print(a)  
# output:
[1 1]
[2 2]
[3 3]
[array([3, 3]), array([3, 3]), array([3, 3])]

but I expect [array([1, 1]), array([2, 2]), array([3, 3])] as the final value of list a. I checked that there is no problem with python numbers:

class class_2():
    def __init__(self):
        self.x = 0

    def update(self):
        self.x += 1
        return self.x

cl_2 = class_2()
a = []
for i in range(3):
    y = cl_2.update()
    print(y)
    a.append(y)
print(a)
#output
1
2
3
[1, 2, 3]

This is a duplicate that has been asked a million times over (no reflection on you OP), but I can't seem to find a good hammer candidate. — Mad Physicist
– Mad Physicist, Commented Aug 2, 2018 at 21:08
Basically, notice that you are always appending a reference to the same array. The whole point of += is that it modifies in place, instead of creating a copy. — Mad Physicist
– Mad Physicist, Commented Aug 2, 2018 at 21:13

Mad Physicist · Accepted Answer · 2018-08-02 21:29:54Z

Unlike Python integers, numpy arrays really allow in-place modification. With a Python integer, which is immutable, x += 1 and x = x + 1 have the same result. In both cases, x is rebound to a new integer object.

When you do self.x += 1 to a numpy array, you do not change the reference that self.x points to. A new array is not allocated, but instead, every element in the internal buffer of the existing array is incremented. Note that an assignment still happens, but with the same reference as before.

To simulate the behavior of an integer in this context, explicitly write out the operation you want:

self.x = self.x + 1

In this case self.x + 1 is an entirely new array, which can then be reassigned to self.x as an integer would be.

You have two options when it comes to how you want to fix your code:

If you are OK with creating a new array every time, and discarding the previous array, change the update method as shown above. In some ways this solution is cleanest because it minimizes the effects on outside references. At the same time, you might not want to create a new copy every time.
If you want to avoid creating unnecessary copies in most cases (or at least have control over when the copies get made), use @DYZ's suggestion. Instead of a.append(y), do
```
a.append(y.copy())
```
or alternatively, make the copy immediately:
```
y = cl_2.update().copy()
```

TL;DR

One of the most common beginner pitfalls is happening here: creating a list from references to the same mutable object, and then mutating the object. All the references will end up with the last value of the object, as expected.

Collectives™ on Stack Overflow

Python appending class attribute with numpy array type

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related