Python attributes and numpy arrays

Question

I have a class that stores some attributes. These attributes are numpy arrays with some floats inside. I want this attributes to be accessed when creating objects. What I don't want is them to be modified if any operation is done to a external variable that holds the value of the attribute.

This is simple to do with getters/setters or properties with other types of variables, but it seems to fail with numpy arrays.

I have wrote a simple script that tests every kind of possible solution I know. It works for integer attributes but fails with numpy arrays.

This is the test class:

class test_class:

    # Initialization
    def __init__(self, attribute1, attribute2, attribute3):

        self.attribute1 = attribute1
        self._attribute2 = attribute2
        self._attribute3 = attribute3

# Attribute 1 with getter and setter
    def get_attr1(self):
        return(self.attribute1)

    def set_attr1(self, value):
        self.attribute1 = value

    # Attribute 2 as a property with getter and setter
    def get_attr2(self):
        return(self._attribute2)

    def set_attr2(self, value):
        self._attribute2 = value

    attribute2 = property(get_attr2, set_attr2)

    # Attribute 3 as a property
    @property
    def attribute3(self):
        return(self._attribute3)

    @attribute3.setter
    def attribute3(self, value):
        self._attribute3 = value

Testting it with integers as attributes:

test = test_class(10, 100, 1000)

print test.get_attr1()
print test.attribute2
print test.attribute3

a1 = test.get_attr1()
a2 = test.attribute2
a3 = test.attribute3

a1 += 5
a2 += 50
a3 += 500

print test.get_attr1()
print test.attribute2
print test.attribute3

Output as expected, without the attributes being modified externally:

10
100
1000
10
100
1000

Testing it with numpy arrays:

import numpy as np

test = test_class(np.array([10,20,30]), np.array([100,200,300]),   np.array([1000,2000,3000]))

print test.get_attr1()
print test.attribute2
print test.attribute3

a1 = test.get_attr1()
a2 = test.attribute2
a3 = test.attribute3

a1 += 5
a2 += 50
a3 += 500

print test.get_attr1()
print test.attribute2
print test.attribute3

Output not as expected, the values have been changed:

[10 20 30]
[100 200 300]
[1000 2000 3000]
[15 25 35]
[150 250 350]
[1500 2500 3500]

So if not getters/setters nor properties work with numpy arrays, what can be done?

EDIT:

Well, I found a solution to this problem using the copy.deepcopy function. Now it works as expected.

Attribute definition:

from copy import deepcopy

class test_class:

    ...

    # Attribute 4 with getter and setter using deepcopy
    def get_attr4(self):
        return(deepcopy(self.attribute4))

    def set_attr4(self, value):
        self.attribute4 = value

Testing:

test = test_class(np.array([10,20,30]), np.array([100,200,300]), np.array([1000,2000,3000]), np.array([10000,20000,30000]))

...
print test.get_attr4()
...
a4 = test.get_attr4()
...
a4 += 5000
...
print test.get_attr4()

Result:

...
[10000 20000 30000]
...
[10000 20000 30000]

Answer 1

NumPy arrays are mutable, integers are not.

>>> a = 1
>>> id(1)
4297261152
>>> a += 1
>>> id(a)
4297261184

Note: The id changes.

As opposed to:

>>> arr = np.arange(5)
>>> d(arr)
4331954736
>>> arr += 10
>>> id(arr)
4331954736
>>> arr
array([10, 11, 12, 13, 14])

Note: The id stays the same.

It does not matter that you use a = test.get_attr1() or a = test.attribute2. The a you get is a Python object, as pretty much everything is an object you deal with in Python. Once you have a it does not matter how you created it either by assignment a = 1 or as return value from a method a = test.get_attr1() (The property just makes for a nicer syntax for a method call.). Then a is just a name for an object and you work with this object. If it is mutable like NumPy arrays the += typically changes the values inside. For immutable this simply not possible.

If you do not want to modify these objects, you can male a copy. Usually with the help of the module copy. NumPy arrays provide their own copy method:

>>> arr2  = arr.copy()
>>> arr
array([10, 11, 12, 13, 14])
>>> arr2 += 100
>>> arr2
array([110, 111, 112, 113, 114])
>>>  arr
array([10, 11, 12, 13, 14])