How to link instance attributes to a numpy array?

Question

I have a list of Python objects, representing several classes. The classes obviously differ, but nevertheless have a couple of common attributes (with different values for each object). For example: ``` class Super1: def __init__(self, value1, value2): self.value1 = value1 self.value2 = value2 #Lot's of other stuff class Sub1(Super1): def __init__(self, value1, value2, value3): Super1.__init__(self, value1, value2) self.value3 = value3 #Lot's of stuff class Sub2(Super1): def __init__(self, value1, value2, value4): Super1.__init__(self, value1, value2) self.value4 = value4 #Lot's of stuff objects = [Sub1(1.,2.,3.), Sub1(213.,2.,23.), Sub2(23.,10.,0.2), Sub1(3.,2.,12.)] ``` Now, for both convenience and performance, I would need to have to a NumPy array of all these values. I know I can read them like this: ``` np.array([objects[ii].value1 for ii in range(4)]) ``` But I also need to change the values, in both the array form as well as individually within instance methods. Is it possible…

Accepted Answer

Numpy arrays are "a contiguous one-dimensional segment of computer memory", so there really is no way you can create a numpy array that is made up of chunks of memory here and there.

The only possibility is to go the other way around, first create the array, then asign single element slices of that array to your objects, e.g.

```
class Super1(object):
    def __init__(self, value1):
        self._value1 = value1

@property
    def value1(self):
        return self._value1[0]

@value1.setter
    def value1(self, value):
        self._value1[0] = value
```

And now:

```
>>> a = np.arange(4)
>>> obj = [Super1(a[j:j+1]) for j in xrange(len(a))]
>>> obj[0].value1
0
>>> obj[0].value1 = 5
>>> a
array([5, 1, 2, 3])
>>> obj[2].value1
2
>>> a[2] = 8
>>> obj[2].value1
8
```

How to link instance attributes to a numpy array?

Solution

Problem