Use NumPy function on arrays containing both numbers and objects

Question

I have a number-like class where I have implemented the sqrt, exp, etc. methods so that NumPy functions will broadcast on them when they are in ndarrays.

class A:
    def sqrt(self):
        return 1.414

This work perfectly in an array of these:

import numpy as np
print(np.sqrt([A(), A()]))  # [1.414 1.414]

Obviously, sqrt also works with pure numbers:

print(np.sqrt([4, 9]))  # [2. 3.]

However, this does not work when numbers and objects are mixed:

print(np.sqrt([4, A()]))

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-38-0c4201337685> in <module>()
----> 1 print(np.sqrt([4, A()]))

AttributeError: 'int' object has no attribute 'sqrt'

This happens because the dtype of the heterogenous array is object and numpy functions broadcast by calling a method of the same name on each object, but numbers do not have methods with these names.

How do I work around this?

Warning: using numpy objects with custom objects is likely to be worse in terms of performance than a pure list. — cs95
– cs95, Commented Feb 15, 2019 at 16:42
The delegation to sqrt method is caused by the object dtype. So all elements of the array have to have method. An object dtype of pure numbers would have the same problem. — hpaulj
– hpaulj, Commented Feb 15, 2019 at 17:27

Ben.T · Accepted Answer · 2019-02-15 20:48:39Z

1

Not sure about the efficiency, but as a work around, you can use boolean indexing created with map and isinstance and then apply to both slices the same operation, changing the type of the element that are not of class A to be able to use the numpy method.

ar = np.array([4, A(), A(), 9.])
ar_t = np.array(list(map(lambda x: isinstance(x, A), ar)))

ar[~ar_t] = np.sqrt(ar[~ar_t].astype(float)) 
ar[ar_t] = np.sqrt(ar[ar_t])

print(ar)
# array([2.0, 1.414, 1.414, 3.0], dtype=object)

Note: in the astype, I used float, but not sure it will be good for your requirement

answered Feb 15, 2019 at 20:48

Ben.T

29.7k6 gold badges39 silver badges57 bronze badges

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

Collectives™ on Stack Overflow

Use NumPy function on arrays containing both numbers and objects

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