Lazy class attribute initialization

Question

I have a class that takes only one input argument. This value will then be used to compute a number of attributes (only one in the following example). What is a pythonic way if I wanted to take the computation place only if I call the attribute. In addition, the result should be cached and attr2 must not be set from outside the class.

class LazyInit:
    def __init__(self, val):
        self.attr1 = val
        self.attr2 = self.compute_attr2()

    def compute_attr2(self):
        return self.attr1 * 2  # potentially costly computation


if __name__ == "__main__":
    obj = LazyInit(10)

    # actual computation should take place when calling the attribute
    print(obj.attr2)

Define a getter (property docs).

Olvin Roght
– Olvin Roght

2021-09-29 15:21:54 +00:00
Commented Sep 29, 2021 at 15:21 — Olvin Roght
– Olvin Roght, Commented Sep 29, 2021 at 15:21

Olvin Roght · Accepted Answer · 2021-09-30 19:10:50Z

5

As hinted above, just use @cached_property decorator.

from functools import cached_property

class LazyInit():
    ...
    @cached_property
    def attr2(self):
        return <perform expensive computation>

Olvin Roght correctly points out that this solution doesn't make attr2 read-only the way that @property does. If that is important to you, another possibility would be to write:

    ...
    @property
    def attr2(self):
        return self.__internal_attr2()

    @functools.cached
    def __internal_attr2(self):
        return <perform expensive calculation>

In any case, Python provides libraries to help you ensure that a value is only calculated once. It is better to use them than to try and write your own.

edited Sep 30, 2021 at 19:10

Olvin Roght

7,8432 gold badges19 silver badges39 bronze badges

answered Sep 29, 2021 at 15:35

Frank Yellin

11.6k1 gold badge15 silver badges27 bronze badges

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

4 Comments

Olvin Roght Over a year ago

As I said in comment under accepted answer, it could be an option. Quote from docs: "The mechanics of cached_property() are somewhat different from property()". If you need a property to be read-only or to define custom setter cached_property is not your choice.

Andi Over a year ago

This is definitely worth looking at! Thanks!

Frank Yellin Over a year ago

@OlvinRoght. You are absolutely right. I edited the solution above to deal with the case that the user wants read-only.

Mike1936 Over a year ago

I was wondering that if the class attribute is different with instance property. What the answer solved is that the time-consuming __internal_attr2 is calculated once for each instance of LazyInit but not for all different instances of LazyInit.

chepner · Accepted Answer · 2021-09-29 15:22:38Z

3

Make attr2 a property, not an instance attribute.

class LazyInit:
    def __init__(self, val):
        self.attr1 = val
        self._attr2 = None

    @property
    def attr2(self):
        if self._attr2 is None:
            self._attr2 = self.compute_attr2()
        return self._attr2

_attr2 is a private instance attribute that both indicates whether the value has been computed yet, and saves the computed value for future access.

answered Sep 29, 2021 at 15:22

chepner

538k77 gold badges594 silver badges746 bronze badges

1 Comment

Olvin Roght Over a year ago

In some cases it could be a good option to use cached_property decorator.

Collectives™ on Stack Overflow

Lazy class attribute initialization

2 Answers 2

4 Comments

1 Comment

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

4 Comments

1 Comment

Your Answer

Sign up or log in

Post as a guest

Related