memoclass

A utility package for memoizing functions and class methods

Introduction

Memoization is a common technique used primarily as an optimizer, allowing the caching of the return values of expensive functions. When a memoized function is called with a given set of arguments for the first time its return value is cached, and that value is then returned on all subsequent calls.

While many memoization packages exist, as well as instructions to write your own simple decorator, this package provides enhanced utility for memoizing methods on classes. Especially, this allows for a couple of ways of automatically clearing the caches related to a particular instance, for example, when changing a member variable would change the result of those functions. This package also uses inspect.getcallargs to correctly treat default argument values, where possible.

Preliminary sphinx documentation can be found at https://jon-burr.github.io/memoclass/memoclass.html

The project is hosted on github at https://github.com/Jon-Burr/memoclass

Installation

pip install memoclass

Overview of core components

memoclass.memoize.memofunc

Memoizes a single free function. The returned object is a MemoFunc object, which has functions that allow you to temporarily disable the caching, or clear it entirely.

>>> from memoclass.memoize import memofunc
>>>
>>> @memofunc
>>> def build_list(x):
>>>     return [1, 2, 3, x]
>>>
>>> a = build_list(5)
>>> b = build_list(5)
>>> a is b
True
>>> build_list.clear_cache()
>>> c = build_list(5)
>>> a is c
False

The MemoFunc class also has some extra properties that can be set while decorating

>>> from memoclass.memoize import memofunc
>>> import copy
>>>
>>> @memofunc(on_return=copy.copy)
>>> def build_list(x):
>>>     return [1, 2, 3, x]
>>>
>>> a = build_list(5)
>>> b = build_list(5)
>>> a is b
False

memoclass.memoize.memomethod

Memoizes a class method. Methods bound to different instances have independent caches, so the cache on one object can be cleared without clearing it for all other objects.

>>> from memoclass.memoize import memomethod
>>>
>>> class ListBuilder(object):
>>>     @memomethod
>>>     def __call__(self, x):
>>>         return [1, 2, 3, x]
>>>
>>> x = ListBuilder()
>>> y = ListBuilder()
>>> a = x()
>>> b = y()
>>> a is b
False
>>> x.__call__.clear_cache()
>>> c = y()
>>> b is c # Clearing x's cache has not touched y's
True

memoclass.memoclass.MemoClass

Base class meant to make interacting with memoized methods easier. It can enabled, disable and clear all memomethods attached to an instance (note that which methods exist is calculated at the class level, so any added onto an instance will not be seen).

By default, setting any attribute will reset the object's caches, unless that attribute has been provided to the mutable_attrs argument of MemoClass.__init__. This behaviour can be disabled by setting mutable_attrs=None. Additionally, any function can have the memoclass.memoclass.mutates decorator applied to it, which will then reset the caches whenever it is called.

>>> from memoclass.memoize import memomethod
>>> from memoclass.memoclass import MemoClass
>>>
>>> class PartialSum(MemoClass):
>>>
>>>     def __init__(self, stored):
>>>         super().__init__()
>>>         self.stored = stored
>>>
>>>     @memomethod
>>>     def __call__(self, other):
>>>         return self.stored + other
>>>
>>> a = PartialSum(5)
>>> a(3)
8
>>> a.stored = 3 # Triggers a cache reset
>>> a(3)
6

A MemoClass can be locked which means that all caches are enabled and calling a function marked mutates or setting a non-mutable attribute results in a ValueError. When the class is then unlocked again, if the caches were previously disabled, they will be disabled and cleared. This means it is possible to create a class whose methods are only temporarily memoized. This might be useful if a class has expensive methods to calculate that rely on a global state. Note that by default, a memomethod declared on a MemoClass will lock its caller while it is called.