lfudacache

Python implementation Less Frequently Used with Dynamic Aging (LFUDA), implementing an ordered dict-like interface, providing efficient cache with reduced cache pollution.

Installation

pip install lfudacache

Usage

import lfudacache

cache = lfudacache.LFUDACache(1000)  # cache the top 1000 most frequently accessed items
cache['key1'] = 1
cache.get('key1')  # 1
list(cache.items())  # [('key1', 1)]
cache['key2'] = 2
list(cache.items())  # [('key1', 1), ('key2', 2)]
cache.get('key2')  # 2
list(cache.items())  # [('key2', 2), ('key1', 1)]
cache.peek('key1')  # 1
list(cache.items())  # [('key2', 2), ('key1', 1)]
cache.get('key3')  # None
list(cache.items())  # [('key2', 2), ('key1', 1)]
cache['key3'] = 3
list(cache.items())  # [('key3', 3), ('key2', 2), ('key1', 1)]

A function memoization decorator is also available. Please note that, unhashable arguments will raise TypeError just like python's standard functools.lru_cache.

import lfudacache

@lfudacache.memoize(1000)  # cache the top 1000 most frequent call results
def my_cached_function(param_1, param_2):
    result = my_very_expensive_logic(param_1, param_2)
    return result

Documentation

LFUDACache objects implement the MutableMapping interface, behaving like any other python dict (actually, more like an OrderedDict as it's ordered). but expiring long unused items when new ones are inserted.

An item is considered unused when cache ages above item hits.

Cache ages when asked for a missing key .

Inserting an item when cache is not enough aged will result on the item not being stored, reducing cache pollution.

LFUDACache iteration will start from most to less frequently used items.

LFUDACache

Less Frequently Used with Dynamic Aging cache, implementing the entire
:class:`collections.abc.MutableMapping' interface.

Implementation notes:
    * Most methods are self-optimizing into closures when referenced.
    * Cache behaves as as dict, all operations (except iteration or peek)
      count as cache MISS or HIT, affecting key ordering or insertion
      permeability.

How it works:
    * Every cache hit increases item HIT counter
      (except :method:`LFUDACache.peek`).
    * Every cache miss increases MISSES counter by 1, up to top HITS.
    * When full, a new cache item will only be accepted if MISSES counter
      reaches the less frequently used item HIT counter, which is evicted.
    * When a new item is cached, its HIT counter is set equal to MISSES
      itself.
    * When an existing item is updated, its HIT counter is incremented
      by 1 to at least MISSES + 1.

LFUDACache.__init__(maxsize)

:param maxsize: number of items to keep on cache
:type maxsize: int

LFUDACache.peek(key [, default])

Get value of key from cache, without updating HIT nor MISSES counters.

If key is not found, and default is not given, KeyError is raised.

:param key: cache item key
:param default: optional default parameter
:returns: cache item value
:raises KeyError: if no default is given and key is not found

Usage
-----
value = lfudacache.peek(key, 'default_value')

memoize(maxsize, fnc=None, key_fnc=make_key)

Memoization decorator using Less Frequenty Used with Dynamic Aging cache
eviction algorithm.

The LFUDACache instance is available on the decorated function, as `cache`
property.

:param maxsize: maximum cache size
:type maxsize: int
:param fnc: optional function to memoize (non-decorating behavior)
:type fnc: callable or None
:param key_fnc: optional custom cache key function, receiving argument
                list and keyword argument dict
:type key_fnc: callable
:returns: decorator if fnc is not given, wrapped function otherwise
:rtype: callable

make_key(args, kwargs)

Hash function for function arguments.

:param args: argument iterable
:type args: iterable
:param kwargs: keyword argument dict-like object
:type kwargs: dict
:returns: hash of arg and kwargs
:rtype: int

Rationale

There is a lot of LFU and LRU cache implementations for Python, but not so many LFUDA ones, and some of them were a bit too slow due to python attribute lookup overhead (thats why python's standard lru_cache implementation is purely functional).

This module addresses this performance concerns by making extensive use of python closures, while providing an object-oriented interface, via a self-optimizing implementation.

Why Less Frequently Used with Dynamic Aging (LFUDA)

LFUDA only evicts less used items when they become too old, rejecting new cache items until then.

In our implementation, an item become too old when its cache hit count lags behind the entire cache misses.

This approach prevents eviction of very used items in favor of potentially less used items.

Why not Less Recently Used (LRU)

LRU is a very simple approach to caching, evicting older unused items.

This approach is optimal when the most relevant items are accessed frequently, but this is not always the case.

The drawback is that very used items will be evicted when a lot of new items are added to cache, even if they will never be accesed again, causing cache pollution.

This problem is usually minimized increasing the cache size until potentially irrelevant cache items become only a tiny fraction of the entire cache but this is not always possible or memory efficient.

Why not Less Frequenty Used (LFU)

LFU is our base algorithm, which evicts less used items.

This approach has one major drawback when cache is full: a new saved item could cause the eviction of an item more frequently used than the new one, causing cache pollution.

To minimize this problem, two different item storage tiers can be implemented with different eviction policies. But while this method reduces said cache pollution issue, it does not fix it.