Cache a python function with lru_cache

edit ✏️

The easiest way to speed up any code is to run less code. A common technique to reduce the amount of repative work is to implement a cache such that the next time you need the same work done, you don't need to recompute anything you can simply retrieve it from a cache.

lru_cache

The easiest and most common to setup in python is a builtin functools.lru_cache.

from functools import lru_cache

@lru_cache
def get_cars():
    print('pulling cars data')
    return pd.read_csv("https://waylonwalker.com/cars.csv", storage_options = {'User-Agent': 'Mozilla/5.0'})

when to use lru_cache

Any time you have a function where you expect the same results each time a function is called with the same inputs, you can use lru_cache.

when same args, * kwargs always return the same value

lru_cache only works for one python process. If you are running multiple subprocesses, or running the same script over and over, lru_cache will not work.

lru_cache only caches in a single python process

max_size

lru_cache can take an optional parameter maxsize to set the size of your cache. By default its set to 128 , if you want to store more or less items in your cache you can adjust this value.

The get_cars example is a bit of a unique one. As anthonywritescode points out this implementation is behaving like a singleton, and we can optimize the size of the cache by allocating exactly how many items we will ever have in it by setting its value to 1.

from functools import lru_cache

@lru_cache(maxsize=1)
def get_cars():
    print('pulling cars data')
    return pd.read_csv("https://waylonwalker.com/cars.csv", storage_options = {'User-Agent': 'Mozilla/5.0'})

My example stretches the rule a little bit

The example above does a web request. As a Data Engineer I often write scripts that run for a short time then stop. I do not expect the output of this function to change during the runtime of this job, and if it did I may actually want them to match anyways.

web request do change their output

If I were building webapps, or some sort of process that was running for a long time. Something that starts and waits for work, this may not be a good application of lru_cache. If this process is running for days or months my assumption that the request does not change is no longer valid.

There's also a typed kwarg for lru_cache

This one is new to me but you can cache not only on the value, but the type of the value being passed into your function.

(from the docstring) If typed is True, arguments of different types will be cached separately. For example, f(3.0) and f(3) will be treated as distinct calls with distinct results.