Caching Function Results in Python: Examples using Joblib and Functools

Caching Function Results in Python: Examples using Joblib and Functools

Last updated:
Table of Contents

All examples use Python 3.5+

Decorators in Python are functions that take other functions as parameters 1

See all examples here: caching-stuff-notebook

In-memory caching

Use @functools.lru_cache decorator.

This is better suited for functions that use small objects as parameters (primitive values like strings, integers, etc) and return small objects.

import functools
import time

# maxsize=4 means that you will only
# cache up to 4 different results
@functools.lru_cache(maxsize=4)
def myfunc(x):

    # using time.sleep to simulate an operation
    # that takes a long time
    time.sleep(2)

    # then return the result
    return x

# this takes 2 seconds
myfunc(1)
# >>> 1

# the results for myfunc(1) are now cached:
# this returns instantaneously
myfunc(1)
# >>> 1

myfunc(2)
# >>> 2

myfunc(3)
# >>> 3

myfunc(4)
# >>> 4

# At this point, the value for myfunc(5) is added to
# the cache, and the value for myfunc(1) is removed.
myfunc(5)
# >>> 5

Since our cache only caches up to 4 results, calling myfunc(5) causes myfunc(1) to be removed from the cache. The results for myfunc(1) are removed because it's the least-recently used (LRU) result.

Disk caching

Use @memory.cache decorator, found in the joblib library.

This is better suited for functions that take large objects as parameters and return large objects too.

The actual results are cached as pickle files, in directory joblib/, under the location you passed as location parameter.

from joblib import Memory
import time


memory = Memory(location="/path/to/cache/directory",verbose=0)

# same example as above
@memory.cache
def myfunc(x):

    # using time.sleep to simulate an operation
    # that takes a long time
    time.sleep(2)

    # then return the result
    return x

1: These are also called higher-order functions in functional programming languages.


References

  • Joblib: the memory class

    • you can pass other options to the Memory class, such as the maximum size to be used for caching, etc.