Search for a command to run...
Learn the easy way to pass multiple params to map(), multiprocessing pool.map, threadpool, processpool executor.map
AI Software Engineer based in London, UK with 6+ years of professional experience developing and releasing software in different programming languages - Hobbyist Technical Writer - Interested in Software Testing, Best Practices, Scalability, Machine Learning/AI, and Python.
Interesting and beginner friendly too! Thanks.
Check if two lists are equal, which elements match, get the difference between two lists, compare lists of dictionaries, list of strings and more!
A guide on how to check if two strings are equal, or similar. Learn how to find the difference between two strings. Make complex comparisons and more!
Learn how to fix this new Pylint rule that raises: 'formatting a regular string which could be a f-string (consider-using-f-string)' error
A mega tutorial with dozens of examples on how to use the pathlib module in Python 3
Learn how to ignore autouse fixtures for one test or more tests
The map() function is everywhere in Python. It's a built in, it's part of the concurrent.futures.Executor, and also multiprocessing.Pool; but... it's limited!
It's limited because you cannot pass multiple arguments to it. However, what if I told you that there's some easy ways you can do that?
In this post, I’m going to show what you can do to map a function that expects multiple arguments. By the end of this article, you'll know:
itertools.starmap()functools.partial to "freeze" and pass multiple arguments to mappool.mapLet's go!
A map() is a function that expects one or more iterables and a function as arguments.
For each item in these iterables, map applies the function passed as argument. The result is an iterator where each element is produced by the function you provided as argument. If you pass multiple iterables, you must pass a function that accepts that many arguments.
Let’s imagine that you have a function called sum_four that takes 4 arguments and returns their sum.
>>> def sum_four(a, b, c, d):
return a + b + c + d
Let’s also suppose that you are solving a very specific problem that requires the first 3 arguments to be fixed. In this problem, you want to compare how the function behaves when you vary only the last parameter.
>>> a, b, c = 1, 2, 3
>>> sum_four(a=a, b=b, c=c, d=1)
7
>>> sum_four(a=a, b=b, c=c, d=2)
8
>>> sum_four(a=a, b=b, c=c, d=3)
9
>>> sum_four(a=a, b=b, c=c, d=4)
10
Now, say that you want to use map, because you like functional programming, or maybe because you come from a language that encourages this paradigm.
Since only d varies, we could store all potential values for d we want to test in a list like this all_d_values = [1, 2, 3, 4].
The issue is, given a function and a list of single elements, if you want to pass that list to a map function and it takes only one element, what can you do?
itertools.starmap()The first solution is to not adopt the map function but use itertools.starmap instead. This function will take a function as arguments and an iterable of tuples. Then, starmap will iterate over each tuple t and call the function by unpacking the arguments, like this for t in tuples: function(*t).
To make things more clear, consider the following example.
>>> import itertools
>>> all_d_values = [1, 2, 3, 4]
>>> items = ((a, b, c, d) for d in all_d_values)
>>> list(items)
[(1, 2, 3, 1), (1, 2, 3, 2), (1, 2, 3, 3), (1, 2, 3, 4)]
>>> list(itertools.starmap(sum_four, items))
[7, 8, 9, 10]
As you can see, there’s a lot of repetition, which may inevitably consume a lot of memory if the list is big. To improve that I made items as a generator, this way we only hold in memory the element we’ll be processing.
functools.partial to “Freeze” the ArgumentsThe second solution is to use currying and create a new partial function. According to the docs, partial() will "freeze" some portion of a function’s arguments and/or keywords resulting in a new function with a simplified signature.
>>> import functools
>>> partial_sum_four = functools.partial(sum_four, a, b, c)
>>> partial_sum_four(3)
9
>>> list(map(partial_sum_four, all_d_values))
[7, 8, 9, 10]
The third alternative is to use the itertools.repeat().
This function produces an iterator that returns object over and over again. It will run indefinitely if you don’t specify the times argument.
If we take a closer look at map()'s signature, it accepts a function and multiple iterables, map(function, iterable, ...).
According to its description,
If additional iterable arguments are passed, function must take that many arguments and is applied to the items from all iterables in parallel. With multiple iterables, the iterator stops when the shortest iterable is exhausted.
Bingo! We can make a, b and c infitnite iterables by using itertools.repeat(). As soon as all_d_values is exhausted, which is the shortest iterable, map() will stop.
>>> import itertools
>>> list(map(sum_four, itertools.repeat(a), itertools.repeat(b), itertools.repeat(c), all_d_values))
[7, 8, 9, 10]
To put it another way, using repeat() is roughly equivalent to:
>>> list(map(sum_four, [1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], all_d_values))
[7, 8, 9, 10]
You don't need to worry too much about memory as repeat produces the elements on the go. In fact, it returns a repeatobject, not list [ref] .
Pool.mapThis problem is very similar to using the regular map(). The only difference is that we need to pass multiple arguments to the multiprocessing's pool map.
Suppose that we want to speed up our code and run sum_four in parallel using processes.
The good news is, you can use the solutions above, with one exception: Pool.map only accepts one iterable. This means we cannot use repeat() here. Let's see the alternatives.
pool.starmapThe Pool class from multiprocessing module implements a starmap function that works the same way as its counterpart from the itertools module.
>>> from multiprocessing import Pool
>>> import itertools
>>> def sum_four(a, b, c, d):
return a + b + c + d
>>> a, b, c = 1, 2, 3
>>> all_d_values = [1, 2, 3, 4]
>>> items = [(a, b, c, d) for d in all_d_values]
>>> items
[(1, 2, 3, 1), (1, 2, 3, 2), (1, 2, 3, 3), (1, 2, 3, 4)]
>>> with Pool(processes=4) as pool:
res = pool.starmap(sum_four, items)
>>> res
[7, 8, 9, 10]
partial()As alternative, we can also rely on the good partial function.
>>> import functools
>>> partial_sum_four = functools.partial(sum_four, a, b, c)
>>> with Pool(processes=4) as pool:
res = pool.map(partial_sum_four, all_d_values)
>>> res
[7, 8, 9, 10]
Executor.map?The concurrent.futures module provides a high-level interface called Executor to run callables asynchronously.
There are two different implementations available, a ThreadPoolExecutor and a ProcessPoolExecutor.
Contrary to multiprocessing.Pool, a Executor does not have a startmap() function. However, its map() implementation supports multiple iterables, which allow us to use repeat(). Another difference is that Executor.map returns a generator, not a list.
partial() With a ProcessPoolExecutor (or ThreadPoolExecutor)By "freezing" the arguments using partial we use the map method from ProcessPoolExecutor like a regular map function. Since they both share the same interface, you can do the same interchangeably with a ThreadPoolExecutor
>>> from concurrent.futures import ProcessPoolExecutor
>>> import functools
>>> def sum_four(a, b, c, d):
return a + b + c + d
>>> a, b, c = 1, 2, 3
>>> all_d_values = [1, 2, 3, 4]
>>> partial_sum_four = functools.partial(sum_four, a, b, c)
>>> with ProcessPoolExecutor(max_workers=4) as pool:
res = list(pool.map(partial_sum_four, all_d_values))
>>> res
[7, 8, 9, 10]
repeat()Again, we can just use itertools.repeat to get the job done like the previous solutions.
>>> from concurrent.futures import ProcessPoolExecutor
>>> from itertools import repeat
>>> def sum_four(a, b, c, d):
return a + b + c + d
>>> a, b, c = 1, 2, 3
>>> all_d_values = [1, 2, 3, 4]
>>> with ProcessPoolExecutor(max_workers=4) as pool:
res = list(pool.map(sum_four, repeat(a), repeat(b), repeat(c), all_d_values))
>>> res
[7, 8, 9, 10]
That’s it for today, folks! I hope you’ve learned something different and useful. The map() function makes Python feel like a functional programming language. map() is available not only as a built-in function but also as methods in the multiprocessing and concurrent.futures module. In this article, I showed what I do to map functions that take several arguments.
Other posts you may like: