I am trying to build a number of Machine Learning models on a single dataset. The output of all models is then to be used in further steps. I would like the training of the models, to happen simultaneously to save time and manual labour.

I am completely new to asynchronous processing, and that has manifested itself in my code below not working. I get the error:

sys:1 RuntimeWarning: coroutine 'level1models' was never awaited

This appears to be a fairly common issue when await isn't used, but wherever I place this command the error persists, and answers I find online do not seem to address functions that return values. To provide a reproducible example I have altered my code while keeping the structure identical to the original.

from time import sleep

nrs_list = [1, 2, 3, 4, 5]


def subtract(n):
    return n - 1


async def subtract_nrs(nrs):
    # Train selected ML models
    numbers = {nr: subtract(nr) for nr in nrs}
    sleep(50)

    # Loop to check if all models are trained
    while True:
        print([i for i in numbers.values()])
        if [i for i in numbers.values()] != [None for _ in range(len(numbers))]:
            break
        sleep(5)
    return numbers

r = subtract_nrs(nrs_list)
print(r)

<coroutine object subtract_nrs at 0x000002A413A4C4C0>
sys:1: RuntimeWarning: coroutine 'subtract_nrs' was never awaited

CodePudding user response：

Anytime you create a coroutine (here when you call subtract_nrs) but don't await it, asyncio will emit the warning you received [0]. The wait you avoid this is by awaiting the coroutine, either via

await subtract_nrs(nrs_list)

or by using asyncio.gather [1], which itself must be awaited

await asyncio.gather(subtract_nrs(nrs_list)

Note that here there's no value in using asyncio.gather. That would only come if you needed to wait for multiple coroutines at once.

Based on your code, you seem to be using subtract_nrs as the entry point to your program. await can't be used outside of an async def, so you need another way to wait for it. For that, you'll typically want to use asyncio.run [2]. This will handle creating, running, and closing the event loop along with waiting for your coroutine.

asyncio.run(subtract_nrs(nrs_list))

Now that we've covered all that, asyncio won't actually help you achieve your goal of simultaneous execution. asyncio never does things simultaneously; it does things concurrently [3]. While one task is waiting for I/O to complete, asyncio's event loop allows another to execute. While you've stated that this is a simplified version of your actual code, the code you've provided isn't I/O-bound; it's CPU-bound. This kind of code doesn't work well with asyncio. To use your CPU-bound code and achieve something more akin to simultaneous execution, you should use processes. not asyncio. The best way to do this is with ProcessPoolExecutor from concurrent.futures [4].