How To Sweep Many Hyperparameter Sets In Parallel In Python?

Note that I have to sweep through more argument sets than available CPUs, so I'm not sure if Python will automatically schedule the use of the CPUs depending on their availability

Solution 1:

The function in multiprocessing.Pool.map expects one argument. One way to adapt your code is to write a small wrapper function that takes env, alg, and seed as one argument, separates them, and passes them to run.

Another option is to use multiprocessing.Pool.starmap, which allows multiple arguments to be passed to the function.

import random
import multiprocessing
import itertools

envs = [
    "AntBulletEnv-v0",
    "HalfCheetahBulletEnv-vo",
    "HopperBulletEnv-v0",
    "ReacherBulletEnv-v0",
    "Walker2DBulletEnv-v0",
    "InvertedDoublePendulumBulletEnv-v0",
]
algs = ["PPO", "A2C"]
seeds = [
    random.randint(0, 200),
    random.randint(200, 400),
    random.randint(400, 600),
    random.randint(600, 800),
    random.randint(800, 1000),
]

args = list(itertools.product(*[envs, algs, seeds]))

num_cpus = multiprocessing.cpu_count()

# sample implementation or `run`defrun(env, alg, seed):
    # do stuffreturn random.randint(0, 200)

defwrapper(env_alg_seed):
    env, alg, seed = env_alg_seed
    return run(env=env, alg=alg, seed=seed)

# use a wrapperwith multiprocessing.Pool(num_cpus) as processing_pool:
    # accumulate results in a dictionary
    results = processing_pool.map(wrapper, args)

# use starmap and call `run` directlywith multiprocessing.Pool(num_cpus) as processing_pool:
    results = processing_pool.starmap(run, args)