mujoco-py 0.5.7 vs 1.50.0.1

bench_new.py

import mujoco_py
import timeit


T = 1000
NSIM = 100


def test():
    model = mujoco_py.load_model_from_path(
        "/Users/jonas/code/openai/pymj/xmls/claw.xml")
    pool = mujoco_py.MjSimPool([mujoco_py.MjSim(model)
                                for _ in range(NSIM)])
    for _ in range(T):
        pool.step()
        for sim in pool.sims:
            sim.data.qpos
            sim.data.qvel
            sim.data.ctrl
            sim.data.active_contacts_efc_pos

for i in range(5):
    NSIM = 1 * 10**i
    print(NSIM, timeit.timeit("test()", setup="from __main__ import test", number=1))

bench_old.py

import mujoco_py
import timeit
import multiprocessing

T = 1000
NSIM = 100


def f(_):
    model = mujoco_py.MjModel(
        "/Users/jonas/code/openai/pymj/xmls/claw.xml")
    for _ in range(T):
        model.step()
        model.data.qpos
        model.data.qvel
        model.data.ctrl
        model.data.efc_pos[model.data.nefc]


def test_pool():
    pool = multiprocessing.Pool(processes=8)
    pool.map(f, range(NSIM))


def test_simple():
    for i in range(NSIM):
        model = mujoco_py.MjModel(
            "/Users/jonas/code/openai/pymj/xmls/claw.xml")
        for _ in range(T):
            model.step()
            model.data.qpos
            model.data.qvel
            model.data.ctrl
            model.data.efc_pos[model.data.nefc]

for i in range(5):
    NSIM = 1 * 10**i
    simple = timeit.timeit(
        "test_simple()", setup="from __main__ import test_simple", number=1)
    pool = timeit.timeit(
        "test_pool()", setup="from __main__ import test_pool", number=1)
    print(NSIM, simple, pool)

data.csv