hopper-v2.py

import os, sys, signal
import random
import numpy as np
from multiprocessing import Process, Queue, current_process, freeze_support
import argparse

parser = argparse.ArgumentParser()
parser.add_argument('--pgmorl', default=False, action='store_true')
parser.add_argument('--ra', default=False, action='store_true')
parser.add_argument('--pfa', default=False, action='store_true')
parser.add_argument('--moead', default=False, action='store_true')
parser.add_argument('--random', default=False, action='store_true')
parser.add_argument('--num-seeds', type=int, default=6)
parser.add_argument('--num-processes', 
                    type=int, 
                    default=1, 
                    help='number of algorithms to be run in parallel (Note: each algorithm needs 4 * num-tasks processors by default, so the total number of processors is 4 * num-tasks * num-processes.)')
parser.add_argument('--save-dir', type=str, default='./results/Hopper-v2')
args = parser.parse_args()

random.seed(2000)
commands = []
save_dir = args.save_dir


test_pgmorl = args.pgmorl
test_ra = args.ra
test_random = args.random
test_pfa = args.pfa
test_moead = args.moead

for i in range(args.num_seeds):
    seed = random.randint(0, 1000000)
    if test_pgmorl:
        cmd = 'python morl/run.py '\
            '--env-name MO-Hopper-v2 '\
            '--seed {} '\
            '--num-env-steps 8000000 '\
            '--warmup-iter 200 '\
            '--update-iter 40 '\
            '--min-weight 0.0 '\
            '--max-weight 1.0 '\
            '--delta-weight 0.2 '\
            '--eval-num 1 '\
            '--pbuffer-num 100 '\
            '--pbuffer-size 2 '\
            '--selection-method prediction-guided '\
            '--num-weight-candidates 7 '\
            '--num-tasks 6 '\
            '--sparsity 1.0 '\
            '--obj-rms '\
            '--ob-rms '\
            '--raw '\
            '--save-dir {}/pgmorl/{}/'\
                .format(seed, save_dir, i)
        commands.append(cmd)
    
    if test_ra:
        cmd = 'python morl/run.py '\
            '--env-name MO-Hopper-v2 '\
            '--seed {} '\
            '--num-env-steps 8000000 '\
            '--warmup-iter 200 '\
            '--update-iter 40 '\
            '--min-weight 0.0 '\
            '--max-weight 1.0 '\
            '--delta-weight 0.2 '\
            '--eval-num 1 '\
            '--pbuffer-num 100 '\
            '--pbuffer-size 2 '\
            '--selection-method ra '\
            '--num-tasks 6 '\
            '--obj-rms '\
            '--ob-rms '\
            '--raw '\
            '--save-dir {}/ra/{}/'\
                .format(seed, save_dir, i)
        commands.append(cmd)

    if test_random:
        cmd = 'python morl/run.py '\
            '--env-name MO-Hopper-v2 '\
            '--seed {} '\
            '--num-env-steps 8000000 '\
            '--warmup-iter 200 '\
            '--update-iter 40 '\
            '--min-weight 0.0 '\
            '--max-weight 1.0 '\
            '--delta-weight 0.2 '\
            '--eval-num 1 '\
            '--pbuffer-num 100 '\
            '--pbuffer-size 2 '\
            '--selection-method random '\
            '--num-tasks 6 '\
            '--obj-rms '\
            '--ob-rms '\
            '--raw '\
            '--save-dir {}/random/{}/'\
                .format(seed, save_dir, i)
        commands.append(cmd)

    if test_pfa:
        cmd = 'python morl/run.py '\
            '--env-name MO-Hopper-v2 '\
            '--seed {} '\
            '--num-env-steps 8000000 '\
            '--warmup-iter 200 '\
            '--update-iter 40 '\
            '--min-weight 0.0 '\
            '--max-weight 1.0 '\
            '--delta-weight 0.2 '\
            '--eval-num 1 '\
            '--pbuffer-num 100 '\
            '--pbuffer-size 2 '\
            '--selection-method pfa '\
            '--num-tasks 6 '\
            '--obj-rms '\
            '--ob-rms '\
            '--raw '\
            '--save-dir {}/pfa/{}/'\
                .format(seed, save_dir, i)
        commands.append(cmd)

    if test_moead:
        cmd = 'python morl/run.py '\
            '--env-name MO-Hopper-v2 '\
            '--seed {} '\
            '--num-env-steps 8000000 '\
            '--warmup-iter 200 '\
            '--update-iter 40 '\
            '--min-weight 0.0 '\
            '--max-weight 1.0 '\
            '--delta-weight 0.2 '\
            '--eval-num 1 '\
            '--pbuffer-num 100 '\
            '--pbuffer-size 2 '\
            '--selection-method moead '\
            '--num-tasks 6 '\
            '--obj-rms '\
            '--ob-rms '\
            '--raw '\
            '--save-dir {}/moead/{}/'\
                .format(seed, save_dir, i)
        commands.append(cmd)

def worker(input, output):
    for cmd in iter(input.get, 'STOP'):
        ret_code = os.system(cmd)
        if ret_code != 0:
            output.put('killed')
            break
    output.put('done')

# Create queues
task_queue = Queue()
done_queue = Queue()

# Submit tasks
for cmd in commands:
    task_queue.put(cmd)

# Submit stop signals
for i in range(args.num_processes):
    task_queue.put('STOP')

# Start worker processes
for i in range(args.num_processes):
    Process(target=worker, args=(task_queue, done_queue)).start()

# Get and print results
for i in range(args.num_processes):
    print(f'Process {i}', done_queue.get())