def main(args, unknown_args):
args, config = parse_args_uargs(args, unknown_args, dump_config=True)
os.makedirs(args.logdir, exist_ok=True)
save_config(config=config, logdir=args.logdir)
if args.expdir is not None:
modules = prepare_modules( # noqa: F841
expdir=args.expdir, dump_dir=args.logdir)
algorithm = Registry.get_fn("algorithm", args.algorithm)
algorithm_kwargs = algorithm.prepare_for_trainer(config)
redis_server = StrictRedis(port=config.get("redis", {}).get("port", 12000))
redis_prefix = config.get("redis", {}).get("prefix", "")
pprint(config["trainer"])
pprint(algorithm_kwargs)
trainer = Trainer(**config["trainer"],
**algorithm_kwargs,
logdir=args.logdir,
redis_server=redis_server,
redis_prefix=redis_prefix)
pprint(trainer)
def on_exit():
for p in trainer.get_processes():
p.terminate()
atexit.register(on_exit)
trainer.run()
def main(args, unknown_args):
args, config = parse_args_uargs(args, unknown_args, dump_config=True)
os.makedirs(args.logdir, exist_ok=True)
save_config(config=config, logdir=args.logdir)
if args.expdir is not None:
modules = prepare_modules( # noqa: F841
expdir=args.expdir, dump_dir=args.logdir)
algorithm = Registry.get_fn("algorithm", args.algorithm)
if args.environment is not None:
# @TODO: remove this hack
# come on, just refactor whole rl
environment_fn = Registry.get_fn("environment", args.environment)
env = environment_fn(**config["env"])
config["shared"]["observation_size"] = env.observation_shape[0]
config["shared"]["action_size"] = env.action_shape[0]
del env
algorithm_kwargs = algorithm.prepare_for_trainer(config)
redis_server = StrictRedis(port=config.get("redis", {}).get("port", 12000))
redis_prefix = config.get("redis", {}).get("prefix", "")
pprint(config["trainer"])
pprint(algorithm_kwargs)
trainer = Trainer(**config["trainer"],
**algorithm_kwargs,
logdir=args.logdir,
redis_server=redis_server,
redis_prefix=redis_prefix)
pprint(trainer)
def on_exit():
for p in trainer.get_processes():
p.terminate()
atexit.register(on_exit)
trainer.run()
def main(args, unknown_args):
args, config = parse_args_uargs(args, unknown_args)
os.makedirs(args.logdir, exist_ok=True)
save_config(config=config, logdir=args.logdir)
if args.expdir is not None:
modules = prepare_modules( # noqa: F841
expdir=args.expdir, dump_dir=args.logdir)
algorithm = Registry.get_fn("algorithm", args.algorithm)
environment = Registry.get_fn("environment", args.environment)
processes = []
sampler_id = 0
def on_exit():
for p in processes:
p.terminate()
atexit.register(on_exit)
params = dict(logdir=args.logdir,
algorithm=algorithm,
environment=environment,
config=config,
resume=args.resume,
redis=args.redis)
if args.debug:
params_ = dict(
vis=False,
infer=False,
action_noise=0.5,
param_noise=0.5,
action_noise_prob=args.action_noise_prob,
param_noise_prob=args.param_noise_prob,
id=sampler_id,
)
run_sampler(**params, **params_)
for i in range(args.vis):
params_ = dict(
vis=False,
infer=False,
action_noise_prob=0,
param_noise_prob=0,
id=sampler_id,
)
p = mp.Process(target=run_sampler, kwargs=dict(**params, **params_))
p.start()
processes.append(p)
sampler_id += 1
for i in range(args.infer):
params_ = dict(
vis=False,
infer=True,
action_noise_prob=0,
param_noise_prob=0,
id=sampler_id,
)
p = mp.Process(target=run_sampler, kwargs=dict(**params, **params_))
p.start()
processes.append(p)
sampler_id += 1
for i in range(1, args.train + 1):
action_noise = args.max_action_noise * i / args.train \
if args.max_action_noise is not None \
else None
param_noise = args.max_param_noise * i / args.train \
if args.max_param_noise is not None \
else None
params_ = dict(
vis=False,
infer=False,
action_noise=action_noise,
param_noise=param_noise,
action_noise_prob=args.action_noise_prob,
param_noise_prob=args.param_noise_prob,
id=sampler_id,
)
p = mp.Process(target=run_sampler, kwargs=dict(**params, **params_))
p.start()
processes.append(p)
sampler_id += 1
for p in processes:
p.join()
