boyan_N0=119,
incremental=True,
)
elif name == "rbf-q":
return methods.rbf_q(
domain,
seed,
num_rbfs=5000,
resolution=8,
initial_learn_rate=0.26,
lambda_=0.9,
boyan_N0=2120,
)
else:
raise NotImplementedError("Method {} is not supported".format(name))
if __name__ == "__main__":
run_experiment(
select_domain,
select_agent,
default_max_steps=30000,
default_num_policy_checks=10,
default_checks_per_policy=50,
other_options=[
click.Option(["--task"], type=str, default="original"),
click.Option(["--good-reward"], type=float, default=0.0),
click.Option(["--kernel"], type=str, default="gaussian_kernel"),
],
)
import click
from rlpy.domains import DeepSea
from rlpy.tools.cli import run_experiment
from fr_gridworld import select_agent
def select_domain(size, noise, **kwargs):
return DeepSea(size, noise=noise)
if __name__ == "__main__":
run_experiment(
select_domain,
select_agent,
default_max_steps=10000,
default_num_policy_checks=10,
default_checks_per_policy=50,
other_options=[
click.Option(["--size"], type=int, default=10),
click.Option(["--noise"], type=float, default=0.0),
click.Option(["--epsilon"], type=float, default=0.1),
click.Option(["--epsilon-min"], type=float, default=None),
click.Option(["--beta"], type=float, default=0.05),
click.Option(["--show-reward"], is_flag=True),
click.Option(["--vi-threshold"], type=float, default=0.001),
],
)
from fr_gridworld import select_agent
def select_domain(map_, noise, episode_cap, **kwargs):
map_ = BernoulliGridWorld.default_map(map_ + ".txt")
return BernoulliGridWorld(
map_,
noise=noise,
episode_cap=episode_cap,
)
if __name__ == "__main__":
run_experiment(
select_domain,
select_agent,
default_max_steps=10000,
default_num_policy_checks=10,
default_checks_per_policy=50,
other_options=[
click.Option(["--map", "map_"], type=str, default="5x5normal"),
click.Option(["--noise"], type=float, default=0.1),
click.Option(["--episode-cap"], type=int, default=20),
click.Option(["--epsilon"], type=float, default=0.1),
click.Option(["--epsilon-min"], type=float, default=None),
click.Option(["--beta"], type=float, default=0.05),
click.Option(["--show-reward"], is_flag=True),
click.Option(["--vi-threshold"], type=float, default=1e-6),
],
)
def select_domain(chain_size):
return ChainMDP(chain_size=chain_size)
def select_agent(name, domain, max_steps, seed, **kwargs):
if name is None or name == "lspi":
return methods.tabular_lspi(domain, max_steps)
elif name == "nac":
return methods.tabular_nac(domain)
elif name == "tabular-q":
return methods.tabular_q(domain, initial_learn_rate=0.1)
elif name == "ifddk-q":
return methods.ifddk_q(domain, initial_learn_rate=0.1)
elif name == "psrl":
return methods.tabular_psrl(domain, seed=seed)
else:
raise NotImplementedError("Method {} is not supported".format(name))
if __name__ == "__main__":
run_experiment(
select_domain,
select_agent,
default_max_steps=10000,
default_num_policy_checks=10,
default_checks_per_policy=50,
other_options=[click.Option(["--chain-size"], type=int, default=4)],
)
from rlpy.domains import BicycleRiding
from rlpy.tools.cli import run_experiment
import methods
def select_agent(name, domain, max_steps, _seed):
if name is None or name == "kifddk-q":
return methods.kifdd_q(
domain,
11.6543336229,
threshold=88044,
lambda_=0.43982644088,
initial_learn_rate=0.920244401,
boyan_N0=64502.0,
kernel="linf_triangle_kernel",
)
else:
raise NotImplementedError("Method {} is not supported".format(name))
if __name__ == "__main__":
run_experiment(
BicycleRiding(),
select_agent,
default_max_steps=150000,
default_num_policy_checks=30,
default_checks_per_policy=1,
)
return methods.kifdd_q(
domain,
8.567677,
threshold=0.0807,
lambda_=0.52738,
initial_learn_rate=0.4244,
boyan_N0=389.56,
)
elif name == "rbfs-q":
return methods.rbf_q(
domain,
seed,
num_rbfs=96,
resolution=21,
initial_learn_rate=0.6633,
lambda_=0.1953,
boyan_N0=13444.0,
)
else:
raise NotImplementedError()
if __name__ == "__main__":
run_experiment(
PuddleWorld(),
select_agent,
default_max_steps=40000,
default_num_policy_checks=20,
default_checks_per_policy=100,
)
return methods.fourier_q(domain, order=5)
elif name == "fourier-sarsa":
return methods.fourier_sarsa(domain, order=5)
elif name == "ifdd-q":
return methods.ifdd_q(domain)
elif name == "ifdd-sarsa":
return methods.ifdd_sarsa(domain)
elif name == "kifdd-q":
return methods.kifdd_q(domain)
elif name == "kifdd-sarsa":
return methods.kifdd_sarsa(domain)
elif name == "rbfs-q":
return methods.rbf_q(domain, seed=seed)
elif name == "rbfs-sarsa":
return methods.rbf_q(domain, seed=seed)
else:
raise NotImplementedError()
if __name__ == "__main__":
run_experiment(
select_domain,
select_agent,
default_max_steps=100000,
default_num_policy_checks=30,
default_checks_per_policy=1,
other_options=[
click.Option(["--cfg"], type=str, default="pinball_simple_single")
],
)
1 1 0 0 0 1;
0 1 0 1 1 0;
0 1 0 0 1 1;
0 1 0 1 0 1;
0 1 1 0 1 0;
0 1 1 0 0 1""")
return methods.tile_ggq(
domain,
mat,
lambda_=lambda_,
initial_learn_rate=0.240155681,
boyan_N0=14.44946,
)
else:
raise NotImplementedError("Method {} is not supported".format(name))
if __name__ == "__main__":
run_experiment(
select_domain,
select_agent,
default_max_steps=100000,
default_num_policy_checks=20,
default_checks_per_policy=1,
other_options=[
click.Option(["--blocks"], type=int, default=6),
click.Option(["--noise"], type=float, default=0.3),
click.Option(["--lambda", "lambda_"], type=float, default=0.0),
],
)
episode_cap=20)
def select_agent(name, domain, max_steps, seed, **kwargs):
if name is None or name == "lspi":
return methods.tabular_lspi(domain, max_steps)
elif name == "nac":
return methods.tabular_nac(domain)
elif name == "tabular-q":
return methods.tabular_q(domain, initial_learn_rate=0.11)
elif name == "ifddk-q":
return methods.ifddk_q(domain, initial_learn_rate=0.11)
elif name == "psrl":
return methods.tabular_psrl(domain, seed=seed)
else:
raise NotImplementedError("Method {} is not supported".format(name))
if __name__ == "__main__":
run_experiment(
select_domain,
select_agent,
default_max_steps=10000,
default_num_policy_checks=10,
default_checks_per_policy=50,
other_options=[
click.Option(["--map", "map_"], type=str, default="4x5"),
click.Option(["--noise"], type=float, default=0.1),
],
)
elif name == "tabular-q":
return methods.tabular_q(
domain,
lambda_=0.9,
initial_learn_rate=0.26,
boyan_N0=119,
incremental=True,
)
elif name == "rbf-q":
return methods.rbf_q(
domain,
seed,
num_rbfs=5000,
resolution=8,
initial_learn_rate=0.26,
lambda_=0.9,
boyan_N0=2120,
)
else:
raise NotImplementedError("Method {} is not supported".format(name))
if __name__ == "__main__":
run_experiment(
select_domain,
select_agent,
default_max_steps=30000,
default_num_policy_checks=10,
default_checks_per_policy=50,
)
