def select_agent(name, domain, _max_steps, _seed, lambda_=0.1, **kwargs):
if name is None or name == "tabular-q":
return methods.tabular_q(domain, initial_learn_rate=0.9)
elif name == "ifdd-ggq":
return methods.ifdd_q(
domain,
lambda_=lambda_,
boyan_N0=1220.247254,
initial_learn_rate=0.27986823,
ifddplus=1.0 - 1e-7,
)
elif name == "ifdd-q":
return methods.ifdd_q(
domain,
threshold=0.03104970,
lambda_=lambda_,
boyan_N0=1220.247254,
initial_learn_rate=0.27986823,
ifddplus=1.0 - 1e-7,
)
elif name == "ifdd-sarsa":
return methods.ifdd_sarsa(
domain,
threshold=0.023476,
lambda_=lambda_,
boyan_N0=20.84362,
initial_learn_rate=0.3356222674,
ifddplus=1.0 - 1e-7,
)
elif name == "tile-ggq":
mat = np.matrix("""1 1 1 0 0 0;
0 1 1 1 0 0;
0 0 1 1 1 0;
0 0 0 1 1 1;
0 0 1 0 1 1;
0 0 1 1 0 1;
1 0 1 1 0 0;
1 0 1 0 1 0;
1 0 0 1 1 0;
1 0 0 0 1 1;
1 0 1 0 0 1;
1 0 0 1 0 1;
1 1 0 1 0 0;
1 1 0 0 1 0;
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))
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))
def select_agent(name, domain, max_steps, seed, **kwargs):
if name is None or name == "ifdd-q":
return methods.ifdd_q(
domain,
discretization=47,
threshold=77.0,
lambda_=0.9,
initial_learn_rate=0.05,
boyan_N0=11,
ifddplus=True,
)
elif name == "kifdd-q":
return methods.kifdd_q(
domain,
kernel_resolution=13.14,
threshold=0.21,
lambda_=0.9,
initial_learn_rate=0.07,
boyan_N0=37.0,
kernel="gaussian_kernel",
)
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))
def select_agent(name, domain, max_steps, seed):
if name is None or name == "lspi":
return methods.tabular_lspi(domain, max_steps)
elif name == "tabular-q":
return methods.tabular_q(domain)
elif name == "tabular-sarsa":
return methods.tabular_sarsa(domain)
elif name == "ifdd-q":
return methods.ifdd_q(
domain,
discretization=18,
lambda_=0.42,
boyan_N0=202,
initial_learn_rate=0.7422,
)
elif name == "kifdd-q":
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()
def select_agent(name, domain, max_steps, seed, epsilon, epsilon_min, beta,
show_reward, vi_threshold, **kwargs):
if epsilon_min is not None:
eps_decay = (epsilon - epsilon_min) / max_steps * 0.9
eps_min = epsilon_min
else:
eps_decay, eps_min = 0.0, 0.0
if name is None or name == "tabular-q":
return methods.tabular_q(
domain,
epsilon=epsilon,
epsilon_decay=eps_decay,
epsilon_min=eps_min,
initial_learn_rate=0.5,
)
elif name == "lspi":
return methods.tabular_lspi(domain, max_steps)
elif name == "nac":
return methods.tabular_nac(domain)
elif name == "ifddk-q":
return methods.ifddk_q(domain, epsilon=epsilon, initial_learn_rate=0.5)
elif name == "count-based-q":
return methods.count_based_tabular_q(
domain,
beta=beta,
epsilon=epsilon,
epsilon_decay=eps_decay,
epsilon_min=eps_min,
initial_learn_rate=0.5,
show_reward=show_reward,
)
elif name == "psrl":
return methods.tabular_psrl(
domain,
seed=seed,
show_reward=show_reward,
epsilon=epsilon,
epsilon_decay=eps_decay,
epsilon_min=eps_min,
)
elif name == "mbie-eb":
return methods.tabular_mbie_eb(
domain,
seed=seed,
show_reward=show_reward,
epsilon=epsilon,
epsilon_decay=eps_decay,
epsilon_min=eps_min,
vi_threshold=vi_threshold,
)
elif name == "opt-psrl":
return methods.tabular_opt_psrl(
domain,
n_samples=10,
seed=seed,
show_reward=show_reward,
epsilon=epsilon,
epsilon_decay=eps_decay,
epsilon_min=eps_min,
)
elif name == "gaussian-psrl":
return methods.tabular_gaussian_psrl(
domain,
seed=seed,
show_reward=show_reward,
epsilon=epsilon,
epsilon_decay=eps_decay,
epsilon_min=eps_min,
)
elif name == "ucbvi":
return methods.tabular_ucbvi(
domain,
seed=seed,
show_reward=show_reward,
epsilon=epsilon,
epsilon_decay=eps_decay,
epsilon_min=eps_min,
)
else:
raise NotImplementedError("Method {} is not supported".format(name))