def create_cartpole_swarm():
swarm = Swarm(
model=lambda x: DiscreteUniform(env=x),
walkers=Walkers,
env=lambda: DiscreteEnv(ClassicControl("CartPole-v0")),
reward_limit=121,
n_walkers=150,
max_epochs=300,
reward_scale=2,
)
return swarm
def create_cartpole_swarm():
swarm = Swarm(
model=lambda x: DiscreteUniform(env=x),
walkers=Walkers,
env=lambda: DiscreteEnv(ClassicControl()),
n_walkers=20,
max_iters=200,
prune_tree=True,
reward_scale=2,
)
return swarm
def create_atari_swarm():
env = AtariEnvironment(name="MsPacman-ram-v0", )
dt = GaussianDt(min_dt=10, max_dt=100, loc_dt=5, scale_dt=2)
swarm = Swarm(
model=lambda x: DiscreteUniform(env=x, critic=dt),
env=lambda: DiscreteEnv(env),
n_walkers=6,
max_epochs=10,
reward_scale=2,
reward_limit=1,
)
return swarm
def create_atari_swarm():
env = AtariEnvironment(name="MsPacman-ram-v0",
clone_seeds=True,
autoreset=True)
dt = GaussianDt(min_dt=3, max_dt=100, loc_dt=5, scale_dt=2)
swarm = Swarm(
model=lambda x: DiscreteUniform(env=x, critic=dt),
walkers=Walkers,
env=lambda: DiscreteEnv(env),
n_walkers=67,
max_epochs=500,
reward_scale=2,
reward_limit=751,
)
return swarm
def create_atari_swarm():
env = ParallelEnvironment(
env_class=AtariEnvironment,
name="MsPacman-ram-v0",
clone_seeds=True,
autoreset=True,
blocking=False,
)
dt = GaussianDt(min_dt=3, max_dt=100, loc_dt=5, scale_dt=2)
swarm = Swarm(
model=lambda x: DiscreteUniform(env=x, critic=dt),
walkers=Walkers,
env=lambda: DiscreteEnv(env),
n_walkers=67,
max_iters=20,
prune_tree=True,
reward_scale=2,
)
return swarm
def mathy_swarm(config: SwarmConfig, env_callable=None) -> Swarm:
if env_callable is None:
env_callable = lambda: FragileMathyEnv(
name="mathy_v0", repeat_problem=config.single_problem)
if config.use_mp:
env_callable = ParallelEnv(env_callable=env_callable)
tree_callable = None
if config.history:
tree_callable = lambda: HistoryTree(prune=True,
names=config.history_names)
swarm = Swarm(
model=lambda env: DiscreteMasked(env=env),
env=env_callable,
tree=tree_callable,
reward_limit=EnvRewards.WIN,
n_walkers=config.n_walkers,
max_epochs=config.max_iters,
reward_scale=1,
distance_scale=3,
distance_function=mathy_dist,
show_pbar=False,
)
return swarm