def classic_control_env():
env = ClassicControl()
env.reset()
env = DiscreteEnv(env)
params = {"actions": {"dtype": np.int64}, "dt": {"dtype": np.float32}}
states = States(state_dict=params, batch_size=N_WALKERS)
states.update(actions=np.ones(N_WALKERS), dt=np.ones(N_WALKERS))
return env, states
def create_majority_step_swarm():
swarm = StepSwarm(
model=lambda x: DiscreteUniform(env=x),
env=lambda: ParallelEnv(lambda: DiscreteEnv(
ClassicControl(name="CartPole-v0"))),
reward_limit=10,
n_walkers=100,
max_epochs=20,
step_epochs=25,
)
return swarm
def create_step_to_best():
swarm = StepToBest(
model=lambda x: DiscreteUniform(env=x),
env=lambda: ParallelEnv(lambda: DiscreteEnv(
ClassicControl("CartPole-v0"))),
reward_limit=16,
n_walkers=100,
max_epochs=5,
step_epochs=25,
)
return swarm
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_follow_best_step_swarm():
swarm = StepSwarm(
root_model=FollowBestModel,
model=lambda x: DiscreteUniform(env=x),
env=lambda: ParallelEnv(lambda: DiscreteEnv(
ClassicControl("CartPole-v0"))),
reward_limit=15,
n_walkers=100,
max_epochs=15,
step_epochs=25,
)
return swarm
def create_cartpole_swarm():
from fragile.core import DiscreteEnv, DiscreteUniform, Swarm
from plangym import ClassicControl
swarm = Swarm(
model=lambda x: DiscreteUniform(env=x),
env=lambda: DiscreteEnv(ClassicControl("CartPole-v0")),
reward_limit=51,
n_walkers=50,
max_epochs=100,
reward_scale=2,
)
return swarm
def cartpole_env():
if judo.Backend.can_use_cuda():
return RayEnv(lambda: DiscreteEnv(ClassicControl(name="CartPole-v0")), n_workers=2)
else:
return ParallelEnv(lambda: DiscreteEnv(ClassicControl(name="CartPole-v0")))
def classic_control_env():
env = ClassicControl()
env.reset()
env = DiscreteEnv(env)
return env