def create_step_to_best():
swarm = StepToBest(
model=lambda x: DiscreteUniform(env=x),
env=cartpole_env,
reward_limit=51,
n_walkers=100,
max_epochs=160,
step_epochs=3,
)
return swarm
def create_majority_step_swarm():
swarm = StepSwarm(
model=lambda x: DiscreteUniform(env=x),
env=lambda: ParallelEnvironment(lambda: DiscreteEnv(ClassicControl())),
reward_limit=10,
n_walkers=100,
max_epochs=20,
step_epochs=25,
)
return swarm
def create_majority_step_swarm():
swarm = StepSwarm(
model=lambda x: DiscreteUniform(env=x),
env=cartpole_env,
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: ParallelEnvironment(lambda: DiscreteEnv(ClassicControl())),
reward_limit=16,
n_walkers=100,
max_epochs=5,
step_epochs=25,
)
return swarm
def create_follow_best_step_swarm():
swarm = StepSwarm(
root_model=FollowBestModel,
model=lambda x: DiscreteUniform(env=x),
env=cartpole_env,
reward_limit=101,
n_walkers=100,
max_epochs=200,
step_epochs=10,
)
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=51,
n_walkers=100,
max_epochs=200,
step_epochs=50,
)
return swarm
def create_follow_best_step_swarm():
swarm = StepSwarm(
root_model=FollowBestModel,
model=lambda x: DiscreteUniform(env=x),
env=lambda: ParallelEnvironment(lambda: DiscreteEnv(ClassicControl())),
reward_limit=15,
n_walkers=100,
max_epochs=15,
step_epochs=25,
)
return swarm
def create_follow_best_step_swarm_after_impr():
swarm = StepSwarm(
root_model=FollowBestModel,
model=lambda x: DiscreteUniform(env=x),
env=cartpole_env,
reward_limit=101,
n_walkers=10, # 0,
max_epochs=2, # 200,
step_epochs=2, # 5,
step_after_improvement=True,
)
return swarm
def swarm_with_tree():
swarm = Swarm(
model=lambda x: DiscreteUniform(env=x),
env=lambda: DiscreteEnv(ClassicControl()),
reward_limit=200,
n_walkers=150,
max_epochs=300,
reward_scale=2,
tree=HistoryTree,
prune_tree=True,
)
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 create_follow_best_step_swarm_after_impr():
swarm = StepSwarm(
root_model=FollowBestModel,
model=lambda x: DiscreteUniform(env=x),
env=lambda: ParallelEnv(lambda: DiscreteEnv(
ClassicControl("CartPole-v0"))),
reward_limit=101,
n_walkers=100,
max_epochs=200,
step_epochs=25,
step_after_improvement=True,
)
return swarm
def create_step_to_best_after_impr():
from plangym import AtariEnvironment
from fragile.core import GaussianDt
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 = StepToBest(
model=lambda x: DiscreteUniform(env=x, critic=dt),
env=lambda: DiscreteEnv(env),
reward_limit=-100,
n_walkers=67,
max_epochs=60,
step_epochs=5,
step_after_improvement=True,
)
return swarm
