def fastgames(seed: Optional[int] = None) -> Iterator[Experiment]:
"""Optimization of policies for games, namely direct policy search."""
funcs: List[ExperimentFunction] = []
funcs += [game.Game("war")]
funcs += [game.Game("batawaf")]
funcs += [game.Game("flip")]
funcs += [game.Game("guesswho")]
funcs += [game.Game("bigguesswho")]
seedg = create_seed_generator(seed)
algos = [
"NaiveTBPSA", "ScrHammersleySearch", "PSO", "OnePlusOne", "CMA",
"TwoPointsDE", "QrDE", "LhsDE", "Zero", "StupidRandom", "RandomSearch",
"HaltonSearch", "RandomScaleRandomSearch", "MiniDE", "SplitOptimizer5",
"NGO", "Shiva", "DiagonalCMA"
]
for budget in [1600, 3200, 6400, 12800]:
for num_workers in [1, 10, 100]:
if num_workers < budget:
for algo in algos:
for fu in funcs:
xp = Experiment(fu,
algo,
budget,
num_workers=num_workers,
seed=next(seedg))
if not xp.is_incoherent:
yield xp
def realworld(seed: Optional[int] = None) -> Iterator[Experiment]:
"""Realworld optimization. This experiment contains:
- a subset of MLDA (excluding the perceptron: 10 functions rescaled or not.
- ARCoating https://arxiv.org/abs/1904.02907: 1 function.
- The 007 game: 1 function, noisy.
- PowerSystem: a power system simulation problem.
- STSP: a simple TSP problem.
MLDA stuff, except the Perceptron.
"""
funcs: List[Union[ExperimentFunction, rl.agents.TorchAgentFunction]] = [
_mlda.Clustering.from_mlda(name, num, rescale) for name, num in [("Ruspini", 5), ("German towns", 10)] for rescale in [True, False]
]
funcs += [
_mlda.SammonMapping.from_mlda("Virus", rescale=False),
_mlda.SammonMapping.from_mlda("Virus", rescale=True),
_mlda.SammonMapping.from_mlda("Employees"),
]
funcs += [_mlda.Landscape(transform) for transform in [None, "square", "gaussian"]]
# Adding ARCoating.
funcs += [ARCoating()]
funcs += [PowerSystem(), PowerSystem(13)]
funcs += [STSP(), STSP(500)]
funcs += [game.Game("war")]
funcs += [game.Game("batawaf")]
funcs += [game.Game("flip")]
funcs += [game.Game("guesswho")]
funcs += [game.Game("bigguesswho")]
# 007 with 100 repetitions, both mono and multi architectures.
base_env = rl.envs.DoubleOSeven(verbose=False)
random_agent = rl.agents.Agent007(base_env)
modules = {'mono': rl.agents.Perceptron, 'multi': rl.agents.DenseNet}
agents = {a: rl.agents.TorchAgent.from_module_maker(base_env, m, deterministic=False) for a, m in modules.items()}
env = base_env.with_agent(player_0=random_agent).as_single_agent()
runner = rl.EnvironmentRunner(env.copy(), num_repetitions=100, max_step=50)
for archi in ["mono", "multi"]:
func = rl.agents.TorchAgentFunction(agents[archi], runner, reward_postprocessing=lambda x: 1 - x)
funcs += [func]
seedg = create_seed_generator(seed)
algos = ["NaiveTBPSA", "LargeScrHammersleySearch", "ScrHammersleySearch", "PSO", "OnePlusOne",
"NGO", "Shiva", "DiagonalCMA", "CMA", "TwoPointsDE", "QrDE", "LhsDE", "Zero", "StupidRandom", "RandomSearch", "HaltonSearch",
"RandomScaleRandomSearch", "MiniDE"]
for budget in [25, 50, 100, 200, 400, 800, 1600, 3200, 6400, 12800]:
for num_workers in [1, 10, 100]:
if num_workers < budget:
for algo in algos:
for fu in funcs:
xp = Experiment(fu, algo, budget, num_workers=num_workers, seed=next(seedg))
if not xp.is_incoherent:
yield xp
def sequential_fastgames(seed: tp.Optional[int] = None) -> tp.Iterator[Experiment]:
"""Optimization of policies for games, namely direct policy search."""
funcs = [game.Game(name) for name in ["war", "batawaf", "flip", "guesswho", "bigguesswho"]]
seedg = create_seed_generator(seed)
algos = ["NaiveTBPSA", "ScrHammersleySearch", "PSO", "OnePlusOne",
"ProgOptimizer3", "ProgOptimizer5", "ProgOptimizer9", "ProgOptimizer13",
"ProgDOptimizer3", "ProgDOptimizer5", "ProgDOptimizer9", "ProgDOptimizer13",
"CMA", "TwoPointsDE", "QrDE", "LhsDE", "Zero", "StupidRandom", "RandomSearch", "HaltonSearch",
"RandomScaleRandomSearch", "MiniDE", "SplitOptimizer5", "NGO", "Shiva", "DiagonalCMA",
"OptimisticNoisyOnePlusOne", "OptimisticDiscreteOnePlusOne"]
for budget in [12800, 25600, 51200, 102400]:
for num_workers in [1]:
if num_workers < budget:
for algo in algos:
for fu in funcs:
xp = Experiment(fu, algo, budget, num_workers=num_workers, seed=next(seedg))
if not xp.is_incoherent:
yield xp
