def run_benchmark(iterations, env_params):
env = Herding(**env_params)
dogs_count = env.env_data.config.dogs_count
env.reset()
start_time = time.time()
for i in range(iterations):
env.step(np.random.rand(dogs_count, 3).astype(np.float32))
end_time = time.time()
env.close()
result = end_time - start_time
print(str(int(iterations / result)) + " iterations / s")
def run_benchmarks(params: BenchmarkParams = BenchmarkParams(),
log_dir: str = None,
show_plot=False):
dogs_counts = list(
range(params.dogs_min, params.dogs_max, params.dogs_step))
sheep_counts = list(
range(params.sheep_min, params.sheep_max, params.sheep_step))
actions = _get_actions(dogs_counts)
results = np.zeros((params.dogs_max, params.sheep_max))
for dogs_count in dogs_counts:
for sheep_count in sheep_counts:
env = Herding(dogs_count=dogs_count, sheep_count=sheep_count)
env.reset()
start_time = time.time()
for i in range(params.benchmark_iterations):
env.step(np.random.rand(dogs_count, 3))
end_time = time.time()
result = end_time - start_time
results[dogs_count, sheep_count] = result
if log_dir is not None:
_write_log(results, log_dir)
if show_plot:
_show_plot(dogs_counts, sheep_counts, results)
"policy_mapping_fn": lambda agent_id: "policy",
},
"horizon": 2000,
"num_gpus": 1,
"explore": False
#"replay_sequence_length": 5,
#"num_workers": 4,
#"num_envs_per_worker": 2,
}
ray.init(local_mode=True)
checkpoint_number = 790
env = Herding({"sheep_count": 3
#"agents_layout": "simple"
})
agent = ppo.PPOTrainer(config=config, env=HerdingEnvWrapper)
agent.restore(
rf"C:\Users\Mateusz\ray_results\Herding\Herding\checkpoint_{checkpoint_number}\checkpoint-{checkpoint_number}"
)
while True:
episode_reward = 0
done = False
steps = 0
obs = env.reset()
while (not done) and (steps != 300):
action = agent.compute_action(obs[0], policy_id="policy")
obs, reward, done, info = env.step(np.array([[2, action]]))
env.render()
def play(my_env=None):
env = my_env or Herding()
manual_steering = ManualSteering(env)
manual_steering.run_env()
episode_reward = 0
self.env.close()
@staticmethod
def print_debug(*args):
print('\r', end='', flush=True)
for arg in args:
print(str(arg) + '\t', end='', flush=True)
def play(my_env=None):
env = my_env or Herding()
manual_steering = ManualSteering(env)
manual_steering.run_env()
if __name__ == '__main__':
args = {'sheep_count': 80, 'dogs_count': 1}
if len(sys.argv) > 1:
config = get_default_configuration()
parser = argparse.ArgumentParser()
for name, value in config.items():
parser.add_argument('--' + name, type=type(value))
args_all = vars(parser.parse_args())
for name, value in args_all.items():
if args_all[name] is not None:
args[name] = value
play(Herding(args))
def play(my_env=None):
env = my_env if my_env is not None else Herding()
manual_steering = ManualSteering(env)
manual_steering.run_env()