def _run_exp(self,
agent_class: AgentBase,
agent_type: str,
n_jobs: int = 1):
# Arrange
exp = AgentExperiment(
name=os.path.join(self._tmp_dir.name, 'test_exp'),
agent_class=agent_class,
agent_config=self._agent_config(agent_type=agent_type,
folder=self._tmp_dir.name),
n_reps=3,
n_jobs=n_jobs,
training_options={
"n_episodes": 4,
"max_episode_steps": 4
})
# Act
exp.run()
# Assert
self.assertEqual(3, len(exp._trained_agents))
self.assertEqual(3, len(exp.agent_scores))
for a in exp._trained_agents:
self.assertEqual(4, len(a.training_history.history))
def run_exp(n_episodes: int = 500, max_episode_steps: int = 1500):
exp = AgentExperiment(agent_class=RandomAgent,
agent_config=MountainCarConfig('random'),
n_reps=8,
n_jobs=4,
training_options={"n_episodes": n_episodes,
"max_episode_steps": max_episode_steps})
exp.run()
exp.save(fn=f"{RandomAgent.__name__}_experiment.pkl")
def run_exp(n_episodes: int = 500, max_episode_steps: int = 1000):
exp = AgentExperiment(agent_class=LinearQAgent,
agent_config=MountainCarConfig(agent_type='linear_1'),
n_reps=6,
n_jobs=6,
training_options={"n_episodes": n_episodes,
"max_episode_steps": max_episode_steps})
exp.run()
exp.save(fn=f"{LinearQAgent.__name__}_experiment.pkl")
def run_exp(n_episodes: int = 1000, max_episode_steps: int = 500):
exp = AgentExperiment(agent_class=ReinforceAgent,
agent_config=CartPoleConfig(agent_type='reinforce'),
n_reps=5,
n_jobs=6,
training_options={"n_episodes": n_episodes,
"max_episode_steps": max_episode_steps,
"update_every": 1})
exp.run()
exp.save(fn=f"{ReinforceAgent.__name__}_experiment.pkl")
def run_exp(agent_type: str,
n_episodes: int = 500,
max_episode_steps: int = 1000):
exp = AgentExperiment(name=f"{agent_type} MountainCar",
agent_class=DeepQAgent,
n_reps=8,
n_jobs=8,
training_options={
"n_episodes": n_episodes,
"max_episode_steps": max_episode_steps
})
exp.run()
exp.save(fn=f"{DeepQAgent.__name__}_{agent_type}experiment.pkl")
def run_exp(agent_type: str,
n_episodes: int = 1000,
max_episode_steps: int = 500):
exp = AgentExperiment(name=f"{agent_type} CartPole",
agent_class=ActorCriticAgent,
agent_config=CartPoleConfig(agent_type=agent_type),
n_reps=6,
n_jobs=6,
training_options={
"n_episodes": n_episodes,
"max_episode_steps": max_episode_steps
})
exp.run()
exp.save(fn=f"{ActorCriticAgent.__name__}_{agent_type}experiment.pkl")
def run_exp(agent_type: str, n_episodes: int = 400, max_episode_steps: int = 10000,
model_mode: str = 'diff'):
config = PongConfig(agent_type=agent_type, mode=model_mode)
exp = AgentExperiment(name=f"{agent_type} {model_mode} Pong",
agent_class=ActorCriticAgent,
agent_config=config,
n_reps=3,
n_jobs=3,
gpu_memory_per_agent=1024,
training_options={"n_episodes": n_episodes,
"verbose": 1,
"max_episode_steps": max_episode_steps})
exp.run()
exp.save(fn=f"{DeepQAgent.__name__}_{agent_type}experiment.pkl")
def run_exp(n_episodes: int = 1000, max_episode_steps: int = 1000):
gpu = VirtualGPU(256)
exp = AgentExperiment(
agent_class=ReinforceAgent,
agent_config=MountainCarConfig(agent_type='reinforce'),
n_reps=5,
n_jobs=1 if gpu.on else 5,
training_options={
"n_episodes": n_episodes,
"max_episode_steps": max_episode_steps,
"update_every": 1
})
exp.run()
exp.save(fn=f"{ReinforceAgent.__name__}_experiment.pkl")
def run_exp(agent_type: str,
n_episodes: int = 1000,
max_episode_steps: int = 10000):
config = SpaceInvadersConfig(agent_type=agent_type, mode='stack')
exp = AgentExperiment(name=f"{agent_type} Space Invaders",
agent_class=DeepQAgent,
agent_config=config,
n_reps=6,
n_jobs=6,
gpu_memory_per_agent=512,
training_options={
"n_episodes": n_episodes,
"verbose": 1,
"max_episode_steps": max_episode_steps
})
exp.run()
exp.save(fn=f"{DeepQAgent.__name__}_{agent_type}experiment.pkl")