def _build_for_dqn(self) -> Dict[str, Any]:
"""This isn't tuned."""
return {
'name':
os.path.join(self.folder, 'DeepQAgent'),
'env_spec':
self.env_spec,
'model_architecture':
DenseNN(observation_shape=(2, ),
n_actions=3,
opt='adam',
learning_rate=0.001,
unit_scale=12,
dueling=False),
'gamma':
0.99,
'final_reward':
650,
'replay_buffer_samples':
32,
'eps':
EpsilonGreedy(eps_initial=0.1,
decay=0.002,
eps_min=0.002,
actions_pool=list(range(3))),
'replay_buffer':
ContinuousBuffer(buffer_size=200)
}
def _build_for_dueling_dqn(self) -> Dict[str, Any]:
config_dict = self._build_for_dqn()
config_dict.update({'name': os.path.join(self.folder, 'DuelingDQN'),
'model_architecture': DenseNN(observation_shape=(4,), n_actions=2, opt='adam',
learning_rate=0.0001, unit_scale=16, dueling=True)})
return config_dict
def _build_for_reinforce(self) -> Dict[str, Any]:
return {'name': os.path.join(self.folder, 'REINFORCEAgent'),
'env_spec': self.env_spec,
'model_architecture': DenseNN(observation_shape=(4,), n_actions=2, opt='adam', unit_scale=16,
output_activation='softmax', learning_rate=0.001, dueling=False),
'final_reward': -2,
'gamma': 0.99,
'alpha': 0.00001}
def _build_with_dense_model(dueling: bool = False) -> Dict[str, Any]:
return {"model_architecture": DenseNN(observation_shape=(115,), n_actions=19, dueling=dueling,
output_activation=None, opt='adam', learning_rate=0.000105)}