def __init__(self,
model_class,
model=None,
env=None,
exploration=None,
gamma=0.99,
memory_size=10000,
batch_size=64,
target_update_frequency=10,
saving_dir=None):
"""
base class for lstm dqn agent
:param model_class: sub class of torch.nn.Module. class reference of the model
:param model: initial model of the policy net. could be None if loading from checkpoint
:param env: environment
:param exploration: exploration object. Must have function value(step) which returns e
:param gamma: gamma
:param memory_size: size of the memory
:param batch_size: size of the mini batch for one step update
:param target_update_frequency: the frequency for updating target net (in episode)
:param saving_dir: the directory for saving checkpoint
"""
DQNAgent.__init__(self, model_class, model, env, exploration, gamma,
memory_size, batch_size, target_update_frequency,
saving_dir)
self.memory = EpisodicReplayMemory(memory_size)
self.hidden_size = 0
if self.policy_net:
self.hidden_size = self.policy_net.hidden_size
self.hidden = None
