openrl.algorithms package

Submodules

openrl.algorithms.base_algorithm module

class openrl.algorithms.base_algorithm.BaseAlgorithm(cfg, init_module, agent_num: int, device=device(type='cpu'))

Bases: abc.ABC

prep_rollout()

prep_training()

abstract train(buffer, turn_on=True)

openrl.algorithms.dqn module

class openrl.algorithms.dqn.DQNAlgorithm(cfg, init_module, agent_num: int = 1, device: Union[str, torch.device] = 'cpu')

Bases: openrl.algorithms.base_algorithm.BaseAlgorithm

cal_value_loss(value_normalizer, values, value_preds_batch, return_batch, active_masks_batch)

dqn_update(sample, turn_on=True)

prepare_loss(obs_batch, next_obs_batch, rnn_states_batch, actions_batch, masks_batch, available_actions_batch, value_preds_batch, rewards_batch, active_masks_batch, turn_on)

to_single_np(input)

train(buffer, turn_on=True)

openrl.algorithms.mat module

class openrl.algorithms.mat.MATAlgorithm(cfg, init_module, agent_num: int = 1, device: Union[str, torch.device] = 'cpu')

Bases: openrl.algorithms.ppo.PPOAlgorithm

construct_loss_list(policy_loss, dist_entropy, value_loss, turn_on)

get_data_generator(buffer, advantages)

openrl.algorithms.ppo module

class openrl.algorithms.ppo.PPOAlgorithm(cfg, init_module, agent_num: int = 1, device: Union[str, torch.device] = 'cpu')

Bases: openrl.algorithms.base_algorithm.BaseAlgorithm

cal_value_loss(value_normalizer, values, value_preds_batch, return_batch, active_masks_batch)

construct_loss_list(policy_loss, dist_entropy, value_loss, turn_on)

get_data_generator(buffer, advantages)

ppo_update(sample, turn_on=True)

prepare_loss(critic_obs_batch, obs_batch, rnn_states_batch, rnn_states_critic_batch, actions_batch, masks_batch, available_actions_batch, old_action_log_probs_batch, adv_targ, value_preds_batch, return_batch, active_masks_batch, turn_on)

to_single_np(input)

train(buffer, turn_on=True)

openrl.algorithms.vdn module

class openrl.algorithms.vdn.VDNAlgorithm(cfg, init_module, agent_num: int = 1, device: Union[str, torch.device] = 'cpu')

Bases: openrl.algorithms.base_algorithm.BaseAlgorithm

cal_value_loss(value_normalizer, values, value_preds_batch, return_batch, active_masks_batch)

prepare_loss(obs_batch, next_obs_batch, rnn_states_batch, actions_batch, masks_batch, available_actions_batch, value_preds_batch, rewards_batch, active_masks_batch, turn_on)

to_single_np(input)

train(buffer, turn_on=True)

vdn_update(sample, turn_on=True)

Module contents