colosseum.mdp.taxi.finite_horizon

View Source

 1from typing import Any, Dict, List
 2
 3import gin
 4
 5from colosseum.mdp import EpisodicMDP
 6from colosseum.mdp.taxi.base import TaxiMDP
 7
 8
 9@gin.configurable
10class TaxiEpisodic(EpisodicMDP, TaxiMDP):
11    @staticmethod
12    def sample_parameters(n: int, seed: int = None) -> List[Dict[str, Any]]:
13        return TaxiMDP.sample_mdp_parameters(n, True, seed)

@gin.configurable

class TaxiEpisodic(colosseum.mdp.base_finite.EpisodicMDP, colosseum.mdp.taxi.base.TaxiMDP): View Source

10@gin.configurable
11class TaxiEpisodic(EpisodicMDP, TaxiMDP):
12    @staticmethod
13    def sample_parameters(n: int, seed: int = None) -> List[Dict[str, Any]]:
14        return TaxiMDP.sample_mdp_parameters(n, True, seed)

The base class for episodic MDPs.

TaxiEpisodic(H: int = None, **kwargs) View Source

 78    def __init__(self, H: int = None, **kwargs):
 79        super(EpisodicMDP, self).__init__(**kwargs)
 80
 81        # Computing the time horizon
 82        self._input_H = H
 83        self._H = None
 84
 85        # Episodic setting specific caching variables
 86        self._reachable_states = None
 87        self._episodic_graph = dict()
 88        self._continuous_form_episodic_transition_matrix_and_rewards = None
 89        self._episodic_transition_matrix_and_rewards = None
 90        self._optimal_policy_cf = dict()
 91        self._worst_policy_cf = dict()
 92        self._optimal_value_cf = None
 93        self._worst_value_cf = None
 94        self._random_value_cf = None
 95        self._eoar = None
 96        self._woar = None
 97        self._roar = None
 98        self._random_policy_cf = None
 99        self._random_policy = None
100        self._average_optimal_episodic_reward = None
101        self._average_worst_episodic_reward = None
102        self._average_random_episodic_reward = None

Parameters

seed (int): The seed used for sampling rewards and next states.
size (int): The size of the grid.
length (int): The length of the walls.
width (int): The width of the walls.
space (int): The space between walls.
n_locations (int): The number of possible spawn locations. It must be a squared number.
optimal_mean_reward (float): If the rewards are made stochastic, this parameter controls the mean reward for the optimal trajectory. By default, it is set to 0.9.
sub_optimal_mean_reward (float): If the rewards are made stochastic, this parameter controls the mean reward for suboptimal trajectories. By default, it is set to 0.1.
default_r
successfully_delivery_r (Union[Tuple, rv_continuous]): The reward distribution for successfully delivering a passenger. It can be either passed as a tuple containing Beta parameters or as a rv_continuous object.
failure_delivery_r: The reward distribution for failing to deliver a passenger. It can be either passed as a tuple containing Beta parameters or as a rv_continuous object.
make_reward_stochastic (bool): If True, the rewards of the MDP will be stochastic. By default, it is set to False.
reward_variance_multiplier (float): A constant that can be used to increase the variance of the reward distributions without changing their means. The lower the value, the higher the variance. By default, it is set to 1.

@staticmethod

def sample_parameters(n: int, seed: int = None) -> List[Dict[str, Any]]: View Source

12    @staticmethod
13    def sample_parameters(n: int, seed: int = None) -> List[Dict[str, Any]]:
14        return TaxiMDP.sample_mdp_parameters(n, True, seed)

Returns

List[Dict[str, Any]]: n sampled parameters that can be used to construct an MDP in a reasonable amount of time.

Inherited Members

colosseum.mdp.base_finite.EpisodicMDP: is_episodic; H; random_policy_cf; random_policy; parameters; reachable_states; T_cf; R_cf; optimal_value_continuous_form; worst_value_continuous_form; random_value_continuous_form; episodic_optimal_average_reward; episodic_worst_average_reward; episodic_random_average_reward; continuous_form_episodic_transition_matrix_and_rewards; episodic_transition_matrix_and_rewards; get_optimal_policy_continuous_form; get_worst_policy_continuous_form; get_random_policy_continuous_form; get_minimal_regret_for_starting_node; get_optimal_policy_starting_value; get_worst_policy_starting_value; get_random_policy_starting_value; get_episodic_graph; get_grid_representation
colosseum.mdp.taxi.base.TaxiMDP: get_unique_symbols; does_seed_change_MDP_structure; sample_mdp_parameters; get_node_class; get_gin_parameters; n_actions
colosseum.mdp.base.BaseMDP: get_available_hardness_measures; produce_gin_file_from_mdp_parameters; get_gin_config; get_node_labels; get_node_action_labels; hash; instantiate_MDP; T; R; recurrent_nodes_set; communication_class; get_optimal_policy; get_worst_policy; get_value_functions; optimal_value_functions; worst_value_functions; random_value_functions; optimal_transition_probabilities; worst_transition_probabilities; random_transition_probabilities; optimal_markov_chain; worst_markov_chain; random_markov_chain; get_stationary_distribution; optimal_stationary_distribution; worst_stationary_distribution; random_stationary_distribution; optimal_average_rewards; worst_average_rewards; random_average_rewards; get_average_reward; optimal_average_reward; worst_average_reward; random_average_reward; transition_matrix_and_rewards; graph_layout; graph_metrics; diameter; sum_reciprocals_suboptimality_gaps; discounted_value_norm; undiscounted_value_norm; value_norm; measures_of_hardness; summary; hardness_report; get_info_class; get_transition_distributions; get_reward_distribution; sample_reward; get_measure_from_name; action_spec; observation_spec; get_observation; reset; step; random_steps; random_step; get_visitation_counts; reset_visitation_counts; get_value_node_labels
dm_env._environment.Environment: reward_spec; discount_spec; close