Method __init__

numpy_ml/rl_models/agents.py:392–420

A Monte-Carlo learning agent trained using either first-visit Monte Carlo updates (on-policy) or incremental weighted importance sampling (off-policy). Parameters ---------- env : :class:`gym.wrappers` or :class:`gym.envs` instance The en

(self, env, off_policy=False, temporal_discount=0.9, epsilon=0.1)

class MonteCarloAgent(AgentBase):
    def __init__(self, env, off_policy=False, temporal_discount=0.9, epsilon=0.1):
        """
        A Monte-Carlo learning agent trained using either first-visit Monte
        Carlo updates (on-policy) or incremental weighted importance sampling
        (off-policy).

        Parameters
        ----------
        env : :class:`gym.wrappers` or :class:`gym.envs` instance
            The environment to run the agent on.
        off_policy : bool
            Whether to use a behavior policy separate from the target policy
            during training. If False, use the same epsilon-soft policy for
            both behavior and target policies. Default is False.
        temporal_discount : float between [0, 1]
            The discount factor used for downweighting future rewards. Smaller
            values result in greater discounting of future rewards. Default is
            0.9.
        epsilon : float between [0, 1]
            The epsilon value in the epsilon-soft policy. Larger values
            encourage greater exploration during training. Default is 0.1.
        """
        super().__init__(env)

        self.epsilon = epsilon
        self.off_policy = off_policy
        self.temporal_discount = temporal_discount

        self._init_params()

    def _init_params(self):
        E = self.env_info