from mushroom_rl.algorithms.value.td import TD
from mushroom_rl.utils.table import Table
[docs]class SARSA(TD):
"""
SARSA algorithm.
"""
[docs] def __init__(self, mdp_info, policy, learning_rate):
Q = Table(mdp_info.size)
super().__init__(mdp_info, policy, Q, learning_rate)
[docs] def _update(self, state, action, reward, next_state, absorbing):
q_current = self.Q[state, action]
self.next_action = self.draw_action(next_state)
q_next = self.Q[next_state, self.next_action] if not absorbing else 0.
self.Q[state, action] = q_current + self.alpha(state, action) * (
reward + self.mdp_info.gamma * q_next - q_current)