def _updateQ(self, s_id, a_id, reward, discount_factor, similarity,
best_future_utility):
if s_id not in self.q:
self.q[s_id] = {}
if a_id not in self.q[s_id]:
self.q[s_id][a_id] = 0
# this is a typical q-learning except for "similarity," which is factored
# in to help deal with how big the state space is
q_value = (
1 -
self.learning_rate) * self.q[s_id][a_id] + self.learning_rate * (
reward + similarity * discount_factor * best_future_utility)
self.q[s_id][a_id] = q_value
Database.updateQ(s_id, a_id, q_value)
