def greedy_action(self, free_qblock_id_lists, collapsed_qttts, mark):
assert len(collapsed_qttts) > 0
states = {}
for i in range(len(collapsed_qttts)):
if free_qblock_id_lists[i] is None:
nstate = after_action_state(collapsed_qttts[i], None, mark)
states[(i, -1, -1)] = GameTree.get_state_val(nstate)
continue
n = len(free_qblock_id_lists[i])
for j in range(n - 1):
for k in range(j + 1, n):
loc1 = free_qblock_id_lists[i][j]
loc2 = free_qblock_id_lists[i][k]
nstate = after_action_state(collapsed_qttts[i],
(loc1, loc2), mark)
states[(i, loc1, loc2)] = GameTree.get_state_val(nstate)
if mark % 2 == 1:
indices = GameTree.best_states(states, min)
else:
indices = GameTree.best_states(states, max)
i, j, k = random.choice(indices)
action = (collapsed_qttts[i], (j, k))
return action
def bellman_backup(self, qttt, next_qttt, reward, mark):
"""
Bellman backup for TD learning
:param Qttt state: current state of qttt
:param Qttt next_state: next state after action is take
:param int reward: immediate reward for this round
:return: None
"""
state_value = GameTree.get_state_val(qttt.get_state())
next_state_value = GameTree.get_state_val(next_qttt.get_state())
updated_state_value = state_value + self.alpha * (
reward + gamma * next_state_value - state_value)
GameTree.set_state_value(qttt.get_state(), updated_state_value)
