def start(self, verbose=False):
'''
対戦の開始
'''
state = State()
current_player_mark = 1
result = None
while (True):
#print("="*30)
current_player = self.players[current_player_mark]
if verbose:
print("%s" % (state.to_array()))
print state.output()
print("-" * 5)
# プレイヤーの行動の選択
index = current_player.select_index(state)
#print("%s selected %i" % (self.players[current_player_mark].mark.to_string(), index))
state = state.set(index, self.players[current_player_mark].mark)
# この時点のstateで報酬が発生する場合はここでrewardを判定して学習できる
# tic_tac_toeでは勝負が決まるまで報酬は0
current_player.learn(0)
if state.is_win(self.players[current_player_mark].mark):
result = self.players[current_player_mark].mark
# 勝者の報酬
current_player.learn(1, True)
# 敗者の報酬
self.players[result.opponent().to_int()].learn(-1, True)
if verbose:
print("%s" % (state.to_array()))
print("-" * 5)
state.output()
print("-" * 5)
print("%s win!!!" %
(self.players[current_player_mark].mark.to_string()))
break
elif state.is_draw():
result = Mark(Empty())
for player in self.players.itervalues():
player.learn(0, True)
if verbose:
state.output()
print("draw.")
break
current_player_mark = self.players[
current_player_mark].mark.opponent().to_int()
arguments:
状態
報酬
事後状態から行動を行った後の状態
'''
if next_state is None:
next_state_value = 0.0
else:
#状態価値テーブルから価値を取得
next_state_value = self.value[next_state]
# 行動価値の更新式
self.value[state] += self.step_size * (reward + next_state_value - self.value[state])
#print("value[state] %f", self.value[state])
if __name__ == '__main__':
value = Value()
state = State()
#print value.get_value(state)
#print value.get_max_action(state, Mark(Maru()))
new_state = state.set(3, Mark(Maru()))
new_state1 = state.set(4, Mark(Maru()))
new_state2 = new_state.set(2, Mark(Batsu()))
value.update(state, 10, new_state)
value.update(new_state, 10, new_state2)
value.update(new_state, 10, new_state2)
value.update(new_state1, 10, new_state2)
value.update(new_state1, 100, new_state2)
#value.update(state, -3, new_state2)
print value.get_max_action(state, Mark(Maru()))
状態
報酬
事後状態から行動を行った後の状態
'''
if next_state is None:
next_state_value = 0.0
else:
#状態価値テーブルから価値を取得
next_state_value = self.value[next_state]
# 行動価値の更新式
self.value[state] += self.step_size * (reward + next_state_value -
self.value[state])
#print("value[state] %f", self.value[state])
if __name__ == '__main__':
value = Value()
state = State()
#print value.get_value(state)
#print value.get_max_action(state, Mark(Maru()))
new_state = state.set(3, Mark(Maru()))
new_state1 = state.set(4, Mark(Maru()))
new_state2 = new_state.set(2, Mark(Batsu()))
value.update(state, 10, new_state)
value.update(new_state, 10, new_state2)
value.update(new_state, 10, new_state2)
value.update(new_state1, 10, new_state2)
value.update(new_state1, 100, new_state2)
#value.update(state, -3, new_state2)
print value.get_max_action(state, Mark(Maru()))
def start(self, verbose=False):
'''
対戦の開始
'''
state = State()
current_player_mark = 1
result = None
while(True):
#print("="*30)
current_player = self.players[current_player_mark]
if verbose:
print("%s" % (state.to_array()))
print state.output()
print("-"*5)
# プレイヤーの行動の選択
index = current_player.select_index(state)
#print("%s selected %i" % (self.players[current_player_mark].mark.to_string(), index))
state = state.set(index, self.players[current_player_mark].mark)
# この時点のstateで報酬が発生する場合はここでrewardを判定して学習できる
# tic_tac_toeでは勝負が決まるまで報酬は0
current_player.learn(0)
if state.is_win(self.players[current_player_mark].mark):
result = self.players[current_player_mark].mark
# 勝者の報酬
current_player.learn(1, True)
# 敗者の報酬
self.players[result.opponent().to_int()].learn(-1, True)
if verbose:
print("%s" % (state.to_array()))
print("-"*5)
state.output()
print("-"*5)
print("%s win!!!" % (self.players[current_player_mark].mark.to_string()))
break
elif state.is_draw():
result = Mark(Empty())
for player in self.players.itervalues():
player.learn(0, True)
if verbose:
state.output()
print("draw.")
break
current_player_mark = self.players[current_player_mark].mark.opponent().to_int()
