def measure_perf(self, s1, s2):
old_epsilon1, old_is_learning1, old_stand_for1 = s1.epsilon, s1.is_learning, s1.stand_for
# old_epsilon2, old_is_learning2, old_stand_for2 = s2.epsilon, s2.is_learning, s2.stand_for
old_is_learning2, old_stand_for2 = s2.is_learning, s2.stand_for
s1.epsilon, s1.is_learning, s1.stand_for = 0, False, Board.STONE_BLACK
# s2.epsilon, s2.is_learning, s2.stand_for = 0, False, Board.STONE_WHITE
s2.is_learning, s2.stand_for = False, Board.STONE_WHITE
s3 = StrategyRand()
probs = [0, 0, 0, 0, 0, 0]
games = 3 # 30
for i in range(games):
# the learner s1 move first(use black)
s1.stand_for = Board.STONE_BLACK
s2.stand_for = Board.STONE_WHITE
g = Game(Board(), s1, s2)
g.step_to_end()
if g.winner == Board.STONE_BLACK:
probs[0] += 1
elif g.winner == Board.STONE_EMPTY:
probs[1] += 1
# the learner s1 move second(use white)
s1.stand_for = Board.STONE_WHITE
s2.stand_for = Board.STONE_BLACK
g = Game(Board(), s1, s2)
g.step_to_end()
if g.winner == Board.STONE_WHITE:
probs[2] += 1
elif g.winner == Board.STONE_EMPTY:
probs[3] += 1
# the learner s1 move first vs. random opponent
s1.stand_for = Board.STONE_BLACK
s3.stand_for = Board.STONE_WHITE
g = Game(Board(), s1, s3)
g.step_to_end()
if g.winner == Board.STONE_BLACK:
probs[4] += 1
# the learner s1 move second vs. random opponent
s1.stand_for = Board.STONE_WHITE
s3.stand_for = Board.STONE_BLACK
g = Game(Board(), s1, s3)
g.step_to_end()
if g.winner == Board.STONE_WHITE:
probs[5] += 1
probs = [i / games for i in probs]
print(probs)
s1.epsilon, s1.is_learning, s1.stand_for = old_epsilon1, old_is_learning1, old_stand_for1
# s2.epsilon, s2.is_learning, s2.stand_for = old_epsilon2, old_is_learning2, old_stand_for2
s2.is_learning, s2.stand_for = old_is_learning2, old_stand_for2
return probs
def measure_perf(self, s1, s2):
old_epsilon1, old_is_learning1, old_stand_for1 = s1.epsilon, s1.is_learning, s1.stand_for
# old_epsilon2, old_is_learning2, old_stand_for2 = s2.epsilon, s2.is_learning, s2.stand_for
old_is_learning2, old_stand_for2 = s2.is_learning, s2.stand_for
s1.epsilon, s1.is_learning, s1.stand_for = 0, False, Board.STONE_BLACK
# s2.epsilon, s2.is_learning, s2.stand_for = 0, False, Board.STONE_WHITE
s2.is_learning, s2.stand_for = False, Board.STONE_WHITE
s3 = StrategyRand()
probs = [0, 0, 0, 0, 0, 0]
games = 3 # 30
for i in range(games):
# the learner s1 move first(use black)
s1.stand_for = Board.STONE_BLACK
s2.stand_for = Board.STONE_WHITE
g = Game(Board(), s1, s2)
g.step_to_end()
if g.winner == Board.STONE_BLACK:
probs[0] += 1
elif g.winner == Board.STONE_EMPTY:
probs[1] += 1
# the learner s1 move second(use white)
s1.stand_for = Board.STONE_WHITE
s2.stand_for = Board.STONE_BLACK
g = Game(Board(), s1, s2)
g.step_to_end()
if g.winner == Board.STONE_WHITE:
probs[2] += 1
elif g.winner == Board.STONE_EMPTY:
probs[3] += 1
# the learner s1 move first vs. random opponent
s1.stand_for = Board.STONE_BLACK
s3.stand_for = Board.STONE_WHITE
g = Game(Board(), s1, s3)
g.step_to_end()
if g.winner == Board.STONE_BLACK:
probs[4] += 1
# the learner s1 move second vs. random opponent
s1.stand_for = Board.STONE_WHITE
s3.stand_for = Board.STONE_BLACK
g = Game(Board(), s1, s3)
g.step_to_end()
if g.winner == Board.STONE_WHITE:
probs[5] += 1
probs = [i / games for i in probs]
print(probs)
s1.epsilon, s1.is_learning, s1.stand_for = old_epsilon1, old_is_learning1, old_stand_for1
# s2.epsilon, s2.is_learning, s2.stand_for = old_epsilon2, old_is_learning2, old_stand_for2
s2.is_learning, s2.stand_for = old_is_learning2, old_stand_for2
return probs
def init_both_sides(self):
# feat = Board.BOARD_SIZE_SQ * 2 + 2
# if self.strategy_1 is None:
# s1 = StrategyTD(feat, feat * 2)
# s1.stand_for = Board.STONE_BLACK
# s1.alpha = 0.3
# s1.beta = 0.3
# s1.lambdaa = 0.05
# s1.epsilon = 0.3
# self.strategy_1 = s1
# else:
# s1 = self.strategy_1
# s1.epsilon = 0.3
if self.strategy_1 is None:
# s1 = StrategyMC()
# s1 = StrategyANN(feat, feat * 2)
file = tf.train.latest_checkpoint(RL_BRAIN_DIR)
s1 = StrategyDNN(from_file=file, part_vars=True)
# s1 = StrategyMCTS1()
self.strategy_1 = s1
else:
s1 = self.strategy_1
s1.is_learning = True
s1.stand_for = Board.STONE_BLACK
# if self.strategy_2 is None:
# s2 = StrategyTD(feat, feat * 2)
# s2.stand_for = Board.STONE_WHITE
# self.strategy_2 = s2
# else:
# s2 = self.strategy_2
# s2.is_learning = False
s2 = StrategyRand()
# s2 = StrategyMinMax()
s2.stand_for = Board.STONE_WHITE
self.strategy_2 = s2
return s1, s2
def init_both_sides(self):
feat = Board.BOARD_SIZE_SQ * 2 + 2
# if self.strategy_1 is None:
# s1 = StrategyTD(feat, feat * 2)
# s1.stand_for = Board.STONE_BLACK
# # s1.alpha = 0.3
# # s1.beta = 0.3
# s1.lambdaa = 0.05
# s1.epsilon = 0.3
# self.strategy_1 = s1
# else:
# s1 = self.strategy_1
# s1.epsilon = 0.3
if self.strategy_1 is None:
# s1 = StrategyMC()
# s1 = StrategyANN(feat, feat * 2)
s1 = StrategyDNN()
self.strategy_1 = s1
else:
s1 = self.strategy_1
s1.is_learning = True
s1.stand_for = Board.STONE_BLACK
# if self.strategy_2 is None:
# s2 = StrategyTD(feat, feat * 2)
# s2.stand_for = Board.STONE_WHITE
# self.strategy_2 = s2
# else:
# s2 = self.strategy_2
# s2.is_learning = False
s2 = StrategyRand()
# s2 = StrategyMinMax()
s2.stand_for = Board.STONE_WHITE
self.strategy_2 = s2
return s1, s2