def run(self):
if self.opponent == "RandomAgent":
log_dir1 = self.log_dir_root + "DeepQLearningAgent_VS_RandomAgent_" + self.time
print(log_dir1)
print(
TensorboardTicTacToeRunner(
DeepQLearningAgent(9, 9),
RandomAgent(),
log_and_reset_score_history_threshold=10000,
log_dir=log_dir1).run(100000000))
elif self.opponent == "TabularQLearningAgent":
log_dir2 = self.log_dir_root + "DeepQLearningAgent_VS_TabularQLearningAgent_" + self.time
print(log_dir2)
print(
TensorboardTicTacToeRunner(
DeepQLearningAgent(9, 9),
TabularQLearningAgent(),
log_and_reset_score_history_threshold=10000,
log_dir=log_dir2).run(100000000))
elif self.opponent == "DeepQLearningAgent":
log_dir3 = self.log_dir_root + "DeepQLearningAgent_VS_DeepQLearningAgent_" + self.time
print(log_dir3)
print(
TensorboardTicTacToeRunner(
DeepQLearningAgent(9, 9),
DeepQLearningAgent(9, 9),
log_and_reset_score_history_threshold=10000,
log_dir=log_dir3).run(100000000))
elif self.opponent == "ReinforceClassicAgent":
log_dir4 = self.log_dir_root + "DeepQLearningAgent_VS_ReinforceClassicAgent_" + self.time
print(log_dir4)
print(
TensorboardTicTacToeRunner(
DeepQLearningAgent(9, 9),
ReinforceClassicAgent(9, 9),
log_and_reset_score_history_threshold=10000,
log_dir=log_dir4).run(100000000))
elif self.opponent == "ReinforceClassicWithMultipleTrajectoriesAgent":
log_dir5 = self.log_dir_root + "DeepQLearningAgent_VS_ReinforceClassicWithMultipleTrajectoriesAgent_" + self.time
print(log_dir5)
print(
TensorboardTicTacToeRunner(
DeepQLearningAgent(9, 9),
ReinforceClassicWithMultipleTrajectoriesAgent(9, 9),
log_and_reset_score_history_threshold=10000,
log_dir=log_dir5).run(100000000))
elif self.opponent == "PPOWithMultipleTrajectoriesMultiOutputsAgent":
log_dir6 = self.log_dir_root + "DeepQLearningAgent_VS_PPOWithMultipleTrajectoriesMultiOutputsAgent_" + self.time
print(log_dir6)
print(
TensorboardTicTacToeRunner(
DeepQLearningAgent(9, 9),
PPOWithMultipleTrajectoriesMultiOutputsAgent(9, 9),
log_and_reset_score_history_threshold=10000,
log_dir=log_dir6).run(100000000))
elif self.opponent == "MOISMCTSWithRandomRolloutsAgent":
log_dir7 = self.log_dir_root + "DeepQLearningAgent_VS_MOISMCTSWithRandomRolloutsAgent_" + self.time
print(log_dir7)
print(
TensorboardTicTacToeRunner(
DeepQLearningAgent(9, 9),
MOISMCTSWithRandomRolloutsAgent(
100,
SafeTicTacToeRunner(RandomAgent(), RandomAgent())),
log_and_reset_score_history_threshold=10000,
log_dir=log_dir7).run(1000000000))
elif self.opponent == "MOISMCTSWithRandomRolloutsExpertThenApprenticeAgent":
log_dir8 = self.log_dir_root + "DeepQLearningAgent_VS_MOISMCTSWithRandomRolloutsExpertThenApprenticeAgent_" + self.time
print(log_dir8)
print(
TensorboardTicTacToeRunner(
DeepQLearningAgent(9, 9),
MOISMCTSWithRandomRolloutsExpertThenApprenticeAgent(
100,
SafeTicTacToeRunner(RandomAgent(), RandomAgent()),
9, 9),
log_and_reset_score_history_threshold=10000,
log_dir=log_dir8).run(1000000000))
elif self.opponent == "MOISMCTSWithValueNetworkAgent":
log_dir9 = self.log_dir_root + "DeepQLearningAgent_VS_MOISMCTSWithValueNetworkAgent_" + self.time
print(log_dir9)
print(
TensorboardTicTacToeRunner(
DeepQLearningAgent(9, 9),
MOISMCTSWithValueNetworkAgent(
100,
SafeTicTacToeRunner(RandomAgent(), RandomAgent())),
log_and_reset_score_history_threshold=10000,
log_dir=log_dir9).run(1000000000))
elif self.opponent == "DoubleQLearningAgent":
log_dir10 = self.log_dir_root + "DeepQLearningAgent_VS_DoubleQLearningAgent_" + self.time
print(log_dir10)
print(
TensorboardTicTacToeRunner(
DeepQLearningAgent(9, 9),
DoubleQLearningAgent(),
log_and_reset_score_history_threshold=10000,
log_dir=log_dir10).run(1000000000))
elif self.opponent == "RandomRolloutAgent":
nb_rollouts = 3
log_dir11 = self.log_dir_root + "DeepQLearningAgent_VS_RandomRolloutAgent(" + str(
nb_rollouts) + ")_" + self.time
print(log_dir11)
print(
TensorboardTicTacToeRunner(
DeepQLearningAgent(9, 9),
RandomRolloutAgent(
nb_rollouts,
SafeTicTacToeRunner(RandomAgent(), RandomAgent())),
log_and_reset_score_history_threshold=10000,
log_dir=log_dir11).run(1000000000))
else:
print("Unknown opponent")
if self.print_and_reset_score_history_threshold is not None and \
round_id % self.print_and_reset_score_history_threshold == 0:
print(score_history /
self.print_and_reset_score_history_threshold)
if self.prev_history is not None and \
score_history[0] == self.prev_history[0] and \
score_history[1] == self.prev_history[1] and \
score_history[2] == self.prev_history[2]:
self.stuck_on_same_score += 1
else:
self.prev_history = score_history
self.stuck_on_same_score = 0
if (self.
replace_player1_with_commandline_after_similar_results
is not None and self.stuck_on_same_score >= self.
replace_player1_with_commandline_after_similar_results
):
self.agents = (CommandLineAgent(), self.agents[1])
self.stuck_on_same_score = 0
return tuple(score_history)
if __name__ == "__main__":
print("MOISMCTSWithRandomRolloutsAgent VS RandomAgent")
print(
BasicTicTacToeRunner(
MOISMCTSWithRandomRolloutsAgent(
100, SafeTicTacToeRunner(RandomAgent(), RandomAgent())),
RandomAgent(),
print_and_reset_score_history_threshold=1000).run(1000))