class LearningBoard(TicTacToeBoard):
winning_action_reward = 1.0
loosing_action_reward = -0.5
neutral_action_reward = 0.0
prohibited_action_reward = -1.0
def MakeMoveWithReward(self, action):
row = action // 3
column = action % 3
if self.SiteIsOccupied(row, column):
return self.prohibited_action_reward, True
elif self.MakeMove(row, column):
return self.winning_action_reward, True
else:
if self.GameEnded():
return self.neutral_action_reward, True
return self.neutral_action_reward, False
store_path = "TicTacToeNewTraining"
TicTacToeANN = ANN.TrainingNetwork(18, 9, 0.05, [50, 50, 50])
if __name__ == '__main__':
board = LearningBoard()
#TicTacToeANN.Training_Episodic_Decorrelated_Batches(board,store_path,500, 50, 5, TicTacToeANN.SARSA_Episodic)
TicTacToeANN.Training_Episodic_Single_Matches_Reverse(
board, store_path, 25000, TicTacToeANN.SARSA_Episodic_Single_Game)
class LearningFourInARow(FourInARowBoard):
winning_action_reward = 1.0
loosing_action_reward = -0.5
neutral_action_reward = 0.0
prohibited_action_reward = -1.0
def MakeMoveWithReward(self, action):
if self.ColumnIsFull(action):
return self.prohibited_action_reward, True
elif self.MakeMove(action):
return self.winning_action_reward, True
else:
if self.GameEnded():
return self.neutral_action_reward, True
return self.neutral_action_reward, False
store_path = "FourInARowNewTraining"
FourInARowANN = ANN.TrainingNetwork(84, 7, 0.05, [50, 50, 50])
if __name__ == '__main__':
board = LearningFourInARow()
FourInARowANN.Training_Episodic_Single_Matches_Reverse(
board, store_path, 10000,
FourInARowANN.Q_Learning_Episodic_Single_Game)
#FourInARowANN.Training_Episodic_Decorrelated_Batches(board,store_path,50, 50, 5, FourInARowANN.Q_Learning_Episodic)
elif np.all(self.current_position == self.goal):
self.current_position = np.array([row, column])
return self.winning_action_reward, True
else:
self.current_position = np.array([row, column])
return self.neutral_action_reward, False
def GetState(self):
return self.current_position
def ResetGame(self):
self.current_position = self.maze.start
store_path = "MazeNewTraining"
MazeANN = ANN.TrainingNetwork(2, 4, 0.05, [50, 50, 50], discount=0.95)
def PrintLearnedPath(name):
runner = MazeRunner(name)
game_ended = False
saver = ANN.tf.train.Saver()
path = "Trainings/" + store_path
with ANN.tf.Session() as sess:
saver.restore(sess, path)
count = 0
while not game_ended:
if runner.maze.hight * runner.maze.width < count:
break
count += 1
current_state = runner.GetState()
