def choose_move(self, backgammon):
""" Chooses a random board from all possible boards. """
# get a list of all possible moves
all_boards = BoardFactory.generate_all_boards(backgammon.current_player,\
backgammon.dice, backgammon.board)
# pick a random move
random_board = random.choice(all_boards)
return random_board
def choose_move(self, backgammon):
""" Function passes the list of all possible boards to the net and
evaluates them. Training can be disabled by the Player class
parameter - 'learning_mode'. In this case the net only
predicts the chances of winning for a board. """
best_board = None
# the value representing the expected contribution of a distinct
# board to win the game
expected_utility = -1.0
# next_out is the network output of the selected new board
next_output = []
# get all possible boards from BoardFactory
all_boards = BoardFactory.generate_all_boards(backgammon.current_player, \
backgammon.dice, \
backgammon.board)
# loop over all boards
for board in all_boards:
board_vector = self.board_to_vector(board)
output = self.neural_network.get_network_output(board_vector)
# translate network output into an actual meaning for player
# eg if output [0.1, 0.3], white odds of winning are lower
# than black's odds
utility = self.compute_utility(output)
if utility > expected_utility:
best_board = board
expected_utility = utility
next_output = output
# learning_mode indicates whether the network propagates back errors
# or only evaluates boards
if self.learning_mode:
current_input = self.board_to_vector(backgammon.board)
# pass the current board (board before this move is about to happen)
# to the network
current_output = self.neural_network.get_network_output(current_input)
self.neural_network.back_prop(current_output, next_output)
return best_board
