def _best_actions(self, board):
alpha, beta = -999999, 999999
states, indexes = next_states(board, self.player, index=True)
values = []
potential_actions = []
for state, index in zip(states, indexes):
value = collapse_state(self.grader, state, self.reach, self.player,
self.alpha_beta, alpha, beta)
values.append(value)
if self.alpha_beta > 0: # Alpha
if value > alpha:
alpha = value
potential_actions = [index]
elif value == alpha:
potential_actions.append(index)
elif self.alpha_beta < 0: # Beta
if value < beta:
beta = value
potential_actions = [index]
elif value == beta:
potential_actions.append(index)
if len(potential_actions) == 1:
return potential_actions[0]
r = np.random.randint(0, len(potential_actions))
return potential_actions[r]
def _best_actions(self, board):
best_value = -9999999 * self.mini_max
states, indexes = next_states(board, self.player, index=True)
values = []
best_actions = []
for state, index in zip(states, indexes):
value = collapse_state(self.grader, state, self.reach, self.player,
(self.mini_max * -1))
values.append(value)
if self.mini_max > 0: # Maximize
if value > best_value:
best_value = value
best_actions = [index]
elif value == best_value:
best_actions.append(index)
elif self.mini_max < 0: # Minimize
if value < best_value:
best_value = value
best_actions = [index]
elif value == best_value:
best_actions.append(index)
print 'values', values
# if len(best_actions) == 1:
# return best_actions[0]
r = np.random.randint(0, len(best_actions))
return best_actions[r]
def collapse_state(grader,
board,
reach,
player,
ab,
alpha=-999999,
beta=999999,
level=0):
if reach == 0 or np.count_nonzero(
board) == board.size or grader.check_board(board, player):
return grader.grade_board(board, player=player) - level * 100
states = next_states(board, (-1 * player))
values = []
for state in states:
value = collapse_state(grader, state, (reach - 1), player, (ab * -1),
alpha, beta, (level + 1))
if ab == 2: # Alpha Pruning
if value > beta:
return value
elif ab == -2: # Beta Pruning
if value < alpha:
return value
values.append(value)
if ab > 0:
return max(values)
elif ab < 0:
return min(values)
return
def _best_action(self, board):
states, indexes = next_states(board, self.player, index=True)
max_value = None
max_indexes = []
for state, index in zip(states, indexes):
value = self._value_state(state)
if max_value is None:
max_value = value
max_indexes = [index]
elif value > max_value:
max_value = value
max_indexes = [index]
elif value == max_value:
max_indexes.append(index)
size = len(max_indexes)
r = np.random.randint(0, size)
return max_indexes[r]