def take_action(self, state):
legal = [op for op in range(4) if board(state).slide(op) != -1]
if legal:
op = self.choice(legal)
return action.slide(op)
else:
return action()
def take_action(self, before):
best_v = float('-inf')
best_a = None
best_op = None
for op in range(4):
after = board(before)
reward = after.slide(op)
if reward != -1:
tmp_v = reward + self.evaluate(after)
if tmp_v > best_v:
best_v = tmp_v
best_a = action.slide(op)
best_op = op
if not self.isFirst:
if best_v != float('-inf'):
self.update(self.last_state, best_v)
else:
self.update(self.last_state, 0)
self.last_state = board(before)
self.last_state.slide(best_op)
self.last_value = self.evaluate(self.last_state)
self.isFirst = False
if best_a == None:
return action()
else:
return best_a
def take_action(self, state):
scores = [board(state).slide(op) for op in range(4)]
max_value = max(scores)
if max_value != -1:
max_index = scores.index(max_value)
return action.slide(max_index)
else:
return action()
def take_action(self, state):
###
### Try your way
### Hint: select the action with maximum (reward + expect value)
###
# random action
legal = [op for op in range(4) if board(state).slide(op) != -1]
if legal:
op = self.choice(legal)
return action.slide(op)
else:
return action()
def take_action(self, state, weight):
#print(state)
legal = list(
filter(lambda x: x[1] != None,
[(op, weight.evaluate(state, op)) for op in range(4)]))
if legal:
argmax = max(legal, key=itemgetter(1))
op = argmax[0]
state.op = op
return action.slide(op)
else:
return action()
def select_best_action(self, board_state):
legal_ops = [
op for op in range(4) if board(board_state).slide(op) != -1
]
if legal_ops:
best_op = 0
best_value = -1
for op in legal_ops:
value = self.evaluate_state_action(board_state, op)
if value > best_value:
best_value = value
best_op = op
return action.slide(best_op)
else:
return action()
def take_action(self, state):
expValues = []
rewards = []
for op in range(4):
tmpBoard = board(state)
# get reward of afterstate
rewards.append(tmpBoard.slide(op))
if rewards[-1] == -1:
# When the action is not allowed (reward==-1),
# it is impossible to take the action
expValues.append(-float("inf"))
else:
expValues.append(rewards[-1] + self.lineValue(tmpBoard))
if max(rewards) == -1:
# if all the reward==-1,
# then gameover
return True, action()
best_move = np.argmax(expValues)
return False, action.slide(best_move)
def take_action(self, state):
max_value = -1
max_op = -1
for op in range(4):
new_board = board(state)
reward = new_board.slide(op)
if reward == -1:
continue
else:
expect = 0
expect = self.get_value(new_board)
expect += reward
if expect > max_value:
max_value = expect
max_op = op
if max_op == -1:
return action()
else:
return action.slide(max_op)
return
def initial_state(self):
return board()
def millisec(self):
return int(round(time.time() * 1000))
if __name__ == '__main__':
print('2048 Demo: episode.py\n')
# action, reward, time usage
moves = []
moves += [(action.place(0,1), 0, 1)]
moves += [(action.place(1,1), 0, 1)]
moves += [(action.slide(3), 2, 1)]
for mv in moves:
print(str(mv[0]) + str(mv[1]) + str(mv[2]))
print("".join([str(move[0]) + ("[" + str(move[1]) + "]" if move[1] else "") + ("(" + str(move[2]) + ")" if move[2] else "") for move in moves]))
sio = io.StringIO("0123")
print(sio.read(1))
print(sio.read(1))
print(sio.read(1))
print(sio.read(1))
print(sio.read(1) == "")
line = "".join([str(move[0]) + ("[" + str(move[1]) + "]" if move[1] else "") + ("(" + str(move[2]) + ")" if move[2] else "") for move in moves])
print(line)
minput = io.StringIO(line)
state = board()
def evaluate_state_action(self, board_state, op):
move = action.slide(op)
board_after_state, reward = self.compute_after_state(board_state, move)
return reward + self.lineValue(board_after_state)