def on_change(dialog):
dialog.close()
self.filename = dialog.filename
# GameWidget(board=dialog.map, bots=dialog.bots)
self.game = game.Game(board=core.Board(dialog.map),
bots=dialog.bots)
# self.game.open()
self.game.start()
def init(self, level=0, bw=core.EMPTY, token=None, board=None):
self._reset()
assert bool(token) == bool(board)
assert (level == 0) == bool(board)
self.level = level
self.bw = bw
self.token = token
self.board = board or core.Board()
def getBoardFromSequence(sequence): # Recreate a game board by replaying the sequence that leads to it ########################################## gameBoard=core.Board(8,16) tempSequence=sequence[:] while 1: if len(tempSequence)==0: return gameBoard chosenBox=tempSequence.pop(0) core.makeMove(gameBoard,chosenBox)
def create(token):
global _games
assert token, token not in _games
with _lock:
board = core.Board()
_games[token] = {
Items.BOARD: board,
Items.TURN: core.EMPTY,
Items.LOCKED: False
}
return board
def __init__(self, wh):
super(Window, self).__init__(resizable=True, width=wh[0], height=wh[1])
self.batch = pyglet.graphics.Batch()
self.board = core.Board(self, self.batch, wh)
self.collisions = core.mouse.Collisions(self, self.board._sprites)
self.input = core.lib.Input(self)
self.input.register_hold([
pyglet.window.key.UP, pyglet.window.key.DOWN,
pyglet.window.key.RIGHT, pyglet.window.key.LEFT
], self.key)
self.input.register_press(pyglet.window.key.F11, self.f_fullscreen)
self.scroll_directions = [0, 0]
self.autoscroll = False
self.active = True
def do_GET(self):
parsed_path = urlparse.urlparse(self.path)
params = {}
name_values = parsed_path.query.split(u'&')
for name_value in name_values:
index = name_value.find(u'=')
name = name_value[:index]
value = name_value[index + 1:]
params[name] = value
stones = [int(i) if i != "2" else -1 for i in params["stones"][:]]
pos = int(params["pos"])
stone = int(params["stone"])
board = core.Board(stones=stones)
board.move(pos, stone)
pre_stones = "".join([str(i) if i >= 0 else "2" for i in board.stones])
print pre_stones
epos = player.move(board, -stone)
print epos
movable_pos = board.calc_movable_pos(stone)
res = json.dumps(
{
"movables":
"".join(
["1" if i in movable_pos else "0" for i in xrange(0, 64)]),
"stones":
"".join([str(i) if i >= 0 else "2" for i in board.stones]),
"pre_stones":
pre_stones,
"pos":
epos,
"gameover":
board.check_gameover()
},
sort_keys=False,
indent=4)
self.send_response(200)
self.send_header('Content-type', 'text/json')
self.send_header('Access-Control-Allow-Origin', '*')
self.end_headers()
self.wfile.write(res)
return
prefs['console_debug'] = console_debug
# Load map
world = core.World(prefs['map_path'])
# Set players
players = []
available_players = agent.all_agents
for p, n in zip(prefs['players'], prefs['players_names']):
ag = available_players[p](n)
ag.console_debug = console_debug and False
players.append(ag)
# Set board
board = core.Board(world, players)
board.setPreferences(prefs)
#%% Draw the board
# Draw game as a simple graph
# Screen size and country box size
WIDTH = prefs['screen_W']
HEIGHT = prefs['screen_H']
SQ_RADIUS = prefs['sq_radius']
CONTINENT_BOX_PAD = prefs['continent_box_pad']
# Position and size of info box (buttons and messages)
INFO_BOX_X = prefs['info_box_x']
INFO_BOX_Y = prefs['info_box_y']
BUTTON_W = prefs['button_w']
BUTTON_H = prefs['button_h']
def play_games(nn_version,j,titer):
"""
Given a NN version, plays 1000 games of SPLT from start to finish and records their states, using a MCTS guided by nn_version.
Saves the states in memory.
Parameters:
----------
nn_version - NN object
NN object that will be playing the game
Returns:
----------
memory - deque object
Effectively a list of states. Each state is a list of the form [gameBoard, search_probs, splits]
"""
global new_max_splits, memory, lck
timedf_dict = {}
for game_num in range(int(nn_config.NUM_SELFTRAIN_GAMES/cpu_count())):
print("PLAYING GAME NUM "+str(game_num)+" ON CORE "+str(j))
gameBoard = core.Board()
core.makeMove(gameBoard,0) #makes first trivial split
split = 1
split_sequence = [0] #in case this ends up being a next best sequence
states = [] #stores each state in the game
nn_times = []
splt_times = []
total_times = []
while len(gameBoard.getMoveOptions()) != 0:
if split>=30: choose_best = True
else: choose_best = False
#print("-------------")
print("Game "+str(game_num) + ", Split "+str(split)+" on core "+str(j))
move, search_probs, nn_time, splt_time, rollout_time =\
make_move_MCTS(gameBoard,nn_version,choose_best,split)
nn_times.append(nn_time)
splt_times.append(splt_time)
total_times.append(rollout_time)
split_sequence.append(move)
gameState = np.squeeze(gameBoard.get_input_array())
states.append([gameState,search_probs])
core.makeMove(gameBoard,move)
split += 1
#add final split amount to state information, then save it in memory
for state in states:
state.append(split/1000)
lck.acquire()
memory.extend(states) #add the new states to memory
if split >= new_max_splits: #save a sequence of new maximum length
new_max_splits = split
pth = os.path.join(os.getcwd(),args.run_name,"highest_splits.txt")
with open(pth,'a') as f:
f.write(str(split)+": " + str(split_sequence)+'\n')
lck.release()
game_df = pd.DataFrame(data={'nn_time':nn_times,'splt_time':splt_times,'total_time':total_times})
timedf_dict[game_num] = game_df
time_df = pd.concat(timedf_dict,axis=1)
time_path = args.run_name+'/trainiter'+str(titer)+'_core'+str(j)+'_timedata_df.pkl'
with open(time_path,'wb') as f:
pickle.dump(time_df,f)
