def _test():
"""
Nothing important here.
"""
state.init()
out = _convert_and_save_model('../notebooks/saved_mlp_model_with_w.h5')
print(out)
def think(dt, events, kpress, mpos): global alpha, playing playing = bool(hud.dialoguebox) if playing: alpha = min(1 * dt + alpha, 1) else: alpha -= 1 * dt if alpha <= 0: state.init() scene.swap(scenes.game) return sound.setvolume(alpha) if kpress[K_ESCAPE]: playing = False for event in events: if event.type == MOUSEBUTTONDOWN: if event.button == 1: hud.dialoguebox.settext(lines.pop(0) if lines else None) if event.type == KEYDOWN and event.key == K_SPACE: hud.dialoguebox.settext(lines.pop(0) if lines else None) if event.type == KEYDOWN and event.key == K_ESCAPE: hud.dialoguebox.settext() if event.type == KEYDOWN and event.key == K_F12: graphics.screenshot() dx, dy, dr, dA = -0.02 * dt, -0.005 * dt, 0, 0 camera.move(1 * dx, 1 * dy, 0.7 * dr, 0.7 * dA) camera.think(dt) hud.think(dt)
def init(targets):
if not os.environ.get('REDO'):
# toplevel call to redo
if len(targets) == 0:
targets.append('all')
exenames = [os.path.abspath(sys.argv[0]),
os.path.realpath(sys.argv[0])]
dirnames = [os.path.dirname(p) for p in exenames]
trynames = ([os.path.abspath(p+'/../lib/redo') for p in dirnames] +
[p+'/redo-sh' for p in dirnames] +
dirnames)
seen = {}
dirs = []
for k in trynames:
if not seen.get(k):
seen[k] = 1
dirs.append(k)
os.environ['PATH'] = ':'.join(dirs) + ':' + os.environ['PATH']
os.environ['REDO'] = os.path.abspath(sys.argv[0])
if not os.environ.get('REDO_BASE'):
base = os.path.commonprefix([os.path.abspath(os.path.dirname(t))
for t in targets] + [os.getcwd()])
bsplit = base.split('/')
for i in range(len(bsplit), 0, -1):
newbase = '/'.join(bsplit[:i])
if os.path.exists(newbase + '/.redo'):
base = newbase
break
os.environ['REDO_BASE'] = base
os.environ['REDO_STARTDIR'] = os.getcwd()
import state
state.init()
def init(targets):
if not os.environ.get('REDO'):
# toplevel call to redo
exenames = [os.path.abspath(sys.argv[0]),
os.path.realpath(sys.argv[0])]
if exenames[0] == exenames[1]:
exenames = [exenames[0]]
dirnames = [os.path.dirname(p) for p in exenames]
os.environ['PATH'] = ':'.join(dirnames) + ':' + os.environ['PATH']
os.environ['REDO'] = os.path.abspath(sys.argv[0])
if not os.environ.get('REDO_BASE'):
base = os.path.commonprefix([os.path.abspath(os.path.dirname(t))
for t in targets] + [os.getcwd()])
bsplit = base.split('/')
for i in range(len(bsplit)-1, 0, -1):
newbase = '/'.join(bsplit[:i])
if os.path.exists(newbase + '/.redo'):
base = newbase
break
os.environ['REDO_BASE'] = base
os.environ['REDO_STARTDIR'] = os.getcwd()
import state
state.init()
def init(targets):
if not os.environ.get('REDO'):
# toplevel call to redo
if len(targets) == 0:
targets.append('all')
exenames = [os.path.abspath(sys.argv[0]),
os.path.realpath(sys.argv[0])]
dirnames = [os.path.dirname(p) for p in exenames]
trynames = ([os.path.abspath(p+'/../lib/redo') for p in dirnames] +
[p+'/redo-sh' for p in dirnames] +
dirnames)
seen = {}
dirs = []
for k in trynames:
if not seen.get(k):
seen[k] = 1
dirs.append(k)
os.environ['PATH'] = ':'.join(dirs) + ':' + os.environ['PATH']
os.environ['REDO'] = os.path.abspath(sys.argv[0])
if not os.environ.get('REDO_BASE'):
base = os.path.expanduser('~')
atoms = os.getcwd().split('/')
for i in range(len(atoms), 0, -1):
possible_base = '/'.join(atoms[:i])
if os.path.exists(possible_base + '/.redo-base'):
base = possible_base
os.environ['REDO_BASE'] = base
os.environ['REDO_STARTDIR'] = os.getcwd()
import state
state.init()
os.environ['REDO_TARGETS_SEEN'] = os.environ.get('REDO_TARGETS_SEEN', '')
def main():
pygame.init()
pygame.mixer.init()
sound.setvolume(0)
pygame.font.init()
graphics.setmode()
graphics.init()
if settings.playback:
state0, record = cPickle.load(open(data.filepath("record.pkl"), "rb"))
state.loadobj(state0)
scene.push(scenes.game if settings.skipmenu else scenes.menu)
elif settings.restart:
state.init()
scene.push(scenes.game if settings.skipmenu else scenes.menu)
else:
scene.push(scenes.menu)
if state.load():
scene.push(scenes.game)
clock = pygame.time.Clock()
if settings.playback:
for rstate, dt, events, kpress, mpos in record:
clock.tick(settings.fps)
s = scene.top()
random.setstate(rstate)
if settings.vidcap:
vidcap._mpos = mpos
s.think(dt, events, kpress, mpos)
s.draw(mpos)
pygame.display.flip()
return
if settings.getrecord:
record = []
state0 = cPickle.loads(cPickle.dumps(state.saveobj()))
while scene.top():
dt = min(0.001 * clock.tick(settings.fps), 0.5)
if settings.fixfps:
dt = 1.0 / settings.fps
s = scene.top()
while dt > 0:
tdt = min(dt, 0.1)
events = map(EventProxy, pygame.event.get())
kpress = pygame.key.get_pressed()
mpos = pygame.mouse.get_pos()
if settings.getrecord:
record.append((random.getstate(), tdt, events, kpress, mpos))
s.think(tdt, events, kpress, mpos)
dt -= tdt
s.draw(mpos)
pygame.display.flip()
graphics.quit()
if settings.getrecord:
cPickle.dump([state0, record], open(data.filepath("record.pkl"), "wb"))
def main():
state_module.init()
clock = pygame.time.Clock()
while True:
if not state_module.state.handle_input(): break
state_module.state.draw()
pygame.display.flip()
clock.tick(30)
pygame.quit()
def main():
state_module.init()
clock = pygame.time.Clock()
while True:
if not state_module.state.handle_input(): break
state_module.state.draw()
pygame.display.flip()
clock.tick(30)
pygame.quit()
def init(): glEnable(GL_COLOR_MATERIAL) glColorMaterial(GL_FRONT_AND_BACK, GL_AMBIENT_AND_DIFFUSE) glEnable(GL_NORMALIZE) lighting.init() state.init() hud.init() hud.menumode() camera.menuinit() global playing, alpha, selected playing = True alpha = 0 selected = None sound.playmusic(info.music["menu"]) state.R = 40
def initialise_tasks():
# Initialise state storage (state.db)
state.init(os.path.join(state_dir_path, 'state'))
# Ush V3 link initialisation
yield ush_v3.get_link().initialise_tasks()
# Datachannel directory initialisation
yield capture_api.get_directory().initialise()
register_task(capture_api.DatachannelDirectoryRefreshTask(task_id="dcdir-refresh", first_delay=30), start=True)
# Initialise pull/push routine for each of the events/channels
for (event_id, event) in ush_v3.get_link().events.iteritems():
logger.info("Starting pull/push routine for event=%s (%s)" % (event_id, event.display_name))
t = pull_push.create_themes_pull_task(event,
chunk_size=theme_pull_chunk_size,
period=theme_pull_period, first_delay=(0,15))
register_task(t, start=True)
def init(targets):
if not os.environ.get('REDO'):
# toplevel call to redo
if len(targets) == 0:
targets.append('all')
exenames = [
os.path.abspath(sys.argv[0]),
os.path.realpath(sys.argv[0])
]
dirnames = [os.path.dirname(p) for p in exenames]
trynames = ([os.path.abspath(p + '/../lib/redo')
for p in dirnames] + [p + '/redo-sh'
for p in dirnames] + dirnames)
seen = {}
dirs = []
for k in trynames:
if not seen.get(k):
seen[k] = 1
dirs.append(k)
os.environ['PATH'] = ':'.join(dirs) + ':' + os.environ['PATH']
os.environ['REDO'] = os.path.abspath(sys.argv[0])
if not os.environ.get('REDO_BASE'):
base = os.path.commonprefix(
[os.path.abspath(os.path.dirname(t))
for t in targets] + [os.getcwd()])
bsplit = base.split('/')
for i in range(len(bsplit) - 1, 0, -1):
newbase = '/'.join(bsplit[:i])
if os.path.exists(newbase + '/.redo'):
base = newbase
break
os.environ['REDO_BASE'] = base
os.environ['REDO_STARTDIR'] = os.getcwd()
import state
state.init()
os.environ['REDO_LOCKS'] = os.environ.get('REDO_LOCKS', '')
def start():
state.init()
storage = TagStorage(get_store_path())
state.set_storage(storage)
def nfcLoaded():
nfc_reader.onTagRead += on_tag_read
nfc_reader.onLoad += nfcLoaded
nfc_reader.start_nfc_thread()
web_interface.init()
player = MediaPlayer()
state.set_player(player)
# TODO: Startup Sound
#player.play_ding()
# Blocks, keep last.
web_interface.run_wait()
print("Posted file: {}".format(request.files['file']))
# def check_timeout_condition():
# """
# TO BE IMPLEMENTED.
# """
# TIMEOUT_DELTA_IN_MINS = 10
# time_now = time.time()
# if time_now > TIMEOUT_DELTA_IN_MINS * 60:
# # Need to trigger the event of broadcasting to all nodes.
# # The nodes to drop everything they were doing.
# pass
if __name__ == '__main__':
log.startLogging(sys.stdout)
factory = CloudNodeFactory()
factory.protocol = CloudNodeProtocol
wsResource = WebSocketResource(factory)
wsgiResource = WSGIResource(reactor, reactor.getThreadPool(), app)
rootResource = WSGIRootResource(wsgiResource, {b'': wsResource})
site = Site(rootResource)
state.init()
reactor.listenTCP(8999, site)
reactor.run()
EVALUATE_EVERY_STEP = 100
EVALUATION_ROUNDS = 20
STEPS_BEFORE_TAU_0 = 10
device = torch.device("cpu")
path = os.getcwd()
model_path = os.path.join(path, f"model4")
net = torch.load(model_path)
mcts = mcts.MCTS()
while True:
won = None
cur_player = s.get_random_player()
cur_state = s.init()
print(s.decode(cur_state))
while won is None:
print(f"Player: {cur_player}")
if cur_player == 1:
mcts.search_batch(20,
16,
cur_state,
cur_player,
net,
device=device)
probs, _ = mcts.get_policy_value(cur_state, tau=0)
def __init__(self): state.init() state.start() self.t = 0
def play_game(mcts_stores,
replay_buffer,
net1,
net2,
steps_before_tau_0,
mcts_searches,
mcts_batch_size,
net1_plays_first=None,
device="cpu"):
"""
Play one single game, memorizing transitions into the replay buffer
:param mcts_stores: could be None or single MCTS or two MCTSes for individual net
:param replay_buffer: queue with (state, probs, values), if None, nothing is stored
:param net1: player1
:param net2: player2
:return: value for the game in respect to player1 (+1 if p1 won, -1 if lost, 0 if draw)
"""
if mcts_stores is None:
mcts_stores = [mcts.MCTS(), mcts.MCTS()]
elif isinstance(mcts_stores, mcts.MCTS):
mcts_stores = [mcts_stores, mcts_stores]
state = s.init()
nets = [net1, net2]
if net1_plays_first is None:
cur_player = s.get_random_player()
else:
cur_player = 1 if net1_plays_first else 2
step = 0
tau = 1 if steps_before_tau_0 > 0 else 0
game_history = []
result = None
net1_result = None
while result is None:
mcts_stores[cur_player - 1].search_batch(mcts_searches,
mcts_batch_size,
state,
cur_player,
nets[cur_player - 1],
device=device)
probs, _ = mcts_stores[cur_player - 1].get_policy_value(state, tau=tau)
game_history.append((state, cur_player, probs))
action = np.random.choice(9, p=probs)
if action not in s.get_action_space(state):
print("Impossible action selected")
state, won = s.move(state, action, cur_player)
if won:
result = 1
net1_result = 1 if cur_player == 1 else -1
break
cur_player = s.next_player(cur_player)
# check the draw case
if len(s.get_action_space(state)) == 0:
result = 0
net1_result = 0
break
step += 1
if step >= steps_before_tau_0:
tau = 0
if replay_buffer is not None:
for state, cur_player, probs in reversed(game_history):
replay_buffer.append((state, cur_player, probs, result))
result = -result
return net1_result, step
