def __init__(self, size, opponent = None, **args):
EpisodicTask.__init__(self, GomokuGame((size, size)))
self.setArgs(**args)
if opponent == None:
opponent = RandomGomokuPlayer(self.env)
elif isclass(opponent):
# assume the agent can be initialized without arguments then.
opponent = opponent(self.env)
if not self.opponentStart:
opponent.color = GomokuGame.WHITE
self.opponent = opponent
self.minmoves = 9
self.maxmoves = self.env.size[0] * self.env.size[1]
self.reset()
def __init__(self, size, opponent=None, **args):
EpisodicTask.__init__(self, CaptureGame(size))
self.setArgs(**args)
if opponent == None:
opponent = RandomCapturePlayer(self.env)
elif isclass(opponent):
# assume the agent can be initialized without arguments then.
opponent = opponent(self.env)
else:
opponent.game = self.env
if not self.opponentStart:
opponent.color = CaptureGame.WHITE
self.opponent = opponent
self.maxmoves = self.env.size * self.env.size
self.minmoves = 3
self.reset()
def __call__(self, x):
""" If a module is given, wrap it into a ModuleDecidingAgent before evaluating it.
Also, if applicable, average the result over multiple games. """
if isinstance(x, Module):
agent = ModuleDecidingPlayer(x, self.env, greedySelection = True)
elif isinstance(x, GomokuPlayer):
agent = x
else:
raise NotImplementedError('Missing implementation for '+x.__class__.__name__+' evaluation')
res = 0
agent.game = self.env
self.opponent.game = self.env
for dummy in range(self.averageOverGames):
agent.color = -self.opponent.color
res += EpisodicTask.__call__(self, agent)
return res / float(self.averageOverGames)
def performAction(self, action):
EpisodicTask.performAction(self, action)
if not self.isFinished():
EpisodicTask.performAction(self, self.opponent.getAction())
def reset(self):
self.switched = False
EpisodicTask.reset(self)
if self.opponent.color == CaptureGame.BLACK:
# first move by opponent
EpisodicTask.performAction(self, self.opponent.getAction())
def reset(self):
EpisodicTask.reset(self)
self.env.reset()
self._ended = False