def __call__(self, p1, p2):
self.temp = self.minTemperature
if self.useNetworks:
p1 = ModuleDecidingPlayer(p1, self.task.env, temperature=self.temp)
p2 = ModuleDecidingPlayer(p2, self.task.env, temperature=self.temp)
else:
assert isinstance(p1, GomokuPlayer)
assert isinstance(p2, GomokuPlayer)
p1.game = self.task.env
p2.game = self.task.env
p1.color = GomokuGame.BLACK
p2.color = -p1.color
self.player = p1
self.opponent = p2
# the games with increasing temperatures and lower coefficients
coeffSum = 0.
res = 0.
for i in range(self.maxGames):
coeff = 1 / (10 * self.temp + 1)
res += coeff * self._oneGame()
coeffSum += coeff
if i > 0:
self._globalWarming()
return res / coeffSum
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)
