def get_move(self, state, moves, side):
self.epochs += 1
if self.no_learn:
self.mepsilon = 1.
else:
self.mepsilon = min(0.99, 0.5 + self.epochs / 1e5)
if self.foil and self.epochs % 1000 == 0:
print "Reloading network"
self.load_network()
if random.random() > self.mepsilon:
action = random.choice(list(valid_columns(state)))
else:
action = self.ms2m(moves)
state1 = moves_to_state(moves + [action])
if not self.no_learn:
self.learner(self.state0, state1=state1)
self.state0 = state1
return action
def moves_to_move(moves0):
state0 = moves_to_state(moves0)
Qs = Q_fn(np.array([state0]))[0]
return np.nanargmax(Qs)