def _build_normalizer_val(self):
mean = 0.0
std = 1.0 / (1.0 - self._discount)
norm = normalizer.Normalizer(sess=self._sess,
scope="val_norm",
size=1,
init_mean=mean,
init_std=std)
return norm
def _build_normalizer_action(self):
size = self.get_action_size()
high = self.get_action_bound_max().copy()
low = self.get_action_bound_min().copy()
inf_mask = np.logical_or((high >= np.finfo(np.float32).max), (low <= np.finfo(np.float32).min))
high[inf_mask] = 1.0
low[inf_mask] = -1.0
mean = 0.5 * (high + low)
std = 0.5 * (high - low)
norm = normalizer.Normalizer(sess=self._sess, scope="a_norm", size=size, init_mean=mean, init_std=std)
return norm
