Error: /paddle/paddle/fluid/operators/elementwise/elementwise_op_function.cu.h:88 Assertion `y_[id] != 0` failed. InvalidArgumentError: Integer division by zero encountered in divide.Please check. about parl HOT 5 CLOSED

def log_prob(self, value):
       var = self.scale * self.scale
        log_scale = nn.log(self.scale)
        return -1. * ((value - self.loc) * (value - self.loc)) / (
            2. * var) - log_scale - math.log(math.sqrt(2. * math.pi))

        values = self.model.value(obs)
        actions_mu, log_std_mu = self.model.policy(obs)
        std_pi = layers.exp(log_std)
        normal_pi = Normal(actions, std_pi)
        x_t1 = normal_pi.sample([1])[0]
        y_t1 = layers.tanh(x_t1)
        # action1 = y_t1 * self.max_action
        log_prob1 = normal_pi.log_prob(x_t1)
        log_prob1 -= layers.log(self.max_action * (1 - layers.pow(y_t1, 2)) + epsilon)
        log_prob1 = layers.reduce_sum(log_prob1, dim=1, keep_dim=True)
        log_prob_pi = layers.squeeze(log_prob1, axes=[1])

        std_mu = layers.exp(log_std_mu)
        normal_mu = Normal(actions_mu, std_mu)
        x_t2 = normal_mu.sample([1])[0]
        y_t2 = layers.tanh(x_t2)
        # action2 = y_t2 * self.max_action
        log_prob2 = normal_mu.log_prob(x_t2)
        log_prob2 -= layers.log(self.max_action * (1 - layers.pow(y_t2, 2)) + epsilon)
        log_prob2 = layers.reduce_sum(log_prob2, dim=1, keep_dim=True)
        log_prob_mu = layers.squeeze(log_prob2, axes=[1])

        # target_policy_distribution = CategoricalDistribution(target_logits)
        # behaviour_policy_distribution = CategoricalDistribution(
        #     behaviour_logits)

        policy_entropy = normal_pi.entropy()
        target_actions_log_probs = log_prob_mu
        behaviour_actions_log_probs = log_prob_pi

        # Calculating kl for debug
        # kl = target_policy_distribution.kl(behaviour_policy_distribution)
        kl = normal_pi.kl_divergence(normal_mu)
        kl = layers.reduce_mean(kl)