def value_net(self, states, for_critic_loss=False):
    """Returns the output of the critic evaluated with the actor.

    Args:
      states: A [batch_size, num_state_dims] tensor representing a batch