aa3c6a5f79d82044306f3a11f1de6b7231163d90,tf_agents/agents/behavioral_cloning/behavioral_cloning_agent.py,BehavioralCloningAgent,_loss,#BehavioralCloningAgent#Any#Any#,226

Before Change

      //   the number of boundary samples increases.
      if weights is not None:
        error *= weights
      loss = tf.reduce_mean(input_tensor=error)

      with tf.name_scope("Losses/"):
        tf.compat.v2.summary.scalar(
            name="loss", data=loss, step=self.train_step_counter)

      if self._summarize_grads_and_vars:
        with tf.name_scope("Variables/"):
          for var in self._cloning_network.trainable_weights:
            tf.compat.v2.summary.histogram(

After Change

      //   their contribution in the loss. Think about what would happen as
      //   the number of boundary samples increases.

      agg_loss = common.aggregate_losses(
          per_example_loss=error,
          sample_weight=weights,
          regularization_loss=self._cloning_network.losses)
      total_loss = agg_loss.total_loss

      dict_losses = {"loss": agg_loss.weighted,
                     "reg_loss": agg_loss.regularization,
                     "total_loss": total_loss}

      common.summarize_scalar_dict(dict_losses,
                                   step=self.train_step_counter,
                                   name_scope="Losses/")

      if self._summarize_grads_and_vars:
        with tf.name_scope("Variables/"):
          for var in self._cloning_network.trainable_weights:

Instances