5e87cb3b9360b4bd10d40cdead521af60f0b0ae3,src/trainer.py,SamplingMultiTaskTrainer,_validate,#SamplingMultiTaskTrainer#Any#Any#Any#Any#Any#Any#Any#,434

Before Change

            if hasattr(task, "name") and g_scheduler is None:  // might be "is not None"?
                scheduler = task_infos[task.name]["scheduler"]
                if isinstance(scheduler, torch.optim.lr_scheduler.ReduceLROnPlateau):
                    scheduler.step(this_epoch_metric, epoch)
                else:
                    scheduler.step(epoch)
            elif g_scheduler is not None and task == "macro":
                scheduler = g_scheduler
                if isinstance(scheduler, torch.optim.lr_scheduler.ReduceLROnPlateau):
                    scheduler.step(this_epoch_metric, epoch)

After Change

                scheduler = g_scheduler
            else:
                scheduler = None
            if scheduler is not None and isinstance(scheduler, ReduceLROnPlateau):
                scheduler.step(this_epoch_metric, epoch)

        return all_val_metrics, should_save, new_best_macro, task_infos, metric_infos

    def _get_lr(self):
        if self._g_optimizer is not None:

Instances