Fix mermaid error

Author: Axect

docs/03_training_loop___trainer___.md

@@ -114,7 +114,7 @@ sequenceDiagram
         Trainer->>Trainer: Calculate loss(y_pred, y)
         Trainer->>+OptimizerScheduler: Backpropagate loss & Update weights (optimizer.step())
         OptimizerScheduler-->>-Trainer: Weights updated
-        Trainer->>-DataLoaderTrain: Repeat for all batches in dl_train
+        Trainer->>DataLoaderTrain: Repeat for all batches in dl_train
         Trainer-->>-Trainer: Return average train_loss for epoch
 
         Trainer->>+Trainer: val_epoch(dl_val) # Ask Trainer to validate
@@ -123,7 +123,7 @@ sequenceDiagram
         Trainer->>+Model: Forward pass: model(x) (No gradient tracking)
         Model-->>-Trainer: Return prediction y_pred
         Trainer->>Trainer: Calculate loss(y_pred, y)
-        Trainer->>-DataLoaderVal: Repeat for all batches in dl_val
+        Trainer->>DataLoaderVal: Repeat for all batches in dl_val
         Trainer-->>-Trainer: Return average val_loss for epoch
 
         Trainer->>+LoggerEarlyStop: Log metrics (train_loss, val_loss, lr)
@@ -141,7 +141,6 @@ sequenceDiagram
     alt Training Finished Normally or Early Stopped
         Trainer-->>-RunFunc: Return final_val_loss
     end
-end
 ```
 
 This diagram shows the cycle: for each epoch, the `Trainer` calls `train_epoch` (which iterates through training batches, performs forward/backward passes, and updates weights) and `val_epoch` (which iterates through validation batches and calculates loss without updating weights). After each epoch, it logs metrics, checks for early stopping, and adjusts the learning rate.

docs/06_pruning_strategy___pflpruner___.md

@@ -103,30 +103,33 @@ sequenceDiagram
     loop For each Epoch (within trainer.train)
         Trainer->>Trainer: Run train_epoch()
         Trainer->>Trainer: Run val_epoch()
-        Trainer->>+PFLPruner: pruner.report(trial_id=N, epoch=E, value=val_loss)
+        Trainer->>+PFLPruner: pruner.report(trial_id=N, epoch=E, value=val_loss) # PFLPruner 활성화 시작
         PFLPruner->>+TrialState: Update loss history for Trial N
         TrialState-->>-PFLPruner: Done
         PFLPruner->>PFLPruner: Calculate Predicted Final Loss (PFL)
         PFLPruner->>PFLPruner: Compare PFL with Top-K finished trials
-        PFLPruner->>+PFLPruner: pruner.should_prune() ?
+        PFLPruner->>PFLPruner: Check if should_prune() is True?
         alt Pruning conditions met
+            # report 호출에 대한 응답으로 True 반환하며 PFLPruner 비활성화
             PFLPruner-->>-Trainer: Return True
             Trainer->>Trainer: Raise optuna.TrialPruned exception
-            Trainer-->>-ObjectiveFunc: Exception caught
-            ObjectiveFunc-->>-OptunaStudy: Report Trial N as Pruned
+            Trainer-->>-ObjectiveFunc: Exception caught # Trainer 비활성화
+            ObjectiveFunc-->>-OptunaStudy: Report Trial N as Pruned # ObjectiveFunc 비활성화
+            # 여기서 루프가 중단될 수 있음 (break 등 명시적 표현은 Mermaid 표준에 없음)
         else Pruning conditions NOT met
+            # report 호출에 대한 응답으로 False 반환하며 PFLPruner 비활성화
             PFLPruner-->>-Trainer: Return False
             Trainer->>Trainer: Continue to next epoch...
         end
     end
-    alt Trial Finishes Normally (or Early Stopping)
+    alt Trial Finishes Normally (or Early Stopping outside pruning)
+        # 루프 종료 후 Trainer 비활성화 (정상 종료 또는 Early Stopping)
         Trainer-->>-ObjectiveFunc: Return final_val_loss
         ObjectiveFunc->>+PFLPruner: pruner.complete_trial(trial_id=N)
         PFLPruner->>PFLPruner: Update Top-K completed trials if necessary
-        PFLPruner-->>-ObjectiveFunc: Done
-        ObjectiveFunc-->>-OptunaStudy: Report Trial N result (final_val_loss)
+        PFLPruner-->>-ObjectiveFunc: Done # PFLPruner 비활성화
+        ObjectiveFunc-->>-OptunaStudy: Report Trial N result (final_val_loss) # ObjectiveFunc 비활성화
     end
-end
 ```
 
 1.  **Setup:** When `main.py` sets up the Optuna study ([Chapter 5]), it also creates the `PFLPruner` instance based on the `optimize_config.yaml`. This pruner instance is passed down through the `objective` function to the `util.run` function, and finally to the `Trainer` when it's initialized for a specific trial.