Include optimizer in experiment directory naming

- Add optimizer to TrainingDefaults (default: sgd)
- Include optimizer in path when non-default (e.g., adamw)
- Update get_experiment_dir callers in config.py and train_kd.py

Fixes issue where SGD and AdamW experiments at same lr would overwrite
each other. Now ConvNeXt AdamW experiments get paths like:
  results/raw/cifar10/convnext_tiny/std_adamw_lr0.004_s42/

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Author: dariocazzani

experiments/config.py

@@ -113,6 +113,7 @@ def __post_init__(self) -> None:
                 augment=self.augment,
                 ablation=self.ablation.value,
                 lr=self.lr,
+                optimizer=self.optimizer,
             )
             self.output_dir = str(experiment_dir)
 

experiments/paths.py

@@ -23,6 +23,7 @@ class TrainingDefaults:
     lr: float = 0.1
     augment: str = "basic"
     ablation: str = "none"
+    optimizer: str = "sgd"
 
 
 @dataclass(frozen=True)
@@ -47,6 +48,7 @@ def get_experiment_dir(
     augment: str = TRAINING_DEFAULTS.augment,
     ablation: str = TRAINING_DEFAULTS.ablation,
     lr: float = TRAINING_DEFAULTS.lr,
+    optimizer: str = TRAINING_DEFAULTS.optimizer,
     kd_temperature: float | None = None,
     kd_alpha: float | None = None,
 ) -> Path:
@@ -78,12 +80,16 @@ def get_experiment_dir(
             parts.append(augment)
         if ablation != TRAINING_DEFAULTS.ablation:
             parts.append(ablation)
+        if optimizer != TRAINING_DEFAULTS.optimizer:
+            parts.append(optimizer)
         if lr != TRAINING_DEFAULTS.lr:
             parts.append(f"lr{lr:g}")
     else:
         parts.append("bit_kd")
         if ablation != TRAINING_DEFAULTS.ablation:
             parts.append(ablation)
+        if optimizer != TRAINING_DEFAULTS.optimizer:
+            parts.append(optimizer)
         if lr != TRAINING_DEFAULTS.lr:
             parts.append(f"lr{lr:g}")
         if kd_temperature is not None and kd_temperature != KD_DEFAULTS.temperature:

experiments/train_kd.py

@@ -254,6 +254,7 @@ def main() -> None:
             args.seed,
             ablation=args.ablation,
             lr=args.lr,
+            optimizer=args.optimizer,
             kd_temperature=args.temperature,
             kd_alpha=args.alpha,
         )