Add CSV logging with pandas (train.csv, val.csv, per-epoch files)

Logs:
  {output_dir}/logs/train.csv — cumulative train metrics (all epochs)
  {output_dir}/logs/val.csv — cumulative val metrics (eval epochs)
  {output_dir}/logs/epoch_NNN_train.csv — per-epoch train
  {output_dir}/logs/epoch_NNN_val.csv — per-epoch val

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

src/training/train_gtauav.py

@@ -16,6 +16,7 @@ from dataclasses import dataclass, field
 from pathlib import Path
 
 import coloredlogs
+import pandas as pd
 import torch
 import torch.nn as nn
 from torch.amp import GradScaler, autocast
@@ -200,6 +201,36 @@ def _evaluate(
     return metrics
 
 
+class CSVLogger:
+    """Log train/val metrics to CSV files using pandas.
+
+    Creates:
+        {output_dir}/logs/train.csv       — all train epochs
+        {output_dir}/logs/val.csv         — all val epochs
+        {output_dir}/logs/epoch_{N}.csv   — per-epoch details
+    """
+
+    def __init__(self, output_dir: Path) -> None:
+        self.log_dir = output_dir / "logs"
+        self.log_dir.mkdir(parents=True, exist_ok=True)
+        self.train_rows: list[dict] = []
+        self.val_rows: list[dict] = []
+
+    def log_train(self, epoch: int, metrics: dict, lr: float, elapsed: float) -> None:
+        row = {"epoch": epoch, "lr": lr, "elapsed_s": round(elapsed, 1), **metrics}
+        self.train_rows.append(row)
+        # Append to cumulative CSV.
+        pd.DataFrame(self.train_rows).to_csv(self.log_dir / "train.csv", index=False)
+        # Per-epoch CSV.
+        pd.DataFrame([row]).to_csv(self.log_dir / f"epoch_{epoch:03d}_train.csv", index=False)
+
+    def log_val(self, epoch: int, metrics: dict) -> None:
+        row = {"epoch": epoch, **metrics}
+        self.val_rows.append(row)
+        pd.DataFrame(self.val_rows).to_csv(self.log_dir / "val.csv", index=False)
+        pd.DataFrame([row]).to_csv(self.log_dir / f"epoch_{epoch:03d}_val.csv", index=False)
+
+
 def _clear_vram() -> None:
     """Free VRAM from previous runs before starting."""
     import gc
@@ -354,6 +385,7 @@ def train(cfg: TrainConfigGTAUAV) -> None:
         LOGGER.info("🔄 Resuming from epoch %d", start_epoch)
 
     history: list[dict] = []
+    csv_logger = CSVLogger(output_dir)
 
     LOGGER.info("🚀 Starting training for %d epochs (from epoch %d)", cfg.epochs, start_epoch)
 
@@ -442,10 +474,14 @@ def train(cfg: TrainConfigGTAUAV) -> None:
             "train": means,
         }
 
+        # Log train metrics to CSV.
+        csv_logger.log_train(epoch, means, optimizer.param_groups[0]["lr"], elapsed)
+
         # Evaluation.
         if (epoch + 1) % cfg.eval_every == 0 or epoch == cfg.epochs - 1:
             val_metrics = _evaluate(model, test_loader, cfg.device)
             epoch_record["val"] = val_metrics
+            csv_logger.log_val(epoch, val_metrics)
             LOGGER.info(
                 "🎯 val epoch=%d  R@1=%.4f  R@5=%.4f  R@10=%.4f  gate_q=%.4f  gate_g=%.4f",
                 epoch,