sync ctt proxy feature mode code 2026-07-03

workspace/scripts/eval_ctt_proxy.py

@@ -44,6 +44,7 @@ def main(argv: list[str] | None = None) -> int:
     thresholds = [float(item) for item in args.thresholds.split(",") if item.strip()]
     checkpoint = torch.load(args.checkpoint, map_location="cpu")
     config = CTTConfig(**checkpoint["config"])
+    chart_feature_mode = str(checkpoint.get("chart_feature_mode", "base"))
     encoder = ChartEncoder(config.chart_feature_dim, output_dim=config.chart_dim)
     ctt = CausalTangentTransport(config)
     encoder.load_state_dict(checkpoint["chart_encoder"])
@@ -52,14 +53,23 @@ def main(argv: list[str] | None = None) -> int:
     ctt.eval()
     normalizer = TangentNormalizer.from_dict(checkpoint["normalizer"])
 
-    source_charts, source_index = load_charts(args.source_index, max_charts=None)
-    target_charts, target_index = load_charts(args.target_index, max_charts=args.max_target_charts)
+    source_charts, source_index = load_charts(
+        args.source_index,
+        max_charts=None,
+        chart_feature_mode=chart_feature_mode,
+    )
+    target_charts, target_index = load_charts(
+        args.target_index,
+        max_charts=args.max_target_charts,
+        chart_feature_mode=chart_feature_mode,
+    )
     _validate_indexes(args.source_index, source_index, args.target_index, target_index)
     rows = []
     log_lines = [
         f"source_charts={len(source_charts)} target_charts={len(target_charts)} k={args.k}",
         f"source_index={args.source_index}",
         f"target_index={args.target_index}",
+        f"chart_feature_mode={chart_feature_mode}",
     ]
     source_by_task: dict[str, list[Any]] = {}
     for chart in source_charts:
@@ -141,6 +151,7 @@ def main(argv: list[str] | None = None) -> int:
         "data_hash": source_index.get("content_hash"),
         "split_hash": source_index.get("split_hash"),
         "target_split_hash": target_index.get("split_hash"),
+        "chart_feature_mode": chart_feature_mode,
     }
     (out_dir / "metrics.json").write_text(json.dumps(metrics, indent=2, sort_keys=True) + "\n")
     (out_dir / "metrics_by_task.json").write_text(json.dumps(_by_task(rows, metric_names), indent=2, sort_keys=True) + "\n")