updated config generation.

tools/make_configs.py

@@ -36,31 +36,38 @@ def build_fewshot_split_map(
     for base_cfg, splits in base_split_map.items():
         base_train = sorted(splits["train"])
         base_id_test = sorted(splits["id_test"])
-        base_ood = sorted(splits["ood_test"])
-
-        k_targets = {
-            "1": 1,
-            "10": 10,
-            "100": 100,
-            "all": len(base_ood),
-        }
-
-        for label, k_target in k_targets.items():
-            k_actual = min(k_target, len(base_ood))
-            chosen = _sample_k(base_ood, k_actual, _stable_seed(seed, base_cfg, f"fs{label}"))
+        base_ood_train = sorted(splits["ood_train"])
+        base_ood_test = sorted(splits["ood_test"])  # never changes across few-shot variants
+
+        # Build a single deterministic ordering of ood_train so that few-shot sets
+        # are guaranteed to be progressive subsets: fs1 ⊂ fs10 ⊂ fs100 ⊂ fsall
+        ordering_seed = _stable_seed(seed, base_cfg, "fsordering")
+        ordered_ood_train = list(base_ood_train)
+        rng = random.Random(ordering_seed)
+        rng.shuffle(ordered_ood_train)
+
+        for label, k in [("1", 1), ("10", 10), ("100", 100)]:
+            k_actual = min(k, len(ordered_ood_train))
+            chosen = ordered_ood_train[:k_actual]
             chosen_set = set(chosen)
 
-            fs_train = sorted(set(base_train) | chosen_set)
-            fs_id_test = list(base_id_test)
-            fs_ood = sorted(set(base_ood) - chosen_set)
-
             fs_cfg_name = f"{base_cfg}__fs{label}"
             out[fs_cfg_name] = {
-                "train": fs_train,
-                "id_test": fs_id_test,
-                "ood_test": fs_ood,
+                "train": sorted(set(base_train) | chosen_set),
+                "id_test": list(base_id_test),
+                "ood_train": sorted(set(base_ood_train) - chosen_set),
+                "ood_test": list(base_ood_test),  # unchanged
             }
 
+        # fsall: move all ood_train into train; ood_test still unchanged
+        fs_cfg_name = f"{base_cfg}__fsall"
+        out[fs_cfg_name] = {
+            "train": sorted(set(base_train) | set(base_ood_train)),
+            "id_test": list(base_id_test),
+            "ood_train": [],  # all moved to train
+            "ood_test": list(base_ood_test),  # unchanged
+        }
+
     return out
 
 
@@ -79,9 +86,12 @@ def write_configs_from_split_map(
         cfg_dir.mkdir(parents=True, exist_ok=True)
         train_n = len(splits["train"])
         id_n = len(splits["id_test"])
+        ood_train_n = len(splits.get("ood_train", []))
         ood_n = len(splits["ood_test"])
-        print(f"[make_configs] -> {cfg_name} (train={train_n}, id_test={id_n}, ood_test={ood_n})")
-        for split in ("train", "id_test", "ood_test"):
+        print(f"[make_configs] -> {cfg_name} (train={train_n}, id_test={id_n}, ood_train={ood_train_n}, ood_test={ood_n})")
+        for split in ("train", "id_test", "ood_train", "ood_test"):
+            if split not in splits or not splits[split]:
+                continue
             write_rows_to_parquet(
                 rows_from_files(splits[split], light_index, images_root),
                 cfg_dir,
@@ -97,6 +107,8 @@ def main() -> None:
     ap.add_argument("--master_dir", required=True, help="hf_repo/data/master (parquet shards)")
     ap.add_argument("--out_configs_dir", required=True, help="hf_repo/data/configs")
     ap.add_argument("--train_ratio", type=float, default=0.9)
+    ap.add_argument("--ood_train_ratio", type=float, default=0.7,
+                    help="Fraction of OOD pool used for ood_train (few-shot source); remainder is ood_test")
     ap.add_argument("--seed", type=int, default=42)
     ap.add_argument("--rows_per_shard", type=int, default=16)
     ap.add_argument("--scan_batch_size", type=int, default=32)
@@ -117,6 +129,7 @@ def main() -> None:
         rows_per_shard=args.rows_per_shard,
         scan_batch_size=args.scan_batch_size,
         min_pool=args.min_pool,
+        ood_train_ratio=args.ood_train_ratio,
         write_parquet=False,
     )
 
@@ -147,6 +160,13 @@ def main() -> None:
         "base_config_count": len(base_cfgs),
         "fewshot_per_base": ["fs1", "fs10", "fs100", "fsall"],
         "total_config_count": len(all_split_map),
+        "splits": ["train", "id_test", "ood_train", "ood_test"],
+        "split_notes": {
+            "train": "ID training set (90% of ID pool)",
+            "id_test": "ID test set (10% of ID pool, fixed across all configs)",
+            "ood_train": "OOD training pool (70% of OOD pool); source for few-shot images",
+            "ood_test": "OOD test set (30% of OOD pool, fixed across all configs including few-shot variants)",
+        },
         "base_configs": base_cfgs,
         "fewshot_configs": fs_cfgs,
         "all_configs": sorted(all_split_map.keys()),

tools/make_fewshot_configs.py

@@ -204,7 +204,15 @@ def materialize_fewshot_config(
 ) -> Dict[str, object]:
     """
     Creates <base_cfg>__fs<k> under data/configs by moving k images
-    from base ood_test into train.
+    from base ood_train into train.
+
+    ood_test is NEVER modified and stays identical across the base config
+    and all few-shot variants, ensuring consistent OOD evaluation.
+
+    Few-shot images are drawn from a single deterministic ordering of ood_train
+    (seeded by base_cfg name only, not k), so that
+      fs1_images ⊂ fs10_images ⊂ fs100_images ⊂ fsall_images.
+    Pass k=-1 (or k >= len(ood_train)) to move all ood_train images (fsall).
     """
     data_dir = repo_root / "data"
     cfgs_dir = data_dir / "configs"
@@ -214,7 +222,7 @@ def materialize_fewshot_config(
     if not base_dir.exists():
         raise FileNotFoundError(f"Base config not found: {base_dir}")
 
-    out_cfg = f"{base_cfg}__fs{k}"
+    out_cfg = f"{base_cfg}__fs{k}" if k >= 0 else f"{base_cfg}__fsall"
     out_dir = cfgs_dir / out_cfg
 
     if out_dir.exists():
@@ -223,62 +231,55 @@ def materialize_fewshot_config(
     # Load filename sets from base config
     train_files = _read_filenames_from_split(base_dir, "train")
     id_test_files = _read_filenames_from_split(base_dir, "id_test")
-    ood_test_files = _read_filenames_from_split(base_dir, "ood_test")
-
-    # Optional density splits
-    ood_same_files = _read_filenames_from_split(base_dir, "ood_same_density")
-    ood_diff_files = _read_filenames_from_split(base_dir, "ood_diff_density")
-    has_density = bool(ood_same_files) or bool(ood_diff_files)
-
-    # Choose few-shot images from OOD pool (the official ood_test split)
-    fs_seed = _stable_seed(seed, base_cfg, k)
-    chosen = _sample_k_from_set(ood_test_files, k=k, seed=fs_seed)
+    ood_train_files = _read_filenames_from_split(base_dir, "ood_train")
+    ood_test_files = _read_filenames_from_split(base_dir, "ood_test")  # never changes
+
+    if not ood_train_files:
+        raise RuntimeError(
+            f"Base config '{base_cfg}' has no ood_train split. "
+            "Regenerate base configs with the updated make_configs.py."
+        )
+
+    # Build a deterministic ordering seeded only on (seed, base_cfg) so that
+    # all k values draw progressive prefixes of the same list.
+    ordering_seed = _stable_seed(seed, base_cfg, 0)  # k=0 → ordering never depends on actual k
+    ordered_ood_train = sorted(ood_train_files)  # stable base ordering
+    rng = random.Random(ordering_seed)
+    rng.shuffle(ordered_ood_train)
+
+    # Determine how many to move
+    k_actual = len(ordered_ood_train) if k < 0 else min(k, len(ordered_ood_train))
+    chosen = ordered_ood_train[:k_actual]
     chosen_set = set(chosen)
 
-    # Move chosen from ood -> train
+    # Move chosen from ood_train -> train; ood_test is completely untouched
     new_train = set(train_files) | chosen_set
     new_id_test = set(id_test_files)
-    new_ood_test = set(ood_test_files) - chosen_set
+    new_ood_train = set(ood_train_files) - chosen_set
+    new_ood_test = set(ood_test_files)  # unchanged
 
-    # If density splits exist, keep them consistent by removing chosen
-    new_ood_same = set(ood_same_files) - chosen_set if ood_same_files else set()
-    new_ood_diff = set(ood_diff_files) - chosen_set if ood_diff_files else set()
-
-    if has_density:
-        # prefer recomputing ood_test from density buckets to guarantee consistency
-        if new_ood_same or new_ood_diff:
-            new_ood_test = (new_ood_same | new_ood_diff)
-        else:
-            # density buckets missing but has_density True shouldn't happen; keep new_ood_test as computed
-            pass
-
-    # Build split membership map
+    # Build split membership map (skip empty splits)
     split_to_files: Dict[str, Set[str]] = {
         "train": new_train,
         "id_test": new_id_test,
         "ood_test": new_ood_test,
     }
-    if has_density:
-        # only write if base had them
-        if ood_same_files:
-            split_to_files["ood_same_density"] = new_ood_same
-        if ood_diff_files:
-            split_to_files["ood_diff_density"] = new_ood_diff
+    if new_ood_train:
+        split_to_files["ood_train"] = new_ood_train
 
     # Write protocol/manifest
     out_dir.mkdir(parents=True, exist_ok=True)
     proto = {
         "base_config": base_cfg,
-        "fewshot_k": k,
+        "fewshot_k": k_actual,
         "seed": seed,
-        "fewshot_seed": fs_seed,
-        "moved_from_ood_to_train": sorted(chosen),
+        "ordering_seed": ordering_seed,
+        "moved_from_ood_train_to_train": sorted(chosen),
         "counts": {
             "train": len(new_train),
             "id_test": len(new_id_test),
+            "ood_train": len(new_ood_train),
             "ood_test": len(new_ood_test),
-            "ood_same_density": len(new_ood_same) if "ood_same_density" in split_to_files else None,
-            "ood_diff_density": len(new_ood_diff) if "ood_diff_density" in split_to_files else None,
         },
     }
     (out_dir / "protocol.json").write_text(json.dumps(proto, indent=2))
@@ -305,6 +306,7 @@ def materialize_fewshot_config(
 
 
 def parse_fewshot_ks(s: str) -> List[int]:
+    """Parse comma-separated k values. Use 'all' or -1 to indicate fsall (all ood_train)."""
     s = s.strip()
     if not s:
         return []
@@ -313,7 +315,10 @@ def parse_fewshot_ks(s: str) -> List[int]:
     for p in parts:
         if not p:
             continue
-        out.append(int(p))
+        if p.lower() == "all":
+            out.append(-1)
+        else:
+            out.append(int(p))
     return out
 
 

tools/make_master_configs.py

@@ -89,6 +89,7 @@ def collect_base_splits(
     train_ratio: float,
     seed: int,
     min_pool: int,
+    ood_train_ratio: float = 0.7,
 ) -> Dict[str, Dict[str, List[str]]]:
     meta = meta.copy()
     meta["biome"] = meta["biome"].astype(str).str.upper().str.strip()
@@ -103,15 +104,18 @@ def collect_base_splits(
 
     def add_single_config(cfg_name: str, id_pool: List[str], ood_pool: List[str]) -> None:
         train_files, id_test_files = split_pool(id_pool, train_ratio, seed)
-        ood_test_files = list(ood_pool)
+        # Split OOD pool 70:30 into ood_train (for few-shot) and ood_test (held-out eval)
+        ood_train_files, ood_test_files = split_pool(ood_pool, ood_train_ratio, seed)
 
         ensure_min(cfg_name + ":train", train_files, min_pool)
         ensure_min(cfg_name + ":id_test", id_test_files, 10)
+        ensure_min(cfg_name + ":ood_train", ood_train_files, 10)
         ensure_min(cfg_name + ":ood_test", ood_test_files, 10)
 
         split_map[cfg_name] = {
             "train": sorted(train_files),
             "id_test": sorted(id_test_files),
+            "ood_train": sorted(ood_train_files),
             "ood_test": sorted(ood_test_files),
         }
 
@@ -158,9 +162,12 @@ def write_config_from_split_map(
         cfg_dir.mkdir(parents=True, exist_ok=True)
         train_n = len(splits["train"])
         id_n = len(splits["id_test"])
+        ood_train_n = len(splits.get("ood_train", []))
         ood_n = len(splits["ood_test"])
-        print(f"[make_master_configs] -> {cfg_name} (train={train_n}, id_test={id_n}, ood_test={ood_n})")
-        for split in ("train", "id_test", "ood_test"):
+        print(f"[make_master_configs] -> {cfg_name} (train={train_n}, id_test={id_n}, ood_train={ood_train_n}, ood_test={ood_n})")
+        for split in ("train", "id_test", "ood_train", "ood_test"):
+            if split not in splits or not splits[split]:
+                continue
             write_rows_to_parquet(
                 rows_from_files(splits[split], light_index, images_root),
                 cfg_dir,
@@ -179,6 +186,7 @@ def build_master_configs(
     rows_per_shard: int,
     scan_batch_size: int,
     min_pool: int,
+    ood_train_ratio: float = 0.7,
     write_parquet: bool = True,
 ) -> Dict[str, Dict[str, List[str]]]:
     meta_path = src_root / "metadata.csv"
@@ -189,7 +197,7 @@ def build_master_configs(
         raise FileNotFoundError(f"world_images/ not found at: {images_root}")
 
     meta = pd.read_csv(meta_path)
-    split_map = collect_base_splits(meta, train_ratio=train_ratio, seed=seed, min_pool=min_pool)
+    split_map = collect_base_splits(meta, train_ratio=train_ratio, seed=seed, min_pool=min_pool, ood_train_ratio=ood_train_ratio)
 
     if write_parquet:
         print("[make_master_configs] Building lightweight master index...")
@@ -212,6 +220,8 @@ def main() -> None:
     ap.add_argument("--master_dir", required=True, help="hf_repo/data/master (parquet shards)")
     ap.add_argument("--out_configs_dir", required=True, help="hf_repo/data/configs")
     ap.add_argument("--train_ratio", type=float, default=0.9)
+    ap.add_argument("--ood_train_ratio", type=float, default=0.7,
+                    help="Fraction of OOD pool used for ood_train (few-shot source); remainder is ood_test")
     ap.add_argument("--seed", type=int, default=42)
     ap.add_argument("--rows_per_shard", type=int, default=16)
     ap.add_argument("--scan_batch_size", type=int, default=32)
@@ -228,12 +238,14 @@ def main() -> None:
         rows_per_shard=args.rows_per_shard,
         scan_batch_size=args.scan_batch_size,
         min_pool=args.min_pool,
+        ood_train_ratio=args.ood_train_ratio,
         write_parquet=True,
     )
 
     manifest = {
         "base_config_count": len(split_map),
         "base_configs": sorted(split_map.keys()),
+        "splits": ["train", "id_test", "ood_train", "ood_test"],
         "split_sizes": {
             cfg: {split: len(files) for split, files in splits.items()}
             for cfg, splits in split_map.items()