Split dataset configs and metadata

.gitattributes

@@ -1,2 +1,2 @@
-metadata.jsonl filter=lfs diff=lfs merge=lfs -text
+*.jsonl filter=lfs diff=lfs merge=lfs -text
 images/* filter=lfs diff=lfs merge=lfs -text

README.md

@@ -9,6 +9,15 @@ CUEBench is a neurosymbolic benchmark that emphasizes **contextual entity predic
 - **Geography / Scenario**: Urban autonomous driving across diverse traffic densities
 - **License**: CC-BY-4.0 (you may adapt if different licensing is desired)
 
+### Configurations
+
+| Config | File | Description |
+| --- | --- | --- |
+| `clue` *(default)* | `clue_metadata.jsonl` | Contextual Unobserved Entity (CLUE) frames with heavy occlusions and single-target predictions. |
+| `mep` | `mep_metadata.jsonl` | Multi-Entity Prediction (MEP) split that introduces complementary metadata and more diverse target sets. |
+
+When this dataset is viewed on Hugging Face, the dataset viewer automatically exposes a **config dropdown** so you can switch between `clue` and `mep` without leaving the UI.
+
 ## Dataset Structure
 
 ### Data Fields
@@ -20,7 +29,7 @@ CUEBench is a neurosymbolic benchmark that emphasizes **contextual entity predic
 | `target_classes` | `list[string]` | Entities inferred to exist but unobserved (occluded, off-frame, sensor failure).
 
 ### Splits
-Currently only a **train** split is defined via `metadata.jsonl`. Additional splits can be created before upload if desired (e.g., hold out 10% for validation).
+Each configuration exposes a single **train** split sourced from either `clue_metadata.jsonl` or `mep_metadata.jsonl`. Feel free to carve out validation/test subsets before upload if you need them.
 
 ### Label Taxonomy
 Representative classes include: `Car`, `Bus`, `Pedestrian`, `PickupTruck`, `MediumSizedTruck`, `Animal`, `Standing`, `VehicleWithRider`, `ConstructionSign`, `TrafficCone`, and more (~40 classes). Extend this section with the final taxonomy before publication if you want exhaustive documentation.
@@ -43,7 +52,8 @@ from datasets import load_dataset
 
 dataset = load_dataset(
   path="ishwarbb23/cuebench",
-    split="train"
+  split="train",
+  config_name="clue",  # or "mep"
 )
 ```
 
@@ -51,9 +61,15 @@ dataset = load_dataset(
 ```python
 from datasets import load_dataset
 
-dataset = load_dataset(path="cuebench", data_files="metadata.jsonl", split="train")
+dataset = load_dataset(
+  path="cuebench",
+  data_files="clue_metadata.jsonl",  # swap with "mep_metadata.jsonl" as needed
+  split="train",
+)
 ```
 
+> **Tip:** From source, you can still switch configurations by pointing `data_files` to `clue_metadata.jsonl` or `mep_metadata.jsonl`.
+
 ## Metrics
 
 `metric.py` defines **Mean Reciprocal Rank**, **Hits@K (1/3/5/10)**, and **Coverage@K (1/3/5/10)** over the predicted class rankings. When publishing to the Hugging Face Metrics Hub, expose the `compute(predictions, references)` signature so leaderboard integrations can consume it.
@@ -81,7 +97,8 @@ The dataset is currently tagged as **CC-BY-4.0** in `cuebench.py`. Update this s
    cuebench/
      README.md
      cuebench.py
-     metadata.jsonl
+     clue_metadata.jsonl
+     mep_metadata.jsonl
      metric.py           # optional metric script
      images/...          # optional or host separately
    ```

clue_metadata.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ad0c1ddeea44f53752037681a153d2409ce549a51132bcd4e4efc7807759c980
+size 689281

cuebench.py

@@ -1,4 +1,6 @@
 import json
+import os
+
 from datasets import (
     BuilderConfig,
     DatasetInfo,
@@ -11,16 +13,37 @@ from datasets import (
     Version,
 )
 
+HF_DATA_BASE_URL = "https://huggingface.co/datasets/ishwarbb23/cuebench/resolve/main"
+DATA_BASE_OVERRIDE = os.getenv("CUEBENCH_DATA_BASE_URL")
+CLUE_METADATA_FILENAME = "clue_metadata.jsonl"
+MEP_METADATA_FILENAME = "mep_metadata.jsonl"
+
+
+class CUEBenchConfig(BuilderConfig):
+    """Builder config that carries the backing metadata file."""
+
+    def __init__(self, *, data_files=None, **kwargs):
+        super().__init__(**kwargs)
+        self.data_files = data_files or {"train": CLUE_METADATA_FILENAME}
+
+
 class CUEBench(GeneratorBasedBuilder):
     VERSION = Version("1.0.0")
     BUILDER_CONFIGS = [
-        BuilderConfig(
-            name="default",
+        CUEBenchConfig(
+            name="clue",
             version=VERSION,
-            description="Contextual Unobserved Entity Benchmark leveraging autonomous driving scenes.",
-        )
+            description="Contextual Unobserved Entity (CLUE) split with occluded-entity targets.",
+            data_files={"train": CLUE_METADATA_FILENAME},
+        ),
+        CUEBenchConfig(
+            name="mep",
+            version=VERSION,
+            description="Multi-Entity Prediction (MEP) split with complementary metadata.",
+            data_files={"train": MEP_METADATA_FILENAME},
+        ),
     ]
-    DEFAULT_CONFIG_NAME = "default"
+    DEFAULT_CONFIG_NAME = "clue"
 
     def _info(self):
         return DatasetInfo(
@@ -37,25 +60,55 @@ class CUEBench(GeneratorBasedBuilder):
         )
 
     def _split_generators(self, dl_manager):
-        data_files = self.config.data_files or {"train": "metadata.jsonl"}
+        data_files = self.config.data_files or {"train": CLUE_METADATA_FILENAME}
         train_files = data_files["train"] if isinstance(data_files, dict) else data_files
-        filepath = dl_manager.download_and_extract(train_files)
-        if isinstance(filepath, list):
-            filepath = filepath[0]
-        return [SplitGenerator(name=Split.TRAIN, gen_kwargs={"filepath": filepath})]
-
-    def _generate_examples(self, filepath):
-        if isinstance(filepath, list):
-            filepath = filepath[0]
-        with open(filepath, "r", encoding="utf-8") as f:
-            for idx, line in enumerate(f):
-                example = json.loads(line)
-                image_id = example.get("aligned_id") or example.get("image_id")
-                if image_id is None:
-                    raise ValueError(f"Missing image identifier for example at line {idx}.")
-                yield idx, {
-                    "image_id": image_id,
-                    "image_path": example["image_path"],
-                    "observed_classes": example["detected_classes"],  # Already a list
-                    "target_classes": example["target_classes"],
-                }
+        if isinstance(train_files, str):
+            train_files = [train_files]
+
+        resolved_files = [self._resolve_path(file_path, dl_manager) for file_path in train_files]
+
+        return [SplitGenerator(name=Split.TRAIN, gen_kwargs={"filepaths": resolved_files})]
+
+    def _resolve_path(self, file_path, dl_manager):
+        if file_path.startswith(("http://", "https://")):
+            resolved = dl_manager.download_and_extract(file_path)
+            return resolved[0] if isinstance(resolved, list) else resolved
+
+        local_path = os.path.join(os.path.dirname(os.path.realpath(__file__)), file_path)
+        if os.path.exists(local_path):
+            return local_path
+
+        if DATA_BASE_OVERRIDE:
+            override = DATA_BASE_OVERRIDE.rstrip("/")
+            if override.startswith(("http://", "https://", "hf://")):
+                remote_path = f"{override}/{file_path}"
+                resolved = dl_manager.download_and_extract(remote_path)
+                return resolved[0] if isinstance(resolved, list) else resolved
+
+            override_candidate = os.path.join(override, file_path)
+            if os.path.exists(override_candidate):
+                return override_candidate
+
+        remote_path = f"{HF_DATA_BASE_URL}/{file_path}"
+        resolved = dl_manager.download_and_extract(remote_path)
+        return resolved[0] if isinstance(resolved, list) else resolved
+
+    def _generate_examples(self, filepaths):
+        if isinstance(filepaths, str):
+            filepaths = [filepaths]
+
+        idx = 0
+        for filepath in filepaths:
+            with open(filepath, "r", encoding="utf-8") as f:
+                for line in f:
+                    example = json.loads(line)
+                    image_id = example.get("aligned_id") or example.get("image_id")
+                    if image_id is None:
+                        raise ValueError(f"Missing image identifier for example at line {idx}.")
+                    yield idx, {
+                        "image_id": image_id,
+                        "image_path": example["image_path"],
+                        "observed_classes": example["detected_classes"],
+                        "target_classes": example["target_classes"],
+                    }
+                    idx += 1

mep_metadata.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0209facd1d858f4be3e32b3c1dd8c581ace65e70218e904079de53d6e21c933b
+size 526853