Add organism-specific 3_species_cls subsets

README.md

@@ -100,6 +100,94 @@ dataset_info:
     num_examples: 600
   download_size: 996782
   dataset_size: 1243475
+- config_name: binary_eukaryote
+  features:
+  - name: id
+    dtype: int64
+  - name: subtask
+    dtype: string
+  - name: question
+    dtype: string
+  - name: options
+    list: string
+  - name: answer_index
+    dtype: int64
+  - name: answer_text
+    dtype: string
+  - name: metadata
+    dtype: string
+  splits:
+  - name: train
+    num_bytes: 198070
+    num_examples: 170
+  download_size: 92539
+  dataset_size: 198070
+- config_name: binary_prokaryote
+  features:
+  - name: id
+    dtype: int64
+  - name: subtask
+    dtype: string
+  - name: question
+    dtype: string
+  - name: options
+    list: string
+  - name: answer_index
+    dtype: int64
+  - name: answer_text
+    dtype: string
+  - name: metadata
+    dtype: string
+  splits:
+  - name: train
+    num_bytes: 174961
+    num_examples: 150
+  download_size: 84047
+  dataset_size: 174961
+- config_name: mcq_eukaryote
+  features:
+  - name: id
+    dtype: int64
+  - name: subtask
+    dtype: string
+  - name: question
+    dtype: string
+  - name: options
+    list: string
+  - name: answer_index
+    dtype: int64
+  - name: answer_text
+    dtype: string
+  - name: metadata
+    dtype: string
+  splits:
+  - name: train
+    num_bytes: 681920
+    num_examples: 177
+  download_size: 567585
+  dataset_size: 681920
+- config_name: mcq_prokaryote
+  features:
+  - name: id
+    dtype: int64
+  - name: subtask
+    dtype: string
+  - name: question
+    dtype: string
+  - name: options
+    list: string
+  - name: answer_index
+    dtype: int64
+  - name: answer_text
+    dtype: string
+  - name: metadata
+    dtype: string
+  splits:
+  - name: train
+    num_bytes: 599809
+    num_examples: 165
+  download_size: 495373
+  dataset_size: 599809
 configs:
 - config_name: binary
   data_files:
@@ -121,6 +209,22 @@ configs:
   data_files:
   - split: train
     path: mcq_600/train-*
+- config_name: binary_eukaryote
+  data_files:
+  - split: train
+    path: binary_eukaryote/train-*
+- config_name: binary_prokaryote
+  data_files:
+  - split: train
+    path: binary_prokaryote/train-*
+- config_name: mcq_eukaryote
+  data_files:
+  - split: train
+    path: mcq_eukaryote/train-*
+- config_name: mcq_prokaryote
+  data_files:
+  - split: train
+    path: mcq_prokaryote/train-*
 ---
 
 # GenomeQA
@@ -174,6 +278,10 @@ headline score is needed.
 | `mcq` | 2,600 | Full benchmark in four-option multiple-choice format. |
 | `binary_600` | 600 | Lower-cost binary subset with 100 examples from each of the six subtasks. |
 | `mcq_600` | 600 | Lower-cost MCQ subset with 100 examples from each of the six subtasks. |
+| `binary_eukaryote` | 170 | Binary `3_species_cls` questions whose target organism is eukaryotic. |
+| `binary_prokaryote` | 150 | Binary `3_species_cls` questions whose target organism is prokaryotic. |
+| `mcq_eukaryote` | 177 | MCQ `3_species_cls` questions whose target organism is eukaryotic. |
+| `mcq_prokaryote` | 165 | MCQ `3_species_cls` questions whose target organism is prokaryotic. |
 | `eukaryote_cls` | 1,000 | Auxiliary four-option eukaryotic taxonomy set stored separately from the original 5,200-example benchmark. |
 
 The `binary_600` and `mcq_600` configurations were sampled without replacement using seed 42 from source revision
@@ -182,6 +290,13 @@ shuffled so the rows are not stored in subtask blocks. These subsets are balance
 by every internal label or question dimension. See
 [`scripts/create_600_subsets.py`](scripts/create_600_subsets.py) for the reproducible generation script.
 
+The organism-specific configurations contain only rows from the `3_species_cls` subtask and preserve their source
+order. The target organism is read from `metadata`: `true` for binary `1_identity` and MCQ `basic_class` questions,
+`target` for MCQ `reverse_search` questions, and `minority` for MCQ `odd_one_out` questions because the minority class
+is the sequence the model must identify. Eukaryote and prokaryote subsets are disjoint within each answer format;
+virus rows are not included. See
+[`scripts/create_organism_subsets.py`](scripts/create_organism_subsets.py) for the reproducible filtering script.
+
 ## Data fields
 
 | Field | Description |
@@ -202,6 +317,7 @@ from datasets import load_dataset
 
 binary = load_dataset("HuggingFaceBio/genomeqa", "binary", split="train")
 representative_mcq = load_dataset("HuggingFaceBio/genomeqa", "mcq_600", split="train")
+eukaryote_mcq = load_dataset("HuggingFaceBio/genomeqa", "mcq_eukaryote", split="train")
 ```
 
 ## Citation

binary_eukaryote/train-00000-of-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:59799e1374c27c6224926a77e4419c8f09e94cc89e62034e8c7c30ee8f19d5f1
+size 92539

binary_prokaryote/train-00000-of-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:30a44db3c5da3c2da566c4b4302ebbb38f2c85070bf65003edcfbdd5485786a7
+size 84047

mcq_eukaryote/train-00000-of-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a9e66c503bf3314196c8c34f6160f0ad7db4f47b7632be0517b1730f4875252b
+size 567585

mcq_prokaryote/train-00000-of-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a6aa8db7c3281415330257d8729288cb999b772f7fc8a29e0487d521b07826bb
+size 495373

scripts/create_organism_subsets.py

@@ -0,0 +1,57 @@
+"""Create GenomeQA 3_species_cls subsets grouped by target organism."""
+
+import argparse
+import json
+from pathlib import Path
+
+from datasets import load_dataset
+
+
+DATASET_ID = "HuggingFaceBio/genomeqa"
+SOURCE_REVISION = "d2213306c6cf5b4568a504f87f86ea50b03eeced"
+SOURCE_CONFIGS = ("binary", "mcq")
+ORGANISMS = ("eukaryote", "prokaryote")
+SUBTASK = "3_species_cls"
+ORGANISM_KEY_BY_DIMENSION = {
+    "1_identity": "true",
+    "basic_class": "true",
+    "reverse_search": "target",
+    "odd_one_out": "minority",
+}
+
+
+def get_organism(metadata: str) -> str:
+    metadata_dict = json.loads(metadata)
+    dimension = metadata_dict["dim"]
+    return metadata_dict[ORGANISM_KEY_BY_DIMENSION[dimension]]
+
+
+def create_subsets(config: str, output_dir: Path) -> None:
+    dataset = load_dataset(DATASET_ID, config, split="train", revision=SOURCE_REVISION)
+    species_dataset = dataset.filter(lambda example: example["subtask"] == SUBTASK)
+
+    for organism in ORGANISMS:
+        subset = species_dataset.filter(
+            lambda example, organism=organism: get_organism(example["metadata"]) == organism
+        )
+        if not subset:
+            raise RuntimeError(f"No {organism} rows found in {config}")
+        if any(get_organism(metadata) != organism for metadata in subset["metadata"]):
+            raise RuntimeError(f"Generated {config}_{organism} subset contains another organism")
+
+        config_dir = output_dir / f"{config}_{organism}"
+        config_dir.mkdir(parents=True, exist_ok=True)
+        subset.to_parquet(config_dir / "train-00000-of-00001.parquet")
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--output-dir", type=Path, default=Path.cwd())
+    args = parser.parse_args()
+
+    for config in SOURCE_CONFIGS:
+        create_subsets(config, args.output_dir)
+
+
+if __name__ == "__main__":
+    main()