Add non-data

.gitignore

@@ -0,0 +1,3 @@
+annotated
+outcomes
+.env

README.md

@@ -1,3 +1,11 @@
 ---
 license: cc-by-sa-4.0
 ---
+## Health Insurance Appeal Adjudication Benchmark
+
+
+This data repository houses manually labeled, and pseudo-labeled background spans corresponding to real external appeal adjudications for coverage denials from U.S. health insurance cases. The data is referenced by and described in a more general work documented here: [https://github.com/TPAFS/hicric](https://github.com/TPAFS/hicric).
+
+
+## Contact
+For questions or comments, please reach out to `[email protected]`.`

construct.py

@@ -0,0 +1,23 @@
+from datasets import DatasetDict, load_dataset
+
+
+def construct_hf_dataset(repo_name: str):
+    """Construct a HF DatasetDict class from the HICRIC outcome data."""
+
+    # Load the datasets from JSONL files
+    train_jsonl_path = "outcomes/train_backgrounds_suff.jsonl"
+    test_jsonl_path = "outcomes/test_backgrounds_suff.jsonl"
+    train_dataset = load_dataset("json", data_files=train_jsonl_path, split="train")
+    test_dataset = load_dataset("json", data_files=test_jsonl_path, split="train")
+
+    # Create a DatasetDict to combine both splits
+    dataset = DatasetDict({"train": train_dataset, "test": test_dataset})
+
+    # Save each sub-directory dataset as a separate dataset within a DatasetDict
+    dataset.push_to_hub(repo_name, private=True)
+
+    return None
+
+
+if __name__ == "__main__":
+    construct_hf_dataset("mike-persius/imr-appeals")