OpinionBench.py

import datasets
import pandas as pd

# Dataset metadata
_CITATION = """"""
_DESCRIPTION = """"""
_HOMEPAGE = ""
_LICENSE = ""

# Updated URLs to dynamically handle task names
_URLS = {
    "train": "https://huggingface.co/datasets/reglab/OpinionBench/resolve/main/data/automated-tasks-subsample/{task_name}/{task_name}_train.csv",
    "validation": "https://huggingface.co/datasets/reglab/OpinionBench/resolve/main/data/automated-tasks-subsample/{task_name}/{task_name}_val.csv",
    "test": "https://huggingface.co/datasets/reglab/OpinionBench/resolve/main/data/automated-tasks-subsample/{task_name}/{task_name}_test.csv",
}

TASK_NAMES = [
    "ATS-Jurisdiction", "ATS-FavorableJudgment", "Chevron-Agency", "Chevron-ChevCited", 
    "Chevron-Dec.Ov.", "Chevron-Deference", "Chevron-Outcome", "Chevron-Subject", 
    "CoA-casetyp1", "CoA-direct1", "CoA-geniss", "CoA-typeiss", 
    "DC-casetype", "DC-category", "DC-libcon", 
    "JRC-AREA1", "JRC-CERT", "JRC-REVERSD",
    "SSC-ca_disp", "SSC-ca_uscty", "SSC-death_c", "SSC-p1_persn",
    "SC-petitioner", "SC-respondent", "SC-respondentState", "SC-jurisdiction", 
    "SC-adminAction", "SC-threeJudgeFdc", "SC-caseOrigin", "SC-caseOriginState", 
    "SC-caseSource", "SC-caseSourceState", "SC-lcDisagreement", 
    "SC-certReason", "SC-lcDisposition", "SC-lcDispositionDirection", 
    "SC-issue", "SC-issueArea", "SC-decisionDirection", "SC-decisionDirectionDissent", 
    "SC-authorityDecision1", "SC-lawType", "SC-decisionType", "SC-declarationUncon", "SC-caseDisposition", 
    "SC-partyWinning", "SC-precedentAlteration", "SC-voteUnclear", "SC-majOpinWriter", "SC-majVotes", "SC-minVotes"
]

_CONFIGS = {
    task_name: {
        "description": f"{task_name} specific legal opinions",  # Dynamic description based on task name
        "features": {
            "idx": datasets.Value("string"),
            "Citation": datasets.Value("string"),
            "Full Case Name": datasets.Value("string"),
            "Opinion Text": datasets.Value("string"),
            "Numerical Label": datasets.Value("string"),  # Will be optional for some tasks
            #"Text Label": datasets.Value("string"), # Will be optional for some tasks
            #"DC Numerical Label": datasets.Value("string")
            #"Syllabus": datasets.Value("string") # Will be optional for some tasks
        },
    }
    for task_name in TASK_NAMES
}


class LongConLDataset(datasets.GeneratorBasedBuilder):
    """Legal opinion classification dataset for LongConL tasks"""

    def _info(self):
        """Return dataset information."""
        features = datasets.Features({
            "idx": datasets.Value("string"),
            "Citation": datasets.Value("string"),
            "Full Case Name": datasets.Value("string"),
            "Opinion Text": datasets.Value("string"),
            "Numerical Label": datasets.Value("string"),  # Will be optional for some tasks
            #"Text Label": datasets.Value("string"), # Will be optional for some tasks
            #"DC Numerical Label": datasets.Value("string")
            #"Syllabus": datasets.Value("string") # Will be optional for some tasks
        })
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=features,
            homepage=_HOMEPAGE,
            citation=_CITATION,
            license=_LICENSE,
        )

    def _split_generators(self, dl_manager):
        task_name = self.config.name
        # Replace hyphens with underscores if your URLs use underscores
    
        urls = {
            key: val.format(task_name=task_name) for key, val in _URLS.items()
        }

        downloaded_files = dl_manager.download_and_extract(urls)
    
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={"file_path": downloaded_files["train"]},
            ),
            datasets.SplitGenerator(
                name=datasets.Split.VALIDATION,
                gen_kwargs={"file_path": downloaded_files["validation"]},
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                gen_kwargs={"file_path": downloaded_files["test"]},
            ),
        ]


    def _generate_examples(self, file_path):
        """Generate examples from the dataset CSV."""
        data = pd.read_csv(file_path)
        print("Loaded file path:", file_path)
        print("Shape of data:", data.shape)
        print(data.head())  # Display first few rows
        data_dict = data.to_dict(orient="records")
        print(f"Number of examples to generate: {len(data_dict)}")

        for id_, row in enumerate(data_dict):
            yield id_, {
                "idx": row["idx"],
                "Citation": row["Citation"],
                "Full Case Name": row["Full Case Name"],
                "Opinion Text": row["Opinion Text"],
                "Numerical Label": row.get("Numerical Label", None),  # Use .get() to handle missing keys
                #"Text Label": row["Text Label"],
                #"DC Numerical Label": row["DC Numerical Label"]
                #"Syllabus": row["Syllabus"]
            }

    # Use a dynamic config
    BUILDER_CONFIGS = [
        datasets.BuilderConfig(name=task_name, version=datasets.Version("1.0.0"), description=task_name)
        for task_name in TASK_NAMES
    ]