dotcausal.py

"""HuggingFace Datasets loader for .causal knowledge graph files."""

import datasets
from datasets import DatasetInfo, Features, Value, Sequence


class CausalConfig(datasets.BuilderConfig):
    """BuilderConfig for .causal files."""

    def __init__(
        self,
        include_inferred: bool = True,
        min_confidence: float = 0.0,
        **kwargs,
    ):
        """
        Args:
            include_inferred: Include inferred triplets (default: True)
            min_confidence: Minimum confidence threshold (default: 0.0)
        """
        super().__init__(**kwargs)
        self.include_inferred = include_inferred
        self.min_confidence = min_confidence


class CausalDataset(datasets.GeneratorBasedBuilder):
    """
    HuggingFace Dataset loader for .causal knowledge graph files.

    The .causal format is a binary knowledge graph with embedded deterministic
    inference. It provides zero-hallucination fact retrieval with full provenance.

    Usage:
        from datasets import load_dataset

        # Load from local file
        ds = load_dataset("chkmie/dotcausal", data_files="knowledge.causal")

        # Load with config
        ds = load_dataset(
            "chkmie/dotcausal",
            data_files="knowledge.causal",
            include_inferred=True,
            min_confidence=0.5,
        )

    Features:
        - trigger (str): The cause/trigger entity
        - mechanism (str): The relationship type
        - outcome (str): The effect/outcome entity
        - confidence (float): Confidence score (0-1)
        - is_inferred (bool): Whether derived or explicit
        - source (str): Original source (e.g., paper)
        - provenance (list): Source triplets for inferred facts

    References:
        - PyPI: https://pypi.org/project/dotcausal/
        - GitHub: https://github.com/DT-Foss/dotcausal
        - Paper: https://doi.org/10.5281/zenodo.18326222
    """

    BUILDER_CONFIG_CLASS = CausalConfig
    BUILDER_CONFIGS = [
        CausalConfig(
            name="default",
            version=datasets.Version("1.0.0"),
            description="Load all triplets from .causal files",
        ),
        CausalConfig(
            name="explicit_only",
            version=datasets.Version("1.0.0"),
            description="Load only explicit triplets (no inferred)",
            include_inferred=False,
        ),
        CausalConfig(
            name="high_confidence",
            version=datasets.Version("1.0.0"),
            description="Load triplets with confidence >= 0.8",
            min_confidence=0.8,
        ),
    ]
    DEFAULT_CONFIG_NAME = "default"

    def _info(self):
        return DatasetInfo(
            description="""\
.causal knowledge graph dataset with embedded deterministic inference.
Each row represents a causal triplet (trigger → mechanism → outcome).
""",
            features=Features(
                {
                    "trigger": Value("string"),
                    "mechanism": Value("string"),
                    "outcome": Value("string"),
                    "confidence": Value("float32"),
                    "is_inferred": Value("bool"),
                    "source": Value("string"),
                    "provenance": Sequence(Value("string")),
                }
            ),
            homepage="https://dotcausal.com",
            license="MIT",
            citation="""\
@article{foss2026causal,
  author = {Foss, David Tom},
  title = {The .causal Format: Deterministic Inference for AI-Assisted Hypothesis Amplification},
  journal = {Zenodo},
  year = {2026},
  doi = {10.5281/zenodo.18326222}
}
""",
        )

    def _split_generators(self, dl_manager):
        """Generate splits from data files."""
        data_files = self.config.data_files

        if not data_files:
            raise ValueError(
                "No data_files specified. Use: load_dataset('chkmie/dotcausal', data_files='your_file.causal')"
            )

        # Handle different data_files formats
        if isinstance(data_files, dict):
            # {"train": ["file1.causal"], "test": ["file2.causal"]}
            splits = []
            for split_name, files in data_files.items():
                if isinstance(files, str):
                    files = [files]
                downloaded = dl_manager.download_and_extract(files)
                splits.append(
                    datasets.SplitGenerator(
                        name=split_name,
                        gen_kwargs={"filepaths": downloaded},
                    )
                )
            return splits
        elif isinstance(data_files, (list, tuple)):
            # ["file1.causal", "file2.causal"]
            downloaded = dl_manager.download_and_extract(list(data_files))
            return [
                datasets.SplitGenerator(
                    name=datasets.Split.TRAIN,
                    gen_kwargs={"filepaths": downloaded},
                )
            ]
        else:
            # Single file string
            downloaded = dl_manager.download_and_extract([data_files])
            return [
                datasets.SplitGenerator(
                    name=datasets.Split.TRAIN,
                    gen_kwargs={"filepaths": downloaded},
                )
            ]

    def _generate_examples(self, filepaths):
        """Generate examples from .causal files."""
        try:
            from dotcausal import CausalReader
        except ImportError:
            raise ImportError(
                "dotcausal package required. Install with: pip install dotcausal"
            )

        if isinstance(filepaths, str):
            filepaths = [filepaths]

        idx = 0
        for filepath in filepaths:
            reader = CausalReader(filepath)

            # Get all triplets via search
            results = reader.search("", limit=100000)

            for r in results:
                # Apply filters from config
                confidence = r.get("confidence", 1.0)
                is_inferred = r.get("is_inferred", False)

                if confidence < self.config.min_confidence:
                    continue
                if not self.config.include_inferred and is_inferred:
                    continue

                # Convert provenance to list of strings
                provenance = r.get("provenance", [])
                if not isinstance(provenance, list):
                    provenance = [str(provenance)] if provenance else []
                else:
                    provenance = [str(p) for p in provenance]

                yield idx, {
                    "trigger": r.get("trigger", ""),
                    "mechanism": r.get("mechanism", ""),
                    "outcome": r.get("outcome", ""),
                    "confidence": float(confidence),
                    "is_inferred": bool(is_inferred),
                    "source": r.get("source", ""),
                    "provenance": provenance,
                }
                idx += 1