crossoint.py

import json
import mlcroissant as mlc
from mlcroissant._src.structure_graph.nodes.source import FileProperty
from mlcroissant.scripts import validate


# 1. Base repo
repo = mlc.FileObject(
    id="huggingface-repo",
    name="huggingface-repo",
    description="ProofWalaDataset GitHub repo",
    content_url="https://huggingface.co/datasets/amitayusht/ProofWalaDataset",
    encoding_formats=["git+https"],
    sha256="main",
)

# 2. Single FileSet that includes all .json files recursively
file_set = mlc.FileSet(
    id="all-json",
    name="all-json",
    description="All .json files across datasets and splits.",
    contained_in=["huggingface-repo"],
    includes="**/*.json",  # recursive glob
    encoding_formats=["application/json"],
)

# 3. Helper to make fields
def make_field(field_id, name, description, json_path):
    return mlc.Field(
        id=field_id,
        name=name,
        description=description,
        data_types=[mlc.DataType.TEXT],
        source=mlc.Source(
            file_set="all-json",
            extract=mlc.Extract(json_path=json_path),
        ),
    )

# 4. Standard fields
fields = [
    make_field("record/proof_id", "proof_id", "Proof ID", "training_data[*].proof_id"),
    make_field("record/goal_description", "goal_description", "Goal description", "training_data[*].goal_description"),
    make_field("record/proof_steps", "proof_steps", "Tactic steps", "training_data[*].proof_steps"),
    make_field("record/theorem_name", "theorem_name", "Theorem name", "training_data[*].theorem_name"),
    make_field("record/project_id", "project_id", "Project ID", "training_data[*].project_id"),
    make_field("record/file_path", "file_path", "File path", "training_data[*].file_path"),
]

# fields += [
#     make_field(
#         "record/start_goal",
#         "start_goal",
#         "Main goal from start_goals list",
#         "training_data[*].start_goals[*].goal",
#     ),
#     make_field(
#         "record/start_hypotheses",
#         "start_hypotheses",
#         "List of hypotheses from start_goals",
#         "training_data[*].start_goals[*].hypotheses",
#     ),
#     make_field(
#         "record/end_goal",
#         "end_goal",
#         "Main goal from end_goals list",
#         "training_data[*].end_goals[*].goal",
#     ),
#     make_field(
#         "record/end_hypotheses",
#         "end_hypotheses",
#         "List of hypotheses from end_goals",
#         "training_data[*].end_goals[*].hypotheses",
#     ),
# ]


# 5. Derived fields: dataset family and split from path
fields += [
    mlc.Field(
        id="record/source_dataset",
        name="source_dataset",
        description="Dataset family (e.g. lean, coq).",
        data_types=[mlc.DataType.TEXT],
        source=mlc.Source(
            file_set="all-json",
            extract=mlc.Extract(file_property=FileProperty.filepath),
            transforms=[mlc.Transform(regex="^([^/]+)/.*")],
        ),
    ),
    mlc.Field(
        id="record/split",
        name="split",
        description="Split name (train/test/eval).",
        data_types=[mlc.DataType.TEXT],
        source=mlc.Source(
            file_set="all-json",
            extract=mlc.Extract(file_property=FileProperty.filepath),
            transforms=[mlc.Transform(regex="^[^/]+/([^/]+)/.*")],
        ),
    ),
]

# 6. RecordSet
record_set = mlc.RecordSet(
    id="proof-records",
    name="proof-records",
    description="All proof records across dataset families and splits.",
    fields=fields,
)

metadata = mlc.Metadata(
    name="ProofWalaDataset",
    description="A dataset of formal theorem-proving steps extracted from Lean, Coq, GeoCoq, MathComp, and more.",
    license="https://opensource.org/licenses/MIT",
    url="https://huggingface.co/datasets/amitayusht/ProofWalaDataset",
    version="1.0.0",
    distribution=[repo, file_set],
    record_sets=[record_set],
)


json_dict = metadata.to_json()
with open("croissant.json", "w") as f:
    json.dump(json_dict, f, indent=2)
# validate.main("croissant.json")