src/tests/test_dataset_schema.py

from datetime import date
from pathlib import Path
from typing import Any, Literal

import pytest
from datasets import load_dataset
from pydantic import AfterValidator, BaseModel, BeforeValidator
from typing_extensions import Annotated

from .conftest import DATASET_NAMES
from .readme_parsing import get_tag_idx, read_frontmatter_and_body


def ensure_tuple(created: str | tuple) -> tuple:
    if isinstance(created, str):
        return tuple(created.split(", "))
    return created


def validate_sample_metadata(metadata: dict[str, Any]) -> dict[str, Any]:
    if "source-pretty" not in metadata:
        raise ValueError("'source-pretty' should be in metadata dict.")
    return metadata


class SampleSchema(BaseModel):
    text: str
    source: str
    id: str
    added: date  # date.fromisoformat
    created: Annotated[tuple[date, date], BeforeValidator(ensure_tuple)]
    license: str  # TODO: should probably be a literal
    domain: str  # TODO: convert to literal
    metadata: Annotated[dict[str, Any], AfterValidator(validate_sample_metadata)]


@pytest.mark.parametrize("dataset_name", DATASET_NAMES)
def test_sample_schema(repo_path: Path, dataset_name: str):
    """Ensure that the dataset samples follow the correct schema"""

    ds = load_dataset(
        str(repo_path.resolve()), dataset_name, split="train", streaming=True
    )
    sample = next(iter(ds))
    SampleSchema(**sample)


class FrontmatterSchema(BaseModel):
    pretty_name: str
    language: list[Literal["da"]]
    license: Literal["cc0-1.0", "other", "cc-by-sa-4.0"]


@pytest.mark.parametrize("dataset_name", DATASET_NAMES)
def test_dataset_readme(repo_path: Path, dataset_name: str):
    """tests that the dataset frontmatter and markdown follows the correct format."""

    readme = repo_path / "data" / dataset_name / f"{dataset_name}.md"

    frontmatter, body = read_frontmatter_and_body(readme)
    frontmatter_validated = FrontmatterSchema(**frontmatter)

    # ensure tags:
    tags = ["SHORT DESCRIPTION", "DESC-STATS", "DATASET PLOTS", "SAMPLE"]
    for tag in tags:
        get_tag_idx(body, tag)

    h2_headings = {line for line in body.splitlines() if line.startswith("## ")}

    if (
        frontmatter_validated.license == "other"
    ):  # ensure description of underspecified licenses
        assert "## License Information" in h2_headings

    # required headings
    req_h2_headings = ["## Dataset Description", "## Additional Information"]
    for req_h2 in req_h2_headings:
        assert req_h2 in h2_headings
    pass


@pytest.mark.parametrize("dataset_name", DATASET_NAMES)
def test_dataset_folder_structure(repo_path: Path, dataset_name: str):
    """tests that the dataset folder structure is as follows.

    dataset_name
    |- dataset_name.md
    |- dataset_name.parquet

    If there is a python file, there should at least be one called `create.py`, but there can be additional.
    """
    path = repo_path / "data" / dataset_name

    assert (path / f"{path.name}.parquet").exists()
    assert (path / f"{path.name}.md").exists()

    if any(p.name.endswith(".py") for p in path.glob("*")):
        assert (path / "create.py").exists()