convert_datasets_to_json.py

import subprocess
import urllib

import typer
from datasets import (DatasetDict, get_dataset_config_names,
                      get_dataset_split_names, load_dataset)
from huggingface_hub import list_datasets

app = typer.Typer()

# These datasets do not agree with the GitHub sources due to small
# inconsistencies in `gem_id` conventions.
# Otherwise, they are identical and we skip then from the validation step.
validation_blocklist = [
    "dart_test",
    "schema_guided_dialog_test",
    "web_nlg",  # Prefix for multiple datasets
    "wiki_auto",  # Prefix for multiple datasets
]


def convert(dataset_id: str):
    dataset_name = dataset_id.split("/")[-1]
    configs = get_dataset_config_names(dataset_id)
    skipped_validation = []

    for config in configs:
        typer.echo(dataset_id)
        raw_datasets = load_dataset(dataset_id, name=config)
        datasets_to_convert = DatasetDict()

        for split, dataset in raw_datasets.items():
            if split not in ["train", "validation"]:
                datasets_to_convert[split] = dataset

        for split, dataset in datasets_to_convert.items():
            columns_to_keep = ["gem_id", "gem_parent_id", "target"]
            # dataset = load_dataset(dataset_id, name=config, split=split)
            # For non-train splits, it seems we use the references column as the target
            if "train" not in split:
                dataset = dataset.map(lambda x: {"target": x["references"]})
            else:
                # Wrap references in list to match GEM schema
                dataset = dataset.map(lambda x: {"target": [x["target"]]})
            # Delete unused columns
            # The test split doesn't have a parent ID
            if split == "test":
                columns_to_keep.remove("gem_parent_id")
            # The `datasets` JSON serializer is buggy - use `pandas` for now
            df = dataset.to_pandas()
            # Exclude dummy config names for comparison with GitHub source dataset
            if config in ["default", "xsum", "totto"]:
                reference_name = f"{dataset_name}_{split}"
            else:
                reference_name = f"{dataset_name}_{config}_{split}"
            df[columns_to_keep].to_json(f"{reference_name}.json", orient="records")
            # Exclude known datasets from validation
            do_validation = list(filter(reference_name.startswith, validation_blocklist)) == []
            if do_validation:
                validated = validate(reference_name)
                if not validated:
                    skipped_validation.append(reference_name)

    typer.echo(f"Skipped validation for {skipped_validation}")


def validate(reference_name: str):
    # Download original references from GitHub repo
    url = f"https://github.com/GEM-benchmark/GEM-metrics/releases/download/data/{reference_name}.json"
    try:
        urllib.request.urlretrieve(
            url,
            f"github_references/{reference_name}.json",
        )
    except Exception:
        typer.echo(f"⛔ Could not download {reference_name} dataset from GitHub. Skipping validation ...")
        return False
    # Run diff - requires `jq`
    process = subprocess.run(
        f"diff <(jq --sort-keys . {reference_name}.json) <(jq --sort-keys . ./github_references/{reference_name}.json)",
        shell=True,
        stdout=subprocess.PIPE,
        executable="/bin/bash",
    )
    if process.stdout:
        # typer.echo(process.stdout)
        raise ValueError(f"❌ Validation failed for {reference_name}! New and original references do not agree 😭")
    else:
        typer.echo(f"✅ Validation successful for {reference_name}!")
        return True


@app.command()
def main():
    all_datasets = list_datasets()
    # Filter for GEM datasets
    gem_datasets = [dataset for dataset in all_datasets if dataset.id.startswith("GEM/")]
    # Filter for blocklist - currently exclude all datasets not found on GitHUb release
    blocklist = [
        "ART",
        "mlb_data_to_text",
        "OrangeSum",
        "split_and_rephrase",
        "wiki_cat_sum",
        "viggo",
        "CrossWOZ",
        "RiSAWOZ",
        "indonlg",
        "squad_v2",
        "BiSECT",
        "surface_realisation_st_2020",
        "SciDuet",
        "cochrane-simplification",
        "turku_paraphrase_corpus",
        "turku_hockey_data2text",
        "sportsett_basketball",
        "Taskmaster",
        "wiki_lingua",
        "SIMPITIKI",
        "conversational_weather",
        "RotoWire_English-German",
        "dstc10_track2_task2",
        "opusparcus",
        "xlsum",
        "references",
    ]
    blocklist = ["GEM/" + dataset for dataset in blocklist]
    gem_datasets = [dataset for dataset in gem_datasets if dataset.id not in blocklist]
    for dataset in gem_datasets:
        typer.echo(f"Converting {dataset.id} ...")
        convert(dataset.id)
    typer.echo(f"🥳 All datasets converted!")


if __name__ == "__main__":
    app()