Hugging Face Dataset Structurer

The Hugging Face Dataset Structurer is a Python wrapper designed to streamline the deployment of multi-config datasets to the Hugging Face Hub. This tool simplifies the process by automating the creation of dataset loading scripts, addressing gaps in the official documentation.

Installation

pip install -U hf-dataset-structurer

Quickstart

Let's demonstrate the tool's capabilities using the Portuguese NER scenario. We'll create a consolidated dataset by merging the official HuggingFace HAREM dataset entry, representing the findings of the First-HAREM meeting, with a complementary "second-HAREM" dataset, accessible here. This "second-HAREM," not initially included in the original HuggingFace efforts, brings valuable additional data. Both HAREM datasets offer two labeling schemes: DEFAULT and SELECTIVE, making them an ideal showcase for highlighting our tool's capabilities.

from datasets import load_dataset, concatenate_datasets, DatasetDict 
from hf_dataset_structurer.DatasetStructure import DatasetStructure

structurer = DatasetStructure("<<TARGET Hugging Face Dataset Name>>")

# Iterate both Labelling Schemes
for config in ["default", "selective"]:
    # Load Official HAREM Dataset
    primeiro_harem = load_dataset("harem", config)
    
    # Start Structuring Process
    structurer.add_dataset(primeiro_harem['train'], f"primeiro_harem_{config}", split="train")

# Load Second HAREM Datasets

second_harem_default = load_dataset("arubenruben/segundo_harem_default")
second_harem_selective = load_dataset("arubenruben/segundo_harem_selective")

# Notice the function used now is add_dataset_dict. A [DatasetDict](https://huggingface.co/docs/datasets/v2.15.0/en/package_reference/main_classes#datasets.DatasetDict) is a native HuggingFace object that represents a dictionary of datasets.
structurer.add_dataset_dict(second_harem_default, "segundo_harem_default")
structurer.add_dataset_dict(second_harem_selective, "segundo_harem_selective")

# Push to Hugging Face Hub
structurer.push_to_hub()

# After creating the bundle. You can append a dataset card to it.
# Create Dataset Card to describe the dataset
structurer.attach_dataset_card(
    language="pt",
    license="cc-by-4.0",
    annotations_creators=["expert-generated"],
    task_categories=["token-classification"],
    tasks_ids=["named-entity-recognition"],
    pretty_name="HAREM",
    multilinguality='monolingual'
)

API Reference

# Initializes a new instance of the DatasetStructure class.
__init__(self, repo_name: str) -> None

# Accepts a DatasetDict and a config_name and adds it to the dataset structure.
add_dataset_dict(self, dataset_dict: DatasetDict, config_name: str) -> None

# Similar to add_dataset_dict, but accepts a Dataset and a split. Internally, it creates a DatasetDict and calls add_dataset_dict.
add_dataset(self, dataset: Dataset, config_name: str, split: str = "train") -> None

# Attaches a dataset card to the dataset structure.
attach_dataset_card(self, language: str,
                    license: str,
                    annotations_creators: str,
                    task_categories: str,
                    tasks_ids: str,
                    pretty_name: str,
                    multilinguality: str = 'monolingual') -> None

# Pushes the dataset structure and dataset card to the Hugging Face Hub.
push_to_hub(self, private: bool = False) -> None

Contributing

Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.

License

MIT

Acknowledgements

This tool was developed by Ruben Almeida as part of the Project PT-Pump-Up. PT-Pump-Up is a project funded by INESC TEC and the Portuguese Government through the Fundacao para a Ciencia e a Tecnologia (FCT) that aims to build Portuguese NLP resources and tools to support the development of NLP applications for Portuguese.

References

• HF Official Docs - Structure your repository

• HF Official Docs - Dataset Card