metadata

dataset_info:
  - config_name: default
    features:
      - name: utterance
        dtype: string
      - name: label
        sequence: int64
    splits:
      - name: train
        num_bytes: 380277965
        num_examples: 53203
      - name: validation
        num_bytes: 40200731
        num_examples: 4834
      - name: test
        num_bytes: 57450762
        num_examples: 4774
    download_size: 389851249
    dataset_size: 477929458
  - config_name: intents
    features:
      - name: id
        dtype: int64
      - name: name
        dtype: 'null'
      - name: tags
        sequence: 'null'
      - name: regex_full_match
        sequence: 'null'
      - name: regex_partial_match
        sequence: 'null'
      - name: description
        dtype: 'null'
    splits:
      - name: intents
        num_bytes: 40
        num_examples: 2
    download_size: 2862
    dataset_size: 40
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: test
        path: data/test-*
      - split: validation
        path: data/validation-*
  - config_name: intents
    data_files:
      - split: intents
        path: intents/intents-*
task_categories:
  - text-classification
language:
  - en

eurlex

This is a text classification dataset. It is intended for machine learning research and experimentation.

This dataset is obtained via formatting another publicly available data to be compatible with our AutoIntent Library.

Usage

It is intended to be used with our AutoIntent Library:

from autointent import Dataset

eurlex = Dataset.from_hub("AutoIntent/eurlex")

Source

This dataset is taken from coastalcph/multi_eurlex and formatted with our AutoIntent Library:

import datasets
from autointent import Dataset


def get_number_of_classes(ds: datasets.Dataset) -> int:
    return len(set(i for example in ds for labels in example for i in labels))


def parse(ds: datasets.Dataset, n_classes: int) -> datasets.Dataset:
    def transform(example: dict):
        return {"utterance": example["text"], "label": [int(i in example["labels"]) for i in range(n_classes)]}
    return ds.map(transform, remove_columns=ds.features.keys())


def get_low_resource_classes_mask(ds: datasets.Dataset, n_classes: int, fraction_thresh: float = 0.01) -> list[bool]:
    res = [0] * n_classes
    for sample in ds:
        for i, indicator in enumerate(sample["label"]):
            res[i] += indicator
    for i in range(n_classes):
        res[i] /= len(ds)
    return [(frac < fraction_thresh) for frac in res]


def remove_low_resource_classes(ds: datasets.Dataset, mask: list[bool]) -> list[dict]:
    res = []
    for sample in ds:
        if sum(sample["label"]) == 1 and mask[sample["label"].index(1)]:
            continue
        sample["label"] = [
            indicator for indicator, low_resource in
            zip(sample["label"], mask, strict=True) if not low_resource
        ]
        res.append(sample)
    return res


def remove_oos(ds: list[dict]):
    return [sample for sample in ds if sum(sample["label"]) != 0]


if __name__ == "__main__":
    eurlex = datasets.load_dataset("coastalcph/multi_eurlex", "en", trust_remote_code=True)

    n_classes = get_number_of_classes(eurlex["train"])

    train = parse(eurlex["train"], n_classes)
    test = parse(eurlex["test"], n_classes)
    validation = parse(eurlex["validation"], n_classes)

    mask = get_low_resource_classes_mask(train, n_classes)
    train = remove_oos(remove_low_resource_classes(train, mask))
    test = remove_oos(remove_low_resource_classes(test, mask))
    validation = remove_oos(remove_low_resource_classes(validation, mask))

    eurlex_converted = Dataset.from_dict({
        "train": train,
        "test": test,
        "validation": validation,
    })