README.md

metadata

dataset_info:
  - config_name: default
    features:
      - name: utterance
        dtype: string
      - name: label
        dtype: int64
    splits:
      - name: train
        num_bytes: 1422940
        num_examples: 15250
      - name: validation
        num_bytes: 286013
        num_examples: 3100
      - name: test
        num_bytes: 517027
        num_examples: 5500
    download_size: 1202523
    dataset_size: 2249306.8852459015
  - config_name: intents
    features:
      - name: id
        dtype: int64
      - name: name
        dtype: 'null'
      - name: tags
        sequence: 'null'
      - name: regexp_full_match
        sequence: 'null'
      - name: regexp_partial_match
        sequence: 'null'
      - name: description
        dtype: 'null'
    splits:
      - name: intents
        num_bytes: 3000
        num_examples: 150
    download_size: 3651
    dataset_size: 3000
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*
      - split: test
        path: data/test-*
  - config_name: intents
    data_files:
      - split: intents
        path: intents/intents-*
task_categories:
  - text-classification
language:
  - ru

Russian clinc150

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

clinc150_ru = Dataset.from_hub("AutoIntent/clinc150_ru")

Source

This dataset is taken from private github repository LadaNikitina/clinc150 and formatted with our AutoIntent Library:

from autointent import Dataset
from autointent.schemas import Sample
from datasets import load_from_disk, Dataset as HFDataset


def convert_ruclinc150(clinc150_train: HFDataset, ood_index=42):
    all_labels = sorted(clinc150_train.unique("intent"))
    assert all_labels == list(range(151))

    in_domain_samples = clinc150_train.filter(lambda x: x["intent"] != ood_index)
    oos_samples = clinc150_train.filter(lambda x: x["intent"] == ood_index)

    classwise_samples = [[] for _ in range(150)]

    for batch in in_domain_samples.iter(batch_size=16, drop_last_batch=False):
        for txt, intent_id in zip(batch["text"], batch["intent"], strict=False):
            intent_id -= int(intent_id > ood_index)
            target_list = classwise_samples[intent_id]
            target_list.append({"utterance": txt, "label": intent_id})

    train_samples = [sample for samples_from_one_class in classwise_samples for sample in samples_from_one_class]
    oos_samples = [{"utterance": txt} for txt in oos_samples["text"]]

    return [Sample(**sample) for sample in train_samples + oos_samples]

if __name__ == "__main__":
    # git clone git@github.com:LadaNikitina/clinc150 data/RuClinc150
    # rm -rf data/RuClinc150/.git

    clinc150 = load_from_disk("data/RuClinc150")
    train_samples = convert_ruclinc150(clinc150["train"])
    val_samples = convert_ruclinc150(clinc150["validation"])
    test_samples = convert_ruclinc150(clinc150["test"])

    clinc150_converted = Dataset.from_dict(
        {"train": train_samples, "validation": val_samples, "test": test_samples}
    )