README.md

metadata

dataset_info:
  - config_name: default
    features:
      - name: utterance
        dtype: string
      - name: label
        dtype: int64
    splits:
      - name: train
        num_bytes: 1315631
        num_examples: 13784
      - name: test
        num_bytes: 68315
        num_examples: 700
    download_size: 581872
    dataset_size: 1383946
  - config_name: intents
    features:
      - name: id
        dtype: int64
      - name: name
        dtype: string
      - 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: 270
        num_examples: 7
    download_size: 3121
    dataset_size: 270
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: test
        path: data/test-*
  - config_name: intents
    data_files:
      - split: intents
        path: intents/intents-*
task_categories:
  - text-classification
language:
  - ru

Russian snips

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

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

Source

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

"""Convert snips dataset to autointent internal format and scheme."""  # noqa: INP001

from datasets import Dataset as HFDataset
from datasets import load_from_disk

from autointent import Dataset
from autointent.schemas import Intent, Sample


def _extract_intents_data(split: HFDataset) -> tuple[dict[str, int], list[Intent]]:
    intent_names = sorted(split.unique("intent"))
    name_to_id = dict(zip(intent_names, range(len(intent_names)), strict=False))
    return name_to_id, [Intent(id=i, name=name) for i, name in enumerate(intent_names)]


def convert_rusnips(snips_split: HFDataset, name_to_id: dict[str, int]) -> list[Sample]:
    """Convert one split into desired format."""
    n_classes = len(name_to_id)

    classwise_utterance_records = [[] for _ in range(n_classes)]

    for sample in snips_split:
        txt, name = sample["text"], sample["intent"]
        intent_id = name_to_id[name]
        target_list = classwise_utterance_records[intent_id]
        target_list.append({"utterance": txt, "label": intent_id})

    return [
        Sample(**sample) for samples_from_one_class in classwise_utterance_records for sample in samples_from_one_class
    ]


if __name__ == "__main__":
    # ! git clone git@github.com:LadaNikitina/Snips data/RuSnips
    # ! rm -rf data/RuSnips/.git
    rusnips = load_from_disk("data/RuSnips")

    name_to_id, intents_data = _extract_intents_data(rusnips["train"])

    train_samples = convert_rusnips(rusnips["train"], name_to_id=name_to_id)
    test_samples = convert_rusnips(rusnips["test"], name_to_id=name_to_id)

    dataset = Dataset.from_dict({"train": train_samples, "test": test_samples, "intents": intents_data})