TFDS now supports the Croissant 🥐 format! Read the documentation to know more.

Dataset Collections

View on TensorFlow.org

Run in Google Colab

View on GitHub

Download notebook

Overview

Dataset collections provide a simple way to group together an arbitrary number of existing TFDS datasets, and to perform simple operations over them.

They can be useful, for example, to group together different datasets related to the same task, or for easy benchmarking of models over a fixed number of different tasks.

Setup

To get started, install a few packages:

# Use tfds-nightly to ensure access to the latest features.
pip install -q tfds-nightly tensorflow
pip install -U conllu

Import TensorFlow and the Tensorflow Datasets package into your development environment:

import pprint

import tensorflow as tf
import tensorflow_datasets as tfds

2023-10-03 09:24:58.961730: E tensorflow/compiler/xla/stream_executor/cuda/cuda_dnn.cc:9342] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered
2023-10-03 09:24:58.961781: E tensorflow/compiler/xla/stream_executor/cuda/cuda_fft.cc:609] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered
2023-10-03 09:24:58.961817: E tensorflow/compiler/xla/stream_executor/cuda/cuda_blas.cc:1518] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered

Dataset collections provide a simple way to group together an arbitrary number of existing datasets from Tensorflow Datasets (TFDS), and to perform simple operations over them.

They can be useful, for example, to group together different datasets related to the same task, or for easy benchmarking of models over a fixed number of different tasks.

Find available dataset collections

All dataset collection builders are a subclass of tfds.core.dataset_collection_builder.DatasetCollection.

To get the list of available builders, use tfds.list_dataset_collections().

tfds.list_dataset_collections()

['longt5', 'xtreme']

Load and inspect a dataset collection

The easiest way of loading a dataset collection is to instantiate a DatasetCollectionLoader object using the tfds.dataset_collection command.

collection_loader = tfds.dataset_collection('xtreme')

Specific dataset collection versions can be loaded following the same syntax as with TFDS datasets:

collection_loader = tfds.dataset_collection('xtreme:1.0.0')

A dataset collection loader can display information about the collection:

collection_loader.print_info()

Dataset collection: xtreme
Version: 1.0.0
Description: # Xtreme Benchmark

The Cross-lingual TRansfer Evaluation of Multilingual Encoders (XTREME)
benchmark is a benchmark for the evaluation of the cross-lingual generalization
ability of pre-trained multilingual models. It covers 40 typologically diverse
languages (spanning 12 language families) and includes nine tasks that
collectively require reasoning about different levels of syntax and semantics.
The languages in XTREME are selected to maximize language diversity, coverage
in existing tasks, and availability of training data. Among these are many
under-studied languages, such as the Dravidian languages Tamil (spoken in
southern India, Sri Lanka, and Singapore), Telugu and Malayalam (spoken
mainly in southern India), and the Niger-Congo languages Swahili and Yoruba,
spoken in Africa.

For a full description of the benchmark,
see the [paper](https://arxiv.org/abs/2003.11080).

Citation:
@article{hu2020xtreme,
    author    = {Junjie Hu and Sebastian Ruder and Aditya Siddhant and Graham
                 Neubig and Orhan Firat and Melvin Johnson},
    title     = {XTREME: A Massively Multilingual Multi-task Benchmark for
                 Evaluating Cross-lingual Generalization},
    journal   = {CoRR},
    volume    = {abs/2003.11080},
    year      = {2020},
    archivePrefix = {arXiv},
    eprint    = {2003.11080}
}

The dataset loader can also display information about the datasets contained in the collection:

collection_loader.print_datasets()

The dataset collection xtreme (version: 1.0.0) contains the datasets:

 - xnli: DatasetReference(dataset_name='xtreme_xnli', namespace=None, config=None, version='1.1.0', data_dir=None, split_mapping=None)
 - pawsx: DatasetReference(dataset_name='xtreme_pawsx', namespace=None, config=None, version='1.0.0', data_dir=None, split_mapping=None)
 - pos: DatasetReference(dataset_name='xtreme_pos', namespace=None, config=None, version='1.0.0', data_dir=None, split_mapping=None)
 - ner: DatasetReference(dataset_name='wikiann', namespace=None, config=None, version='1.0.0', data_dir=None, split_mapping=None)
 - xquad: DatasetReference(dataset_name='xquad', namespace=None, config=None, version='3.0.0', data_dir=None, split_mapping=None)
 - mlqa: DatasetReference(dataset_name='mlqa', namespace=None, config=None, version='1.0.0', data_dir=None, split_mapping=None)
 - tydiqa: DatasetReference(dataset_name='tydi_qa', namespace=None, config=None, version='3.0.0', data_dir=None, split_mapping=None)
 - bucc: DatasetReference(dataset_name='bucc', namespace=None, config=None, version='1.0.0', data_dir=None, split_mapping=None)
 - tatoeba: DatasetReference(dataset_name='tatoeba', namespace=None, config=None, version='1.0.0', data_dir=None, split_mapping=None)

Loading datasets from a dataset collection

The easiest way to load one dataset from a collection is to use a DatasetCollectionLoader object's load_dataset method, which loads the required dataset by calling tfds.load.

This call returns a dictionary of split names and the corresponding tf.data.Datasets:

splits = collection_loader.load_dataset("ner")

pprint.pprint(splits)

{'test': <_PrefetchDataset element_spec={'langs': TensorSpec(shape=(None,), dtype=tf.string, name=None), 'spans': TensorSpec(shape=(None,), dtype=tf.string, name=None), 'tags': TensorSpec(shape=(None,), dtype=tf.int64, name=None), 'tokens': TensorSpec(shape=(None,), dtype=tf.string, name=None)}>,
 'train': <_PrefetchDataset element_spec={'langs': TensorSpec(shape=(None,), dtype=tf.string, name=None), 'spans': TensorSpec(shape=(None,), dtype=tf.string, name=None), 'tags': TensorSpec(shape=(None,), dtype=tf.int64, name=None), 'tokens': TensorSpec(shape=(None,), dtype=tf.string, name=None)}>,
 'validation': <_PrefetchDataset element_spec={'langs': TensorSpec(shape=(None,), dtype=tf.string, name=None), 'spans': TensorSpec(shape=(None,), dtype=tf.string, name=None), 'tags': TensorSpec(shape=(None,), dtype=tf.int64, name=None), 'tokens': TensorSpec(shape=(None,), dtype=tf.string, name=None)}>}
2023-10-03 09:25:02.792501: E tensorflow/compiler/xla/stream_executor/cuda/cuda_driver.cc:268] failed call to cuInit: CUDA_ERROR_NO_DEVICE: no CUDA-capable device is detected

load_dataset accepts the following optional parameters:

split: which split(s) to load. It accepts a single split (split="test") or a list of splits: (split=["train", "test"]). If not specified, it will load all splits for the given dataset.
loader_kwargs: keyword arguments to be passed to the tfds.load function. Refer to the tfds.load documentation for a comprehensive overview of the different loading options.

Loading multiple datasets from a dataset collection

The easiest way to load multiple datasets from a collection is to use the DatasetCollectionLoader object's load_datasets method, which loads the required datasets by calling tfds.load.

It returns a dictionary of dataset names, each one of which is associated with a dictionary of split names and the corresponding tf.data.Datasets, as in the following example:

datasets = collection_loader.load_datasets(['xnli', 'bucc'])

pprint.pprint(datasets)

{'bucc': {'test': <_PrefetchDataset element_spec={'source_id': TensorSpec(shape=(), dtype=tf.string, name=None), 'source_sentence': TensorSpec(shape=(), dtype=tf.string, name=None), 'target_id': TensorSpec(shape=(), dtype=tf.string, name=None), 'target_sentence': TensorSpec(shape=(), dtype=tf.string, name=None)}>,
          'validation': <_PrefetchDataset element_spec={'source_id': TensorSpec(shape=(), dtype=tf.string, name=None), 'source_sentence': TensorSpec(shape=(), dtype=tf.string, name=None), 'target_id': TensorSpec(shape=(), dtype=tf.string, name=None), 'target_sentence': TensorSpec(shape=(), dtype=tf.string, name=None)}>},
 'xnli': {'train': <_PrefetchDataset element_spec={'hypothesis': {'language': TensorSpec(shape=(None,), dtype=tf.string, name=None), 'translation': TensorSpec(shape=(None,), dtype=tf.string, name=None)}, 'label': TensorSpec(shape=(), dtype=tf.int64, name=None), 'premise': {'ar': TensorSpec(shape=(), dtype=tf.string, name=None), 'bg': TensorSpec(shape=(), dtype=tf.string, name=None), 'de': TensorSpec(shape=(), dtype=tf.string, name=None), 'el': TensorSpec(shape=(), dtype=tf.string, name=None), 'en': TensorSpec(shape=(), dtype=tf.string, name=None), 'es': TensorSpec(shape=(), dtype=tf.string, name=None), 'fr': TensorSpec(shape=(), dtype=tf.string, name=None), 'hi': TensorSpec(shape=(), dtype=tf.string, name=None), 'ru': TensorSpec(shape=(), dtype=tf.string, name=None), 'sw': TensorSpec(shape=(), dtype=tf.string, name=None), 'th': TensorSpec(shape=(), dtype=tf.string, name=None), 'tr': TensorSpec(shape=(), dtype=tf.string, name=None), 'ur': TensorSpec(shape=(), dtype=tf.string, name=None), 'vi': TensorSpec(shape=(), dtype=tf.string, name=None), 'zh': TensorSpec(shape=(), dtype=tf.string, name=None)} }>} }

The load_all_datasets method loads all available datasets for a given collection:

all_datasets = collection_loader.load_all_datasets()

pprint.pprint(all_datasets)

{'bucc': {'test': <_PrefetchDataset element_spec={'source_id': TensorSpec(shape=(), dtype=tf.string, name=None), 'source_sentence': TensorSpec(shape=(), dtype=tf.string, name=None), 'target_id': TensorSpec(shape=(), dtype=tf.string, name=None), 'target_sentence': TensorSpec(shape=(), dtype=tf.string, name=None)}>,
          'validation': <_PrefetchDataset element_spec={'source_id': TensorSpec(shape=(), dtype=tf.string, name=None), 'source_sentence': TensorSpec(shape=(), dtype=tf.string, name=None), 'target_id': TensorSpec(shape=(), dtype=tf.string, name=None), 'target_sentence': TensorSpec(shape=(), dtype=tf.string, name=None)}>},
 'mlqa': {'test': <_PrefetchDataset element_spec={'answers': {'answer_start': TensorSpec(shape=(None,), dtype=tf.int32, name=None), 'text': TensorSpec(shape=(None,), dtype=tf.string, name=None)}, 'context': TensorSpec(shape=(), dtype=tf.string, name=None), 'id': TensorSpec(shape=(), dtype=tf.string, name=None), 'question': TensorSpec(shape=(), dtype=tf.string, name=None), 'title': TensorSpec(shape=(), dtype=tf.string, name=None)}>,
          'validation': <_PrefetchDataset element_spec={'answers': {'answer_start': TensorSpec(shape=(None,), dtype=tf.int32, name=None), 'text': TensorSpec(shape=(None,), dtype=tf.string, name=None)}, 'context': TensorSpec(shape=(), dtype=tf.string, name=None), 'id': TensorSpec(shape=(), dtype=tf.string, name=None), 'question': TensorSpec(shape=(), dtype=tf.string, name=None), 'title': TensorSpec(shape=(), dtype=tf.string, name=None)}>},
 'ner': {'test': <_PrefetchDataset element_spec={'langs': TensorSpec(shape=(None,), dtype=tf.string, name=None), 'spans': TensorSpec(shape=(None,), dtype=tf.string, name=None), 'tags': TensorSpec(shape=(None,), dtype=tf.int64, name=None), 'tokens': TensorSpec(shape=(None,), dtype=tf.string, name=None)}>,
         'train': <_PrefetchDataset element_spec={'langs': TensorSpec(shape=(None,), dtype=tf.string, name=None), 'spans': TensorSpec(shape=(None,), dtype=tf.string, name=None), 'tags': TensorSpec(shape=(None,), dtype=tf.int64, name=None), 'tokens': TensorSpec(shape=(None,), dtype=tf.string, name=None)}>,
         'validation': <_PrefetchDataset element_spec={'langs': TensorSpec(shape=(None,), dtype=tf.string, name=None), 'spans': TensorSpec(shape=(None,), dtype=tf.string, name=None), 'tags': TensorSpec(shape=(None,), dtype=tf.int64, name=None), 'tokens': TensorSpec(shape=(None,), dtype=tf.string, name=None)}>},
 'pawsx': {'train': <_PrefetchDataset element_spec={'label': TensorSpec(shape=(), dtype=tf.int64, name=None), 'sentence1': TensorSpec(shape=(), dtype=tf.string, name=None), 'sentence2': TensorSpec(shape=(), dtype=tf.string, name=None)}>},
 'pos': {'dev': <_PrefetchDataset element_spec={'tokens': TensorSpec(shape=(None,), dtype=tf.string, name=None), 'upos': TensorSpec(shape=(None,), dtype=tf.int64, name=None)}>,
         'test': <_PrefetchDataset element_spec={'tokens': TensorSpec(shape=(None,), dtype=tf.string, name=None), 'upos': TensorSpec(shape=(None,), dtype=tf.int64, name=None)}>,
         'train': <_PrefetchDataset element_spec={'tokens': TensorSpec(shape=(None,), dtype=tf.string, name=None), 'upos': TensorSpec(shape=(None,), dtype=tf.int64, name=None)}>},
 'tatoeba': {'train': <_PrefetchDataset element_spec={'source_language': TensorSpec(shape=(), dtype=tf.string, name=None), 'source_sentence': TensorSpec(shape=(), dtype=tf.string, name=None), 'target_language': TensorSpec(shape=(), dtype=tf.string, name=None), 'target_sentence': TensorSpec(shape=(), dtype=tf.string, name=None)}>},
 'tydiqa': {'train': <_PrefetchDataset element_spec={'answers': {'answer_start': TensorSpec(shape=(None,), dtype=tf.int32, name=None), 'text': TensorSpec(shape=(None,), dtype=tf.string, name=None)}, 'context': TensorSpec(shape=(), dtype=tf.string, name=None), 'id': TensorSpec(shape=(), dtype=tf.string, name=None), 'question': TensorSpec(shape=(), dtype=tf.string, name=None), 'title': TensorSpec(shape=(), dtype=tf.string, name=None)}>,
            'translate-train-ar': <_PrefetchDataset element_spec={'answers': {'answer_start': TensorSpec(shape=(None,), dtype=tf.int32, name=None), 'text': TensorSpec(shape=(None,), dtype=tf.string, name=None)}, 'context': TensorSpec(shape=(), dtype=tf.string, name=None), 'id': TensorSpec(shape=(), dtype=tf.string, name=None), 'question': TensorSpec(shape=(), dtype=tf.string, name=None), 'title': TensorSpec(shape=(), dtype=tf.string, name=None)}>,
            'translate-train-bn': <_PrefetchDataset element_spec={'answers': {'answer_start': TensorSpec(shape=(None,), dtype=tf.int32, name=None), 'text': TensorSpec(shape=(None,), dtype=tf.string, name=None)}, 'context': TensorSpec(shape=(), dtype=tf.string, name=None), 'id': TensorSpec(shape=(), dtype=tf.string, name=None), 'question': TensorSpec(shape=(), dtype=tf.string, name=None), 'title': TensorSpec(shape=(), dtype=tf.string, name=None)}>,
            'translate-train-fi': <_PrefetchDataset element_spec={'answers': {'answer_start': TensorSpec(shape=(None,), dtype=tf.int32, name=None), 'text': TensorSpec(shape=(None,), dtype=tf.string, name=None)}, 'context': TensorSpec(shape=(), dtype=tf.string, name=None), 'id': TensorSpec(shape=(), dtype=tf.string, name=None), 'question': TensorSpec(shape=(), dtype=tf.string, name=None), 'title': TensorSpec(shape=(), dtype=tf.string, name=None)}>,
            'translate-train-id': <_PrefetchDataset element_spec={'answers': {'answer_start': TensorSpec(shape=(None,), dtype=tf.int32, name=None), 'text': TensorSpec(shape=(None,), dtype=tf.string, name=None)}, 'context': TensorSpec(shape=(), dtype=tf.string, name=None), 'id': TensorSpec(shape=(), dtype=tf.string, name=None), 'question': TensorSpec(shape=(), dtype=tf.string, name=None), 'title': TensorSpec(shape=(), dtype=tf.string, name=None)}>,
            'translate-train-ko': <_PrefetchDataset element_spec={'answers': {'answer_start': TensorSpec(shape=(None,), dtype=tf.int32, name=None), 'text': TensorSpec(shape=(None,), dtype=tf.string, name=None)}, 'context': TensorSpec(shape=(), dtype=tf.string, name=None), 'id': TensorSpec(shape=(), dtype=tf.string, name=None), 'question': TensorSpec(shape=(), dtype=tf.string, name=None), 'title': TensorSpec(shape=(), dtype=tf.string, name=None)}>,
            'translate-train-ru': <_PrefetchDataset element_spec={'answers': {'answer_start': TensorSpec(shape=(None,), dtype=tf.int32, name=None), 'text': TensorSpec(shape=(None,), dtype=tf.string, name=None)}, 'context': TensorSpec(shape=(), dtype=tf.string, name=None), 'id': TensorSpec(shape=(), dtype=tf.string, name=None), 'question': TensorSpec(shape=(), dtype=tf.string, name=None), 'title': TensorSpec(shape=(), dtype=tf.string, name=None)}>,
            'translate-train-sw': <_PrefetchDataset element_spec={'answers': {'answer_start': TensorSpec(shape=(None,), dtype=tf.int32, name=None), 'text': TensorSpec(shape=(None,), dtype=tf.string, name=None)}, 'context': TensorSpec(shape=(), dtype=tf.string, name=None), 'id': TensorSpec(shape=(), dtype=tf.string, name=None), 'question': TensorSpec(shape=(), dtype=tf.string, name=None), 'title': TensorSpec(shape=(), dtype=tf.string, name=None)}>,
            'translate-train-te': <_PrefetchDataset element_spec={'answers': {'answer_start': TensorSpec(shape=(None,), dtype=tf.int32, name=None), 'text': TensorSpec(shape=(None,), dtype=tf.string, name=None)}, 'context': TensorSpec(shape=(), dtype=tf.string, name=None), 'id': TensorSpec(shape=(), dtype=tf.string, name=None), 'question': TensorSpec(shape=(), dtype=tf.string, name=None), 'title': TensorSpec(shape=(), dtype=tf.string, name=None)}>,
            'validation': <_PrefetchDataset element_spec={'answers': {'answer_start': TensorSpec(shape=(None,), dtype=tf.int32, name=None), 'text': TensorSpec(shape=(None,), dtype=tf.string, name=None)}, 'context': TensorSpec(shape=(), dtype=tf.string, name=None), 'id': TensorSpec(shape=(), dtype=tf.string, name=None), 'question': TensorSpec(shape=(), dtype=tf.string, name=None), 'title': TensorSpec(shape=(), dtype=tf.string, name=None)}>,
            'validation-ar': <_PrefetchDataset element_spec={'answers': {'answer_start': TensorSpec(shape=(None,), dtype=tf.int32, name=None), 'text': TensorSpec(shape=(None,), dtype=tf.string, name=None)}, 'context': TensorSpec(shape=(), dtype=tf.string, name=None), 'id': TensorSpec(shape=(), dtype=tf.string, name=None), 'question': TensorSpec(shape=(), dtype=tf.string, name=None), 'title': TensorSpec(shape=(), dtype=tf.string, name=None)}>,
            'validation-bn': <_PrefetchDataset element_spec={'answers': {'answer_start': TensorSpec(shape=(None,), dtype=tf.int32, name=None), 'text': TensorSpec(shape=(None,), dtype=tf.string, name=None)}, 'context': TensorSpec(shape=(), dtype=tf.string, name=None), 'id': TensorSpec(shape=(), dtype=tf.string, name=None), 'question': TensorSpec(shape=(), dtype=tf.string, name=None), 'title': TensorSpec(shape=(), dtype=tf.string, name=None)}>,
            'validation-en': <_PrefetchDataset element_spec={'answers': {'answer_start': TensorSpec(shape=(None,), dtype=tf.int32, name=None), 'text': TensorSpec(shape=(None,), dtype=tf.string, name=None)}, 'context': TensorSpec(shape=(), dtype=tf.string, name=None), 'id': TensorSpec(shape=(), dtype=tf.string, name=None), 'question': TensorSpec(shape=(), dtype=tf.string, name=None), 'title': TensorSpec(shape=(), dtype=tf.string, name=None)}>,
            'validation-fi': <_PrefetchDataset element_spec={'answers': {'answer_start': TensorSpec(shape=(None,), dtype=tf.int32, name=None), 'text': TensorSpec(shape=(None,), dtype=tf.string, name=None)}, 'context': TensorSpec(shape=(), dtype=tf.string, name=None), 'id': TensorSpec(shape=(), dtype=tf.string, name=None), 'question': TensorSpec(shape=(), dtype=tf.string, name=None), 'title': TensorSpec(shape=(), dtype=tf.string, name=None)}>,
            'validation-id': <_PrefetchDataset element_spec={'answers': {'answer_start': TensorSpec(shape=(None,), dtype=tf.int32, name=None), 'text': TensorSpec(shape=(None,), dtype=tf.string, name=None)}, 'context': TensorSpec(shape=(), dtype=tf.string, name=None), 'id': TensorSpec(shape=(), dtype=tf.string, name=None), 'question': TensorSpec(shape=(), dtype=tf.string, name=None), 'title': TensorSpec(shape=(), dtype=tf.string, name=None)}>,
            'validation-ko': <_PrefetchDataset element_spec={'answers': {'answer_start': TensorSpec(shape=(None,), dtype=tf.int32, name=None), 'text': TensorSpec(shape=(None,), dtype=tf.string, name=None)}, 'context': TensorSpec(shape=(), dtype=tf.string, name=None), 'id': TensorSpec(shape=(), dtype=tf.string, name=None), 'question': TensorSpec(shape=(), dtype=tf.string, name=None), 'title': TensorSpec(shape=(), dtype=tf.string, name=None)}>,
            'validation-ru': <_PrefetchDataset element_spec={'answers': {'answer_start': TensorSpec(shape=(None,), dtype=tf.int32, name=None), 'text': TensorSpec(shape=(None,), dtype=tf.string, name=None)}, 'context': TensorSpec(shape=(), dtype=tf.string, name=None), 'id': TensorSpec(shape=(), dtype=tf.string, name=None), 'question': TensorSpec(shape=(), dtype=tf.string, name=None), 'title': TensorSpec(shape=(), dtype=tf.string, name=None)}>,
            'validation-sw': <_PrefetchDataset element_spec={'answers': {'answer_start': TensorSpec(shape=(None,), dtype=tf.int32, name=None), 'text': TensorSpec(shape=(None,), dtype=tf.string, name=None)}, 'context': TensorSpec(shape=(), dtype=tf.string, name=None), 'id': TensorSpec(shape=(), dtype=tf.string, name=None), 'question': TensorSpec(shape=(), dtype=tf.string, name=None), 'title': TensorSpec(shape=(), dtype=tf.string, name=None)}>,
            'validation-te': <_PrefetchDataset element_spec={'answers': {'answer_start': TensorSpec(shape=(None,), dtype=tf.int32, name=None), 'text': TensorSpec(shape=(None,), dtype=tf.string, name=None)}, 'context': TensorSpec(shape=(), dtype=tf.string, name=None), 'id': TensorSpec(shape=(), dtype=tf.string, name=None), 'question': TensorSpec(shape=(), dtype=tf.string, name=None), 'title': TensorSpec(shape=(), dtype=tf.string, name=None)}>},
 'xnli': {'train': <_PrefetchDataset element_spec={'hypothesis': {'language': TensorSpec(shape=(None,), dtype=tf.string, name=None), 'translation': TensorSpec(shape=(None,), dtype=tf.string, name=None)}, 'label': TensorSpec(shape=(), dtype=tf.int64, name=None), 'premise': {'ar': TensorSpec(shape=(), dtype=tf.string, name=None), 'bg': TensorSpec(shape=(), dtype=tf.string, name=None), 'de': TensorSpec(shape=(), dtype=tf.string, name=None), 'el': TensorSpec(shape=(), dtype=tf.string, name=None), 'en': TensorSpec(shape=(), dtype=tf.string, name=None), 'es': TensorSpec(shape=(), dtype=tf.string, name=None), 'fr': TensorSpec(shape=(), dtype=tf.string, name=None), 'hi': TensorSpec(shape=(), dtype=tf.string, name=None), 'ru': TensorSpec(shape=(), dtype=tf.string, name=None), 'sw': TensorSpec(shape=(), dtype=tf.string, name=None), 'th': TensorSpec(shape=(), dtype=tf.string, name=None), 'tr': TensorSpec(shape=(), dtype=tf.string, name=None), 'ur': TensorSpec(shape=(), dtype=tf.string, name=None), 'vi': TensorSpec(shape=(), dtype=tf.string, name=None), 'zh': TensorSpec(shape=(), dtype=tf.string, name=None)} }>},
 'xquad': {'test': <_PrefetchDataset element_spec={'answers': {'answer_start': TensorSpec(shape=(None,), dtype=tf.int32, name=None), 'text': TensorSpec(shape=(None,), dtype=tf.string, name=None)}, 'context': TensorSpec(shape=(), dtype=tf.string, name=None), 'id': TensorSpec(shape=(), dtype=tf.string, name=None), 'question': TensorSpec(shape=(), dtype=tf.string, name=None), 'title': TensorSpec(shape=(), dtype=tf.string, name=None)}>,
           'translate-dev': <_PrefetchDataset element_spec={'answers': {'answer_start': TensorSpec(shape=(None,), dtype=tf.int32, name=None), 'text': TensorSpec(shape=(None,), dtype=tf.string, name=None)}, 'context': TensorSpec(shape=(), dtype=tf.string, name=None), 'id': TensorSpec(shape=(), dtype=tf.string, name=None), 'question': TensorSpec(shape=(), dtype=tf.string, name=None), 'title': TensorSpec(shape=(), dtype=tf.string, name=None)}>,
           'translate-test': <_PrefetchDataset element_spec={'answers': {'answer_start': TensorSpec(shape=(None,), dtype=tf.int32, name=None), 'text': TensorSpec(shape=(None,), dtype=tf.string, name=None)}, 'context': TensorSpec(shape=(), dtype=tf.string, name=None), 'id': TensorSpec(shape=(), dtype=tf.string, name=None), 'question': TensorSpec(shape=(), dtype=tf.string, name=None), 'title': TensorSpec(shape=(), dtype=tf.string, name=None)}>,
           'translate-train': <_PrefetchDataset element_spec={'answers': {'answer_start': TensorSpec(shape=(None,), dtype=tf.int32, name=None), 'text': TensorSpec(shape=(None,), dtype=tf.string, name=None)}, 'context': TensorSpec(shape=(), dtype=tf.string, name=None), 'id': TensorSpec(shape=(), dtype=tf.string, name=None), 'question': TensorSpec(shape=(), dtype=tf.string, name=None), 'title': TensorSpec(shape=(), dtype=tf.string, name=None)}>} }

The load_datasets method accepts the following optional parameters:

split: which split(s) to load. It accepts a single split (split="test") or a list of splits: (split=["train", "test"]). If not specified, it will load all splits for the given dataset.
loader_kwargs: keyword arguments to be passed to the tfds.load function. Refer to the tfds.load documentation for a comprehensive overview of the different loading options.

Specifying `loader_kwargs`

The loader_kwargs are optional keyword arguments to be passed to the tfds.load function. They can be specified in three ways:

When initializing the DatasetCollectionLoader class:

collection_loader = tfds.dataset_collection('xtreme', loader_kwargs=dict(split='train', batch_size=10, try_gcs=False))

Using DatasetCollectioLoader's set_loader_kwargs method:

collection_loader.set_loader_kwargs(dict(split='train', batch_size=10, try_gcs=False))

As optional parameters to the load_dataset, load_datasets and load_all_datasets methods.

dataset = collection_loader.load_dataset('ner', loader_kwargs=dict(split='train', batch_size=10, try_gcs=False))

Feedback

We are continuously trying to improve the dataset creation workflow, but can only do so if we are aware of the issues. Which issues, errors did you encountered while creating the dataset collection? Was there a part which was confusing, boilerplate or wasn't working the first time? Please share your feedback on GitHub.