Skip to main content

Text classification datasets

Project description

Textbook: Universal NLP Datasets

Current support few commonsense reasoning datsets(alphanli, hellaswag, physicaliqa, socialiqa, codah, and commonsenseqa). It adopts ray's multiprocessing in loading/processing the datasets.

Architecture

Architecture Image

Dependency

pip install -r requirements.txt

Download raw datasets

    bash fetch.sh

It downloads alphanli, hellaswag, physicaliqa, socialiqa, codah, and commonsenseqa from AWS. In case you want to use something-something, pelase download the dataset from 20bn's website.

Usage

Initialize ray

    import ray
    ray.init(memory=1024 * 1024 * 1024, num_cpus=2)

Load a dataset

    from transformers import BertTokenizer
    from textbook import *

    tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
    text_renderer = TextRenderer.remote(tokenizer)

    anli_tool = BatchTool(tokenizer, max_seq_len=128, source="anli")
    anli_dataset = TextDataset(path='data_cache/alphanli/eval.jsonl',
                                config=ANLIConfiguration.remote(), renderers=[text_renderer])
    # Batch by number of examples
    anli_iter = DataLoader(anli_dataset, batch_size=2, collate_fn=anli_tool.collate_fn)

    # Batch by number of tokens
    anli_iter = DataLoader(anli_dataset, batch_sampler=TokenBasedSampler(anli_dataset, batch_size=128), collate_fn=anli_tool.collate_fn)

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

textbook-0.2.2.tar.gz (8.7 kB view hashes)

Uploaded Source

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page