textbook 0.2.1

Text classification datasets

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

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)