ragwiki-terrier

Description

This is the sparse (Terrier) index for the Wikipedia 2018 corpus used by the Natural Questions and Textbook Question Answering datasets. By making this available as a Huggingface dataset, experiments can be performed using the existing index.

Usage

# Load the artifact
import pyterrier as pt
artifact = pt.Artifact.from_hf('pyterrier/ragwiki-terrier')
bm25 = artifact.bm25()
bm25.search("what are chemical reactions")

Benchmarks

TODO: Provide benchmarks for the artifact.

Reproduction

This index was created uing PyTerrier's sparse (Terrier) indexer.

import pyterrier as pt, pyterrier_rag
ref = pt.IterDictIndexer(
        index_dir, 
        text_attrs=['title', 'text'], 
        meta={'docno' : 20, 'text' : 1750, 'title' : 125}
    ).index(pt.get_dataset('rag:nq_wiki').get_corpus_iter())
index = pt.terrier.TerrierIndex(ref)
index.to_hf('pyterrier/ragwiki-terrier')

Full code to reproduce this index is included in ragwiki_indexing-terrier.ipynb

Metadata

{
  "type": "sparse_index",
  "format": "terrier",
  "package_hint": "python-terrier"
}