nq.terrier

Description

Terrier index for NQ (Natural Questions)

Usage

# Load the artifact
import pyterrier as pt
index = pt.Artifact.from_hf('pyterrier/nq.terrier')
index.bm25()

Benchmarks

name	nDCG@10	R@1000
bm25	0.2814	0.8906
dph	0.2846	0.8926

Reproduction

import pyterrier as pt
from tqdm import tqdm
import pandas as pd
import ir_datasets
from pyterrier_pisa import PisaIndex
dataset = ir_datasets.load('beir/nq')
meta_docno_len = dataset.metadata()['docs']['fields']['doc_id']['max_len']
indexer = pt.IterDictIndexer("./nq.terrier", meta={'docno': meta_docno_len, 'text': 4096})
docs = ({'docno': d.doc_id, 'text': d.default_text()} for d in tqdm(dataset.docs))
indexer.index(docs)

Metadata

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