README.md

metadata

tags:
  - pyterrier
  - pyterrier-artifact
  - pyterrier-artifact.sparse_index
  - pyterrier-artifact.sparse_index.terrier
task_categories:
  - text-retrieval
viewer: false

scidocs.terrier

Description

Terrier index for SciDocs

Usage

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

Benchmarks

name	nDCG@10	R@1000
bm25	0.1582	0.5713
dph	0.1499	0.5649

Reproduction

import pyterrier as pt
from tqdm import tqdm
import ir_datasets
dataset = ir_datasets.load('beir/scidocs')
meta_docno_len = dataset.metadata()['docs']['fields']['doc_id']['max_len']
indexer = pt.IterDictIndexer("./scidocs.terrier", meta={'docno': meta_docno_len, 'text': 4096})
docs = ({'docno': d.doc_id, 'text': '{title}\n{text}'.format(**d._asdict())} for d in tqdm(dataset.docs))
indexer.index(docs)

Metadata

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