# HotpotQA Subset

## Purpose

This subset provides a standardized benchmark for evaluating sparse model performance on HotpotQA data, with negative examples pooled using BM25 retrieval.

## Creation

This subset was created using the `make_beir_subset.py` script with the following command:

```bash
python ./make_beir_subset.py --dataset hotpotqa --es-host http://localhost:9200 --force-reindex --split test
```

## Parameters Used

- **Dataset**: hotpotqa
- **Split**: test (test set)
- **ES Host**: http://localhost:9200
- **Force Reindex**: Enabled (overwrote existing Elasticsearch index)
- **Default settings**:
  - Number of queries: 1000 (or available queries if fewer)
  - Pool size: 100 documents per query
  - Pooler: elastic (BM25 retrieval)
  - Seed: 42