| # 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 |