Skip to content

Add MS MARCO reproduction results - #3430

Merged
lintool merged 4 commits into
castorini:masterfrom
Ajay9450:ajay-msmarco-reproduction
Sep 15, 2026
Merged

lintool merged 4 commits into
castorini:masterfrom
Ajay9450:ajay-msmarco-reproduction

Conversation

@Ajay9450

@Ajay9450 Ajay9450 commented Sep 12, 2026

Copy link
Copy Markdown
Contributor

Foundations of Retrieval — Anserini Reproduction Log (Lessons 1–3)

Reproduction record for Lessons 1–3 of the Foundations of Retrieval onboarding path, submitted as castorini/anserini PR #3430.

Environment: Windows 11; Python 3.10.11; Java JDK 21.0.12.1; Maven 3.9.16; Lucene 10.5.0.


Lesson 1 — Anserini: Start Here

Results reproduced by @Ajay9450 on 2026-09-12.

  • Collection passages: 8,841,823
  • Development queries: 6,980
  • Development relevance judgments: 7,437
  • Training relevance judgments: 532,761
  • Converted collection: 9 JSONL files (first eight with 1,000,000 passages each, final file with 841,823)

Lesson 2 — Anserini: BM25 Baselines for MS MARCO Passage Ranking

Results reproduced by @Ajay9450 on 2026-09-12.

  • Collection: 8,841,823 documents indexed, 0 errors/empty/skipped
  • Archive MD5: 31644046b18952c1386cd4564ba2ae69
  • BM25 parameters: k1 = 0.82, b = 0.68, hits = 1000, queries = 6,980

Evaluation: MRR@10 = 0.18741227770955546 · MAP = 0.1957 · Recall@1000 = 0.8573

For query 1048585, document 7187158 was relevant at rank 1 (reciprocal rank 1.0).


Lesson 3 — Anserini: Dense Retrieval for MS MARCO Passage Ranking

Results reproduced by @Ajay9450 on 2026-09-12.

  • Prebuilt BM25 archive MD5: 678876e8c99a89933d553609a0fd8793 (2,170,758,745 bytes); MRR@10 = 0.18741227770955546
  • BGE-base HNSW: encoder = BgeBaseEn15, efSearch = 1000, hits = 1000, threads = 4, queries = 6,980
    MRR@10 = 0.3574197252922182 · Recall@1000 = 0.9787726839

Windows trec_eval reported unsupported win-amd64; affected metrics were independently computed from official qrels and run files.


Summary of Reproduced Retrieval Results

Lesson Dataset Method Metric Result
2 MS MARCO BM25 MRR@10 0.18741227770955546
2 MS MARCO BM25 MAP 0.1957
2 MS MARCO BM25 Recall@1000 0.8573
3 MS MARCO BGE-base HNSW MRR@10 0.3574197252922182
3 MS MARCO BGE-base HNSW Recall@1000 0.9787726839

Status

  • ✓ Lesson 1 — Anserini: Start Here
  • ✓ Lesson 2 — Anserini: BM25 Baselines for MS MARCO Passage Ranking
  • ✓ Lesson 3 — Anserini: Dense Retrieval for MS MARCO Passage Ranking

Comment thread docs/experiments-msmarco-passage.md
Comment thread docs/experiments-msmarco-passage2.md
Comment thread docs/start-here.md
@Ajay9450
Ajay9450 force-pushed the ajay-msmarco-reproduction branch from 918f2e7 to 7533048 Compare September 15, 2026 05:14
@Ajay9450
Ajay9450 requested a review from lintool September 15, 2026 05:39

@Ajay9450 Ajay9450 left a comment

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

All changes are made as instructed.

Comment thread docs/experiments-msmarco-passage.md
@Ajay9450
Ajay9450 force-pushed the ajay-msmarco-reproduction branch from 7533048 to 3777f80 Compare September 15, 2026 11:57
@Ajay9450
Ajay9450 requested a review from lintool September 15, 2026 17:04
@codecov

codecov Bot commented Sep 15, 2026

Copy link
Copy Markdown

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 72.65%. Comparing base (ace2729) to head (76d782b).

Additional details and impacted files
@@            Coverage Diff            @@
##             master    #3430   +/-   ##
=========================================
  Coverage     72.65%   72.65%           
  Complexity     1985     1985           
=========================================
  Files           207      207           
  Lines         13098    13098           
  Branches       1866     1866           
=========================================
  Hits           9516     9516           
  Misses         2846     2846           
  Partials        736      736           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:
  • ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

@lintool
lintool merged commit ad5a11c into castorini:master Sep 15, 2026
3 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants