Skip to content

Latest commit

 

History

History
103 lines (73 loc) · 6.42 KB

regressions-msmarco-v1-passage.cos-dpr-distil.flat.cached.md

File metadata and controls

103 lines (73 loc) · 6.42 KB
Raw

Anserini Regressions: MS MARCO Passage Ranking

Model: cosDPR-distil with flat indexes (using cached queries)

This page describes regression experiments, integrated into Anserini's regression testing framework, using the cosDPR-distil model on the MS MARCO passage ranking task, as described in the following paper:

Xueguang Ma, Tommaso Teofili, and Jimmy Lin. Anserini Gets Dense Retrieval: Integration of Lucene's HNSW Indexes. Proceedings of the 32nd International Conference on Information and Knowledge Management (CIKM 2023), October 2023, pages 5366–5370, Birmingham, the United Kingdom.

In these experiments, we are using cached queries (i.e., cached results of query encoding).

The exact configurations for these regressions are stored in this YAML file. Note that this page is automatically generated from this template as part of Anserini's regression pipeline, so do not modify this page directly; modify the template instead and then run bin/build.sh to rebuild the documentation.

From one of our Waterloo servers (e.g., orca), the following command will perform the complete regression, end to end:

python src/main/python/run_regression.py --index --verify --search --regression msmarco-v1-passage.cos-dpr-distil.flat.cached

We make available a version of the MS MARCO Passage Corpus that has already been encoded with cosDPR-distil.

From any machine, the following command will download the corpus and perform the complete regression, end to end:

python src/main/python/run_regression.py --download --index --verify --search --regression msmarco-v1-passage.cos-dpr-distil.flat.cached

The run_regression.py script automates the following steps, but if you want to perform each step manually, simply copy/paste from the commands below and you'll obtain the same regression results.

Corpus Download

Download the corpus and unpack into collections/:

wget https://rgw.cs.uwaterloo.ca/pyserini/data/msmarco-passage-cos-dpr-distil.tar -P collections/
tar xvf collections/msmarco-passage-cos-dpr-distil.tar -C collections/

To confirm, msmarco-passage-cos-dpr-distil.tar is 57 GB and has MD5 checksum e20ffbc8b5e7f760af31298aefeaebbd. With the corpus downloaded, the following command will perform the remaining steps below:

python src/main/python/run_regression.py --index --verify --search --regression msmarco-v1-passage.cos-dpr-distil.flat.cached \
  --corpus-path collections/msmarco-passage-cos-dpr-distil

Indexing

Sample indexing command, building flat indexes:

bin/run.sh io.anserini.index.IndexCollection \
  -collection JsonDenseVectorCollection \
  -input /path/to/msmarco-passage-cos-dpr-distil \
  -generator DenseVectorDocumentGenerator \
  -index indexes/lucene-flat.msmarco-v1-passage.cos-dpr-distil/ \
  -threads 16  \
  >& logs/log.msmarco-passage-cos-dpr-distil &

The path /path/to/msmarco-passage-cos-dpr-distil/ should point to the corpus downloaded above. Upon completion, we should have an index with 8,841,823 documents.

Retrieval

Topics and qrels are stored here, which is linked to the Anserini repo as a submodule. The regression experiments here evaluate on the 6980 dev set questions; see this page for more details.

After indexing has completed, you should be able to perform retrieval as follows using HNSW indexes:

bin/run.sh io.anserini.search.SearchCollection \
  -index indexes/lucene-flat.msmarco-v1-passage.cos-dpr-distil/ \
  -topics tools/topics-and-qrels/topics.msmarco-passage.dev-subset.cos-dpr-distil.jsonl.gz \
  -topicReader JsonIntVector \
  -output runs/run.msmarco-passage-cos-dpr-distil.cos-dpr-distil-flat-cached.topics.msmarco-passage.dev-subset.cos-dpr-distil.jsonl.txt \
  -generator VectorQueryGenerator -topicField vector -threads 16 -hits 1000 &

Evaluation can be performed using trec_eval:

bin/trec_eval -c -m map tools/topics-and-qrels/qrels.msmarco-passage.dev-subset.txt runs/run.msmarco-passage-cos-dpr-distil.cos-dpr-distil-flat-cached.topics.msmarco-passage.dev-subset.cos-dpr-distil.jsonl.txt
bin/trec_eval -c -M 10 -m recip_rank tools/topics-and-qrels/qrels.msmarco-passage.dev-subset.txt runs/run.msmarco-passage-cos-dpr-distil.cos-dpr-distil-flat-cached.topics.msmarco-passage.dev-subset.cos-dpr-distil.jsonl.txt
bin/trec_eval -c -m recall.100 tools/topics-and-qrels/qrels.msmarco-passage.dev-subset.txt runs/run.msmarco-passage-cos-dpr-distil.cos-dpr-distil-flat-cached.topics.msmarco-passage.dev-subset.cos-dpr-distil.jsonl.txt
bin/trec_eval -c -m recall.1000 tools/topics-and-qrels/qrels.msmarco-passage.dev-subset.txt runs/run.msmarco-passage-cos-dpr-distil.cos-dpr-distil-flat-cached.topics.msmarco-passage.dev-subset.cos-dpr-distil.jsonl.txt

Effectiveness

With the above commands, you should be able to reproduce the following results:

AP@1000 cosDPR-distil
MS MARCO Passage: Dev 0.3942
RR@10 cosDPR-distil
MS MARCO Passage: Dev 0.3896
R@100 cosDPR-distil
MS MARCO Passage: Dev 0.9075
R@1000 cosDPR-distil
MS MARCO Passage: Dev 0.9796

Note that since we're running brute-force search with cached queries on non-quantized indexes, the results should be reproducible exactly.