Anserini: Regressions for MS MARCO Passage Retrieval
This page documents regression experiments for the MS MARCO Passage Retrieval Task with docTTTTTquery expansions. These experiments are integrated into Anserini's regression testing framework.
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.
Typical indexing command:
nohup sh target/appassembler/bin/IndexCollection -collection JsonCollection -input /path/to/msmarco-passage-docTTTTTquery \
-index lucene-index.msmarco-passage-docTTTTTquery.pos+docvectors+rawdocs -generator LuceneDocumentGenerator -threads 9 \
-storePositions -storeDocvectors -storeRawDocs >& log.msmarco-passage-docTTTTTquery.pos+docvectors+rawdocs &
The directory /path/to/msmarco-passage/
should be a directory containing jsonl
files converted from the official passage collection, appended with the docTTTTTquery expansions.
For additional details, see explanation of common indexing options.
Topics and qrels are stored in src/main/resources/topics-and-qrels/
.
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:
nohup target/appassembler/bin/SearchCollection -index lucene-index.msmarco-passage-docTTTTTquery.pos+docvectors+rawdocs \
-topicreader TsvInt -topics src/main/resources/topics-and-qrels/topics.msmarco-passage.dev-subset.txt \
-bm25 -output run.msmarco-passage-docTTTTTquery.bm25-default.topics.msmarco-passage.dev-subset.txt &
nohup target/appassembler/bin/SearchCollection -index lucene-index.msmarco-passage-docTTTTTquery.pos+docvectors+rawdocs \
-topicreader TsvInt -topics src/main/resources/topics-and-qrels/topics.msmarco-passage.dev-subset.txt \
-bm25 -rm3 -output run.msmarco-passage-docTTTTTquery.bm25-default+rm3.topics.msmarco-passage.dev-subset.txt &
nohup target/appassembler/bin/SearchCollection -index lucene-index.msmarco-passage-docTTTTTquery.pos+docvectors+rawdocs \
-topicreader TsvInt -topics src/main/resources/topics-and-qrels/topics.msmarco-passage.dev-subset.txt \
-bm25 -bm25.k1 0.82 -bm25.b 0.68 -output run.msmarco-passage-docTTTTTquery.bm25-tuned.topics.msmarco-passage.dev-subset.txt &
nohup target/appassembler/bin/SearchCollection -index lucene-index.msmarco-passage-docTTTTTquery.pos+docvectors+rawdocs \
-topicreader TsvInt -topics src/main/resources/topics-and-qrels/topics.msmarco-passage.dev-subset.txt \
-bm25 -bm25.k1 0.82 -bm25.b 0.68 -rm3 -output run.msmarco-passage-docTTTTTquery.bm25-tuned+rm3.topics.msmarco-passage.dev-subset.txt &
Evaluation can be performed using trec_eval
:
eval/trec_eval.9.0.4/trec_eval -m map -c -m recall.1000 -c src/main/resources/topics-and-qrels/qrels.msmarco-passage.dev-subset.txt run.msmarco-passage-docTTTTTquery.bm25-default.topics.msmarco-passage.dev-subset.txt
eval/trec_eval.9.0.4/trec_eval -m map -c -m recall.1000 -c src/main/resources/topics-and-qrels/qrels.msmarco-passage.dev-subset.txt run.msmarco-passage-docTTTTTquery.bm25-default+rm3.topics.msmarco-passage.dev-subset.txt
eval/trec_eval.9.0.4/trec_eval -m map -c -m recall.1000 -c src/main/resources/topics-and-qrels/qrels.msmarco-passage.dev-subset.txt run.msmarco-passage-docTTTTTquery.bm25-tuned.topics.msmarco-passage.dev-subset.txt
eval/trec_eval.9.0.4/trec_eval -m map -c -m recall.1000 -c src/main/resources/topics-and-qrels/qrels.msmarco-passage.dev-subset.txt run.msmarco-passage-docTTTTTquery.bm25-tuned+rm3.topics.msmarco-passage.dev-subset.txt
With the above commands, you should be able to replicate the following results:
MAP | BM25 (Default) | +RM3 | BM25 (Tuned) | +RM3 |
---|---|---|---|---|
MS MARCO Passage Ranking: Dev Queries | 0.2805 | 0.2243 | 0.2850 | 0.2266 |
R@1000 | BM25 (Default) | +RM3 | BM25 (Tuned) | +RM3 |
---|---|---|---|---|
MS MARCO Passage Ranking: Dev Queries | 0.9470 | 0.9463 | 0.9471 | 0.9479 |
The setting "default" refers the default BM25 settings of k1=0.9
, b=0.4
, while "tuned" refers to the tuned setting of k1=0.82
, b=0.72
on the original passages.
See this page for more details.
Note that these results are slightly different from the above referenced page because those experiments make up "fake" scores when converting runs from MS MARCO format into TREC format for evaluation by trec_eval
.
To replicate the exact conditions for a leaderboard submission, retrieve using the following command:
wget https://storage.googleapis.com/doctttttquery_git/queries.dev.small.tsv
sh target/appassembler/bin/SearchMsmarco \
-index lucene-index.msmarco-passage-docTTTTTquery.pos+docvectors+rawdocs \
-qid_queries queries.dev.small.tsv \
-output run.msmarco-passage-docTTTTTquery -hits 1000
Evaluate using the MS MARCO eval script:
wget https://storage.googleapis.com/doctttttquery_git/qrels.dev.small.tsv
python src/main/python/msmarco/msmarco_eval.py qrels.dev.small.tsv run.msmarco-passage-docTTTTTquery
The results should be:
#####################
MRR @10: 0.2767497271114737
QueriesRanked: 6980
#####################
Which matches the score described in the docTTTTTquery repo and also on the official MS MARCO leaderboard.