Anserini: Regressions for MS MARCO Passage Retrieval
This page documents regression experiments for the MS MARCO Passage Retrieval Task with docTTTTTquery (also called doc2query-T5) expansions, as proposed in the following paper:
Rodrigo Nogueira and Jimmy Lin. From doc2query to docTTTTTquery. December 2019.
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 indexes/lucene-index.msmarco-passage-docTTTTTquery.pos+docvectors+raw \
-generator DefaultLuceneDocumentGenerator \
-threads 9 -storePositions -storeDocvectors -storeRaw \
>& logs/log.msmarco-passage-docTTTTTquery &
The directory /path/to/msmarco-passage-docTTTTTquery
should be a directory containing jsonl
files containing the expanded passage collection.
Instructions in the docTTTTTquery repo explain how to perform this data preparation.
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 indexes/lucene-index.msmarco-passage-docTTTTTquery.pos+docvectors+raw \
-topicreader TsvInt -topics src/main/resources/topics-and-qrels/topics.msmarco-passage.dev-subset.txt \
-output runs/run.msmarco-passage-docTTTTTquery.bm25-default.topics.msmarco-passage.dev-subset.txt \
-bm25 &
nohup target/appassembler/bin/SearchCollection -index indexes/lucene-index.msmarco-passage-docTTTTTquery.pos+docvectors+raw \
-topicreader TsvInt -topics src/main/resources/topics-and-qrels/topics.msmarco-passage.dev-subset.txt \
-output runs/run.msmarco-passage-docTTTTTquery.bm25-default+rm3.topics.msmarco-passage.dev-subset.txt \
-bm25 -rm3 &
nohup target/appassembler/bin/SearchCollection -index indexes/lucene-index.msmarco-passage-docTTTTTquery.pos+docvectors+raw \
-topicreader TsvInt -topics src/main/resources/topics-and-qrels/topics.msmarco-passage.dev-subset.txt \
-output runs/run.msmarco-passage-docTTTTTquery.bm25-tuned.topics.msmarco-passage.dev-subset.txt \
-bm25 -bm25.k1 0.82 -bm25.b 0.68 &
nohup target/appassembler/bin/SearchCollection -index indexes/lucene-index.msmarco-passage-docTTTTTquery.pos+docvectors+raw \
-topicreader TsvInt -topics src/main/resources/topics-and-qrels/topics.msmarco-passage.dev-subset.txt \
-output runs/run.msmarco-passage-docTTTTTquery.bm25-tuned+rm3.topics.msmarco-passage.dev-subset.txt \
-bm25 -bm25.k1 0.82 -bm25.b 0.68 -rm3 &
nohup target/appassembler/bin/SearchCollection -index indexes/lucene-index.msmarco-passage-docTTTTTquery.pos+docvectors+raw \
-topicreader TsvInt -topics src/main/resources/topics-and-qrels/topics.msmarco-passage.dev-subset.txt \
-output runs/run.msmarco-passage-docTTTTTquery.bm25-tuned2.topics.msmarco-passage.dev-subset.txt \
-bm25 -bm25.k1 2.18 -bm25.b 0.86 &
nohup target/appassembler/bin/SearchCollection -index indexes/lucene-index.msmarco-passage-docTTTTTquery.pos+docvectors+raw \
-topicreader TsvInt -topics src/main/resources/topics-and-qrels/topics.msmarco-passage.dev-subset.txt \
-output runs/run.msmarco-passage-docTTTTTquery.bm25-tuned2+rm3.topics.msmarco-passage.dev-subset.txt \
-bm25 -bm25.k1 2.18 -bm25.b 0.86 -rm3 &
Evaluation can be performed using trec_eval
:
tools/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 runs/run.msmarco-passage-docTTTTTquery.bm25-default.topics.msmarco-passage.dev-subset.txt
tools/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 runs/run.msmarco-passage-docTTTTTquery.bm25-default+rm3.topics.msmarco-passage.dev-subset.txt
tools/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 runs/run.msmarco-passage-docTTTTTquery.bm25-tuned.topics.msmarco-passage.dev-subset.txt
tools/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 runs/run.msmarco-passage-docTTTTTquery.bm25-tuned+rm3.topics.msmarco-passage.dev-subset.txt
tools/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 runs/run.msmarco-passage-docTTTTTquery.bm25-tuned2.topics.msmarco-passage.dev-subset.txt
tools/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 runs/run.msmarco-passage-docTTTTTquery.bm25-tuned2+rm3.topics.msmarco-passage.dev-subset.txt
With the above commands, you should be able to reproduce the following results:
MAP | BM25 (default) | +RM3 | BM25 (tuned) | +RM3 | BM25 (tuned2) | +RM3 |
---|---|---|---|---|---|---|
MS MARCO Passage: Dev | 0.2805 | 0.2243 | 0.2850 | 0.2266 | 0.2893 | 0.2464 |
R@1000 | BM25 (default) | +RM3 | BM25 (tuned) | +RM3 | BM25 (tuned2) | +RM3 |
---|---|---|---|---|---|---|
MS MARCO Passage: Dev | 0.9470 | 0.9463 | 0.9471 | 0.9479 | 0.9506 | 0.9528 |
Explanation of settings:
- The setting "default" refers the default BM25 settings of
k1=0.9
,b=0.4
. - The setting "tuned" refers to
k1=0.82
,b=0.68
, tuned on on the original passages, as described in this page. - The setting "tuned2" refers to
k1=2.18
,b=0.86
, tuned to optimize for recall@1000 directly on the expanded passages (in 2020/12); this is the configuration reported in the Lin et al. (SIGIR 2021) Pyserini paper.
Note that the above runs are generated with SearchCollection
in the TREC format, which due to tie-breaking effects gives slightly different results from SearchMsmarco
in the MS MARCO format.
The following command generates with SearchMsmarco
the run denoted "BM25 (tuned)" above (k1=0.82
, b=0.68
), which corresponds to the entry "docTTTTTquery" dated 2019/11/27 on the MS MARCO Passage Ranking Leaderboard:
$ sh target/appassembler/bin/SearchMsmarco -hits 1000 -threads 8 \
-index indexes/lucene-index.msmarco-passage-docTTTTTquery.pos+docvectors+raw \
-queries collections/msmarco-passage/queries.dev.small.tsv \
-k1 0.82 -b 0.68 \
-output runs/run.msmarco-passage-docTTTTTquery.1
$ python tools/scripts/msmarco/msmarco_passage_eval.py \
collections/msmarco-passage/qrels.dev.small.tsv runs/run.msmarco-passage-docTTTTTquery.1
#####################
MRR @10: 0.27680089370991834
QueriesRanked: 6980
#####################
This corresponds to the scores reported in the following paper:
Rodrigo Nogueira and Jimmy Lin. From doc2query to docTTTTTquery. December 2019.
And are identical to the scores reported in the docTTTTTquery repo.
The following command generates with SearchMsmarco
the run denoted "BM25 (tuned2)" above (k1=2.18
, b=0.86
).
This does not correspond to an official leaderboard submission.
$ sh target/appassembler/bin/SearchMsmarco -hits 1000 -threads 8 \
-index indexes/lucene-index.msmarco-passage-docTTTTTquery.pos+docvectors+raw \
-queries collections/msmarco-passage/queries.dev.small.tsv \
-k1 2.18 -b 0.86 \
-output runs/run.msmarco-passage-docTTTTTquery.2
$ python tools/scripts/msmarco/msmarco_passage_eval.py \
collections/msmarco-passage/qrels.dev.small.tsv runs/run.msmarco-passage-docTTTTTquery.2
#####################
MRR @10: 0.281560751807885
QueriesRanked: 6980
#####################
This corresponds to the scores reported in the Lin et al. (SIGIR 2021) Pyserini paper.