Skip to content

Latest commit

 

History

History
165 lines (122 loc) · 10.3 KB

regressions-msmarco-v1-doc-segmented.unicoil.cached.md

File metadata and controls

165 lines (122 loc) · 10.3 KB

Anserini Regressions: MS MARCO Document Ranking

Model: uniCOIL (with doc2query-T5 expansions) on segmented documents (title/segment encoding)

This page describes regression experiments, integrated into Anserini's regression testing framework, using uniCOIL (with doc2query-T5 expansions) on the MS MARCO document ranking task. The uniCOIL model is described in the following paper:

Jimmy Lin and Xueguang Ma. A Few Brief Notes on DeepImpact, COIL, and a Conceptual Framework for Information Retrieval Techniques. arXiv:2106.14807.

The experiments on this page are not actually reported in the paper. However, the model is the same, applied to the MS MARCO segmented document corpus (with doc2query-T5 expansions). Retrieval uses MaxP technique, where we select the score of the highest-scoring passage from a document as the score for that document to produce a document ranking.

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-doc-segmented.unicoil.cached

We make available a version of the MS MARCO document corpus that has already been processed with uniCOIL, i.e., we have applied doc2query-T5 expansions, performed model inference on every document, and stored the output sparse vectors. Thus, no neural inference is involved.

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-doc-segmented.unicoil.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/JIMMYLIN-bucket0/data/msmarco-doc-segmented-unicoil.tar -P collections/
tar xvf collections/msmarco-doc-segmented-unicoil.tar -C collections/

To confirm, msmarco-doc-segmented-unicoil.tar is 19 GB and has MD5 checksum 6a00e2c0c375cb1e52c83ae5ac377ebb. 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-doc-segmented.unicoil.cached \
  --corpus-path collections/msmarco-doc-segmented-unicoil

Indexing

Sample indexing command:

bin/run.sh io.anserini.index.IndexCollection \
  -threads 16 \
  -collection JsonVectorCollection \
  -input /path/to/msmarco-doc-segmented-unicoil \
  -generator DefaultLuceneDocumentGenerator \
  -index indexes/lucene-inverted.msmarco-v1-doc-segmented.unicoil/ \
  -impact -pretokenized -storeDocvectors \
  >& logs/log.msmarco-doc-segmented-unicoil &

The directory /path/to/msmarco-doc-segmented-unicoil/ should point to the corpus downloaded above.

The important indexing options to note here are -impact -pretokenized: the first tells Anserini not to encode BM25 doclengths into Lucene's norms (which is the default) and the second option says not to apply any additional tokenization on the uniCOIL tokens. Upon completion, we should have an index with 20,545,677 documents.

For additional details, see explanation of common indexing options.

Retrieval

Topics and qrels are stored here, which is linked to the Anserini repo as a submodule. The regression experiments here evaluate on the 5193 dev set questions.

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

bin/run.sh io.anserini.search.SearchCollection \
  -index indexes/lucene-inverted.msmarco-v1-doc-segmented.unicoil/ \
  -topics tools/topics-and-qrels/topics.msmarco-doc.dev.unicoil.tsv.gz \
  -topicReader TsvInt \
  -output runs/run.msmarco-doc-segmented-unicoil.unicoil-cached_q.topics.msmarco-doc.dev.unicoil.txt \
  -impact -pretokenized -hits 10000 -selectMaxPassage -selectMaxPassage.delimiter "#" -selectMaxPassage.hits 1000 &

Evaluation can be performed using trec_eval:

bin/trec_eval -c -m map tools/topics-and-qrels/qrels.msmarco-doc.dev.txt runs/run.msmarco-doc-segmented-unicoil.unicoil-cached_q.topics.msmarco-doc.dev.unicoil.txt
bin/trec_eval -c -M 100 -m recip_rank tools/topics-and-qrels/qrels.msmarco-doc.dev.txt runs/run.msmarco-doc-segmented-unicoil.unicoil-cached_q.topics.msmarco-doc.dev.unicoil.txt
bin/trec_eval -c -m recall.100 tools/topics-and-qrels/qrels.msmarco-doc.dev.txt runs/run.msmarco-doc-segmented-unicoil.unicoil-cached_q.topics.msmarco-doc.dev.unicoil.txt
bin/trec_eval -c -m recall.1000 tools/topics-and-qrels/qrels.msmarco-doc.dev.txt runs/run.msmarco-doc-segmented-unicoil.unicoil-cached_q.topics.msmarco-doc.dev.unicoil.txt

Effectiveness

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

AP@1000 uniCOIL (with doc2query-T5 expansions)
MS MARCO Doc: Dev 0.3535
RR@100 uniCOIL (with doc2query-T5 expansions)
MS MARCO Doc: Dev 0.3531
R@100 uniCOIL (with doc2query-T5 expansions)
MS MARCO Doc: Dev 0.8858
R@1000 uniCOIL (with doc2query-T5 expansions)
MS MARCO Doc: Dev 0.9546

Additional Notes

This model corresponds to the run named "uniCOIL-d2q" on the official MS MARCO Document Ranking Leaderboard, submitted 2021/09/16. The following command generates a comparable run:

target/appassembler/bin/SearchCollection \
  -index indexes/lucene-index.msmarco-doc-segmented-unicoil/ \
  -topics tools/topics-and-qrels/topics.msmarco-doc.dev.unicoil.tsv.gz \
  -topicreader TsvInt \
  -output runs/run.msmarco-doc-segmented-unicoil.msmarco-doc.dev.txt \
  -format msmarco \
  -impact -pretokenized -hits 10000 -selectMaxPassage -selectMaxPassage.delimiter "#" -selectMaxPassage.hits 100

Note that the above command uses -format msmarco to directly generate a run in the MS MARCO output format. And to evaluate:

python tools/scripts/msmarco/msmarco_doc_eval.py \
  --judgments tools/topics-and-qrels/qrels.msmarco-doc.dev.txt \
  --run runs/run.msmarco-doc-segmented-unicoil.msmarco-doc.dev.txt

The results should be as follows:

#####################
MRR @100: 0.352997702662614
QueriesRanked: 5193
#####################

Note that due to MaxP and the need to generate runs to different depths, we can set -hits and -selectMaxPassage.hits differently. Because of tie-breaking effects, we get slightly different results:

Condition AP@1000 RR@100 R@100 R@1000 MS MARCO MRR @100
-hits 10000 -selectMaxPassage.hits 1000 (as above) 0.3535 0.3531 0.8858 0.9546 0.3533301973179882
-hits 10000 -selectMaxPassage.hits 100 0.3531 0.3531 0.8860 - 0.352997702662614
-hits 1000 -selectMaxPassage.hits 100 0.3531 0.3531 0.8860 - 0.352997702662614

Reproduction Log*

To add to this reproduction log, modify this template and run bin/build.sh to rebuild the documentation.