Skip to content

Latest commit

 

History

History
137 lines (98 loc) · 7.98 KB

regressions-msmarco-passage-deepimpact.md

File metadata and controls

137 lines (98 loc) · 7.98 KB

Anserini Regressions: MS MARCO Passage Ranking

Model: DeepImpact

This page describes regression experiments, integrated into Anserini's regression testing framework, using DeepImpact on the MS MARCO passage ranking task. The DeepImpact model is described in the following paper:

Antonio Mallia, Omar Khattab, Nicola Tonellotto, and Torsten Suel. Learning Passage Impacts for Inverted Indexes. SIGIR 2021.

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.

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-passage-deepimpact

We make available a version of the MS MARCO Passage Corpus that has already been processed with DeepImpact, i.e., we have applied neural inference and stored the output sparse vectors.

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-passage-deepimpact

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-passage-deepimpact.tar -P collections/
tar xvf collections/msmarco-passage-deepimpact.tar -C collections/

To confirm, msmarco-passage-deepimpact.tar is 3.6 GB and has MD5 checksum 73843885b503af3c8b3ee62e5f5a9900. 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-passage-deepimpact \
  --corpus-path collections/msmarco-passage-deepimpact

Indexing

Sample indexing command:

target/appassembler/bin/IndexCollection \
  -collection JsonVectorCollection \
  -input /path/to/msmarco-passage-deepimpact \
  -generator DefaultLuceneDocumentGenerator \
  -index indexes/lucene-index.msmarco-passage-deepimpact/ \
  -threads 16 -impact -pretokenized \
  >& logs/log.msmarco-passage-deepimpact &

The path /path/to/msmarco-passage-deepimpact/ 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 doc lengths into Lucene's norms (which is the default) and the second option says not to apply any additional tokenization on the SPLADEv2 tokens. Upon completion, we should have an index with 8,841,823 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 6980 dev set questions; see this page for more details.

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

target/appassembler/bin/SearchCollection \
  -index indexes/lucene-index.msmarco-passage-deepimpact/ \
  -topics tools/topics-and-qrels/topics.msmarco-passage.dev-subset.deepimpact.tsv.gz \
  -topicReader TsvInt \
  -output runs/run.msmarco-passage-deepimpact.deepimpact.topics.msmarco-passage.dev-subset.deepimpact.txt \
  -impact -pretokenized &

Evaluation can be performed using trec_eval:

target/appassembler/bin/trec_eval -c -m map tools/topics-and-qrels/qrels.msmarco-passage.dev-subset.txt runs/run.msmarco-passage-deepimpact.deepimpact.topics.msmarco-passage.dev-subset.deepimpact.txt
target/appassembler/bin/trec_eval -c -M 10 -m recip_rank tools/topics-and-qrels/qrels.msmarco-passage.dev-subset.txt runs/run.msmarco-passage-deepimpact.deepimpact.topics.msmarco-passage.dev-subset.deepimpact.txt
target/appassembler/bin/trec_eval -c -m recall.100 tools/topics-and-qrels/qrels.msmarco-passage.dev-subset.txt runs/run.msmarco-passage-deepimpact.deepimpact.topics.msmarco-passage.dev-subset.deepimpact.txt
target/appassembler/bin/trec_eval -c -m recall.1000 tools/topics-and-qrels/qrels.msmarco-passage.dev-subset.txt runs/run.msmarco-passage-deepimpact.deepimpact.topics.msmarco-passage.dev-subset.deepimpact.txt

Effectiveness

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

AP@1000 DeepImpact
MS MARCO Passage: Dev 0.3334
RR@10 DeepImpact
MS MARCO Passage: Dev 0.3274
R@100 DeepImpact
MS MARCO Passage: Dev 0.8421
R@1000 DeepImpact
MS MARCO Passage: Dev 0.9476

The above runs are in TREC output format and evaluated with trec_eval. In order to reproduce results reported in the paper, we need to convert to MS MARCO output format and then evaluate:

python tools/scripts/msmarco/convert_trec_to_msmarco_run.py \
   --input runs/run.msmarco-passage-deepimpact.deepimpact.topics.msmarco-passage.dev-subset.deepimpact.txt \
   --output runs/run.msmarco-passage-deepimpact.deepimpact.topics.msmarco-passage.dev-subset.deepimpact.tsv --quiet

python tools/scripts/msmarco/msmarco_passage_eval.py \
   tools/topics-and-qrels/qrels.msmarco-passage.dev-subset.txt \
   runs/run.msmarco-passage-deepimpact.deepimpact.topics.msmarco-passage.dev-subset.deepimpact.tsv

The results should be as follows:

#####################
MRR @10: 0.3252764133351524
QueriesRanked: 6980
#####################

The final evaluation metric is very close to the one reported in the paper (0.326).

Reproduction Log*

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