Skip to content

Latest commit

 

History

History
140 lines (103 loc) · 10.1 KB

regressions-dl22-passage.splade-pp-ed.cached.md

File metadata and controls

140 lines (103 loc) · 10.1 KB

Anserini Regressions: TREC 2022 Deep Learning Track (Passage)

Model: SPLADE++ CoCondenser-EnsembleDistil (using cached queries)

This page describes regression experiments, integrated into Anserini's regression testing framework, applying the SPLADE++ CoCondenser-EnsembleDistil model to the MS MARCO V2 passage corpus. Here, we evaluate on the TREC 2022 Deep Learning Track passage ranking task, using cached queries (i.e., cached results of query encoding).

The model is described in the following paper:

Thibault Formal, Carlos Lassance, Benjamin Piwowarski, and Stéphane Clinchant. From Distillation to Hard Negative Sampling: Making Sparse Neural IR Models More Effective. Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval, pages 2353–2359.

For additional instructions on working with the MS MARCO V2 passage corpus, refer to this page.

Note that the NIST relevance judgments provide far more relevant passages per topic, unlike the "sparse" judgments provided by Microsoft (these are sometimes called "dense" judgments to emphasize this contrast).

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 dl22-passage.splade-pp-ed.cached

We make available a version of the corpus that has already been encoded with SPLADE++ CoCondenser-EnsembleDistil.

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 dl22-passage.splade-pp-ed.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_v2_passage_splade_pp_ed.tar -P collections/
tar xvf collections/msmarco_v2_passage_splade_pp_ed.tar -C collections/

To confirm, msmarco_v2_passage_splade_pp_ed.tar is 66 GB and has MD5 checksum 2cdb2adc259b8fa6caf666b20ebdc0e8. With the corpus downloaded, the following command will perform the remaining steps below:

python src/main/python/run_regression.py --index --verify --search --regression dl22-passage.splade-pp-ed.cached \
  --corpus-path collections/msmarco_v2_passage_splade_pp_ed

Indexing

Sample indexing command:

bin/run.sh io.anserini.index.IndexCollection \
  -threads 24 \
  -collection JsonVectorCollection \
  -input /path/to/msmarco-v2-passage-splade-pp-ed \
  -generator DefaultLuceneDocumentGenerator \
  -index indexes/lucene-inverted.msmarco-v2-passage.splade-pp-ed/ \
  -impact -pretokenized -storeRaw \
  >& logs/log.msmarco-v2-passage-splade-pp-ed &

The path /path/to/msmarco-v2-passage-splade-pp-ed/ 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 138,364,198 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 76 topics for which NIST has provided judgments as part of the TREC 2022 Deep Learning Track.

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-v2-passage.splade-pp-ed/ \
  -topics tools/topics-and-qrels/topics.dl22.splade-pp-ed.tsv.gz \
  -topicReader TsvInt \
  -output runs/run.msmarco-v2-passage-splade-pp-ed.splade-pp-ed-cached.topics.dl22.splade-pp-ed.txt \
  -parallelism 16 -impact -pretokenized &

bin/run.sh io.anserini.search.SearchCollection \
  -index indexes/lucene-inverted.msmarco-v2-passage.splade-pp-ed/ \
  -topics tools/topics-and-qrels/topics.dl22.splade-pp-ed.tsv.gz \
  -topicReader TsvInt \
  -output runs/run.msmarco-v2-passage-splade-pp-ed.splade-pp-ed-cached+rm3.topics.dl22.splade-pp-ed.txt \
  -parallelism 16 -impact -pretokenized -rm3 -collection JsonVectorCollection &

bin/run.sh io.anserini.search.SearchCollection \
  -index indexes/lucene-inverted.msmarco-v2-passage.splade-pp-ed/ \
  -topics tools/topics-and-qrels/topics.dl22.splade-pp-ed.tsv.gz \
  -topicReader TsvInt \
  -output runs/run.msmarco-v2-passage-splade-pp-ed.splade-pp-ed-cached+rocchio.topics.dl22.splade-pp-ed.txt \
  -parallelism 16 -impact -pretokenized -rocchio -collection JsonVectorCollection &

Evaluation can be performed using trec_eval:

bin/trec_eval -c -M 100 -m map -l 2 tools/topics-and-qrels/qrels.dl22-passage.txt runs/run.msmarco-v2-passage-splade-pp-ed.splade-pp-ed-cached.topics.dl22.splade-pp-ed.txt
bin/trec_eval -c -M 100 -m recip_rank -l 2 tools/topics-and-qrels/qrels.dl22-passage.txt runs/run.msmarco-v2-passage-splade-pp-ed.splade-pp-ed-cached.topics.dl22.splade-pp-ed.txt
bin/trec_eval -c -m ndcg_cut.10 tools/topics-and-qrels/qrels.dl22-passage.txt runs/run.msmarco-v2-passage-splade-pp-ed.splade-pp-ed-cached.topics.dl22.splade-pp-ed.txt
bin/trec_eval -c -m recall.100 -l 2 tools/topics-and-qrels/qrels.dl22-passage.txt runs/run.msmarco-v2-passage-splade-pp-ed.splade-pp-ed-cached.topics.dl22.splade-pp-ed.txt
bin/trec_eval -c -m recall.1000 -l 2 tools/topics-and-qrels/qrels.dl22-passage.txt runs/run.msmarco-v2-passage-splade-pp-ed.splade-pp-ed-cached.topics.dl22.splade-pp-ed.txt

bin/trec_eval -c -M 100 -m map -l 2 tools/topics-and-qrels/qrels.dl22-passage.txt runs/run.msmarco-v2-passage-splade-pp-ed.splade-pp-ed-cached+rm3.topics.dl22.splade-pp-ed.txt
bin/trec_eval -c -M 100 -m recip_rank -l 2 tools/topics-and-qrels/qrels.dl22-passage.txt runs/run.msmarco-v2-passage-splade-pp-ed.splade-pp-ed-cached+rm3.topics.dl22.splade-pp-ed.txt
bin/trec_eval -c -m ndcg_cut.10 tools/topics-and-qrels/qrels.dl22-passage.txt runs/run.msmarco-v2-passage-splade-pp-ed.splade-pp-ed-cached+rm3.topics.dl22.splade-pp-ed.txt
bin/trec_eval -c -m recall.100 -l 2 tools/topics-and-qrels/qrels.dl22-passage.txt runs/run.msmarco-v2-passage-splade-pp-ed.splade-pp-ed-cached+rm3.topics.dl22.splade-pp-ed.txt
bin/trec_eval -c -m recall.1000 -l 2 tools/topics-and-qrels/qrels.dl22-passage.txt runs/run.msmarco-v2-passage-splade-pp-ed.splade-pp-ed-cached+rm3.topics.dl22.splade-pp-ed.txt

bin/trec_eval -c -M 100 -m map -l 2 tools/topics-and-qrels/qrels.dl22-passage.txt runs/run.msmarco-v2-passage-splade-pp-ed.splade-pp-ed-cached+rocchio.topics.dl22.splade-pp-ed.txt
bin/trec_eval -c -M 100 -m recip_rank -l 2 tools/topics-and-qrels/qrels.dl22-passage.txt runs/run.msmarco-v2-passage-splade-pp-ed.splade-pp-ed-cached+rocchio.topics.dl22.splade-pp-ed.txt
bin/trec_eval -c -m ndcg_cut.10 tools/topics-and-qrels/qrels.dl22-passage.txt runs/run.msmarco-v2-passage-splade-pp-ed.splade-pp-ed-cached+rocchio.topics.dl22.splade-pp-ed.txt
bin/trec_eval -c -m recall.100 -l 2 tools/topics-and-qrels/qrels.dl22-passage.txt runs/run.msmarco-v2-passage-splade-pp-ed.splade-pp-ed-cached+rocchio.topics.dl22.splade-pp-ed.txt
bin/trec_eval -c -m recall.1000 -l 2 tools/topics-and-qrels/qrels.dl22-passage.txt runs/run.msmarco-v2-passage-splade-pp-ed.splade-pp-ed-cached+rocchio.topics.dl22.splade-pp-ed.txt

Effectiveness

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

MAP@100 SPLADE++ CoCondenser-EnsembleDistil +RM3 +Rocchio
DL22 (Passage) 0.1846 0.1930 0.1968
MRR@100 SPLADE++ CoCondenser-EnsembleDistil +RM3 +Rocchio
DL22 (Passage) 0.7005 0.7161 0.6938
nDCG@10 SPLADE++ CoCondenser-EnsembleDistil +RM3 +Rocchio
DL22 (Passage) 0.5705 0.5711 0.5897
R@100 SPLADE++ CoCondenser-EnsembleDistil +RM3 +Rocchio
DL22 (Passage) 0.3742 0.3561 0.3822
R@1000 SPLADE++ CoCondenser-EnsembleDistil +RM3 +Rocchio
DL22 (Passage) 0.6629 0.6367 0.6778