Skip to content

Latest commit

 

History

History
114 lines (78 loc) · 7.26 KB

regressions-msmarco-passage-splade-distil-cocodenser-medium.md

File metadata and controls

114 lines (78 loc) · 7.26 KB
Raw

Anserini Regressions: MS MARCO Passage Ranking

Model: SPLADE-distil CoCodenser Medium

This page describes regression experiments, integrated into Anserini's regression testing framework, using the SPLADE-distil CoCodenser Medium model on the MS MARCO passage ranking task. The SPLADE-distil CoCodenser Medium model is open-sourced by Naver Labs Europe.

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-passage-splade-distil-cocodenser-medium

Corpus

We make available a version of the MS MARCO passage corpus that has already been processed with the model (i.e., with inference applied to generate the lexical representations). Thus, no neural inference is involved. For details on how to train SPLADE-distil CoCodenser Medium and perform inference, please see guide provided by Naver Labs Europe.

Download the corpus and unpack into collections/:

wget https://rgw.cs.uwaterloo.ca/JIMMYLIN-bucket0/data/msmarco-passage-splade_distil_cocodenser_medium.tar -P collections/
tar xvf collections/msmarco-passage-splade_distil_cocodenser_medium.tar -C collections/

To confirm, msmarco.tar is 4.9 GB and has MD5 checksum 54a81e855a7678bc83ecb3ecf1ac5c1c.

With the corpus downloaded, the following command will perform the complete regression, end to end, on any machine:

python src/main/python/run_regression.py --index --verify --search \
  --regression msmarco-passage-splade-distil-cocodenser-medium \
  --corpus-path collections/msmarco-passage-splade_distil_cocodenser_medium

Alternatively, you can simply copy/paste from the commands below and obtain the same results.

Indexing

Sample indexing command:

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

The path /path/to/msmarco-passage-splade_distil_cocodenser_medium/ 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 pre-encoded 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 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:

target/appassembler/bin/SearchCollection \
  -index indexes/lucene-index.msmarco-passage-splade_distil_cocodenser_medium/ \
  -topics src/main/resources/topics-and-qrels/topics.msmarco-passage.dev-subset.splade_distil_cocodenser_medium.tsv.gz \
  -topicreader TsvInt \
  -output runs/run.msmarco-passage-splade_distil_cocodenser_medium.splade_distil_cocodenser_medium.topics.msmarco-passage.dev-subset.splade_distil_cocodenser_medium.txt \
  -impact -pretokenized &

Evaluation can be performed using trec_eval:

tools/eval/trec_eval.9.0.4/trec_eval -c -m map src/main/resources/topics-and-qrels/qrels.msmarco-passage.dev-subset.txt runs/run.msmarco-passage-splade_distil_cocodenser_medium.splade_distil_cocodenser_medium.topics.msmarco-passage.dev-subset.splade_distil_cocodenser_medium.txt
tools/eval/trec_eval.9.0.4/trec_eval -c -M 10 -m recip_rank src/main/resources/topics-and-qrels/qrels.msmarco-passage.dev-subset.txt runs/run.msmarco-passage-splade_distil_cocodenser_medium.splade_distil_cocodenser_medium.topics.msmarco-passage.dev-subset.splade_distil_cocodenser_medium.txt
tools/eval/trec_eval.9.0.4/trec_eval -c -m recall.100 src/main/resources/topics-and-qrels/qrels.msmarco-passage.dev-subset.txt runs/run.msmarco-passage-splade_distil_cocodenser_medium.splade_distil_cocodenser_medium.topics.msmarco-passage.dev-subset.splade_distil_cocodenser_medium.txt
tools/eval/trec_eval.9.0.4/trec_eval -c -m recall.1000 src/main/resources/topics-and-qrels/qrels.msmarco-passage.dev-subset.txt runs/run.msmarco-passage-splade_distil_cocodenser_medium.splade_distil_cocodenser_medium.topics.msmarco-passage.dev-subset.splade_distil_cocodenser_medium.txt

Effectiveness

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

AP@1000 SPLADE-distill CoCodenser Medium
MS MARCO Passage: Dev 0.3943
RR@10 SPLADE-distill CoCodenser Medium
MS MARCO Passage: Dev 0.3892
R@100 SPLADE-distill CoCodenser Medium
MS MARCO Passage: Dev 0.9111
R@1000 SPLADE-distill CoCodenser Medium
MS MARCO Passage: Dev 0.9817

Reproduction Log*

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