Skip to content

Latest commit

 

History

History
185 lines (155 loc) · 15.2 KB

regressions-neuclir22-zh-dt-splade.md

File metadata and controls

185 lines (155 loc) · 15.2 KB
Raw

Anserini Regressions: NeuCLIR22 — Chinese (Document Translation)

This page presents document translation regression experiments for the TREC 2022 NeuCLIR Track, Chinese, with the following configuration:

  • Queries: English
  • Documents: Machine-translated documents from Chinese into English (corpus provided by the organizers)
  • Model: SPLADE CoCondenser SelfDistil

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.

We make available a version of the corpus that has already been encoded with SPLADE CoCondenser SelfDistil, i.e., we performed model inference on every document and stored the output sparse vectors. Thus, no neural inference is required to reproduce these experiments; see instructions below.

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 neuclir22-zh-dt-splade

Corpus Download

Download the corpus and unpack into collections/:

wget https://rgw.cs.uwaterloo.ca/pyserini/data/neuclir22-zh-en-splade.tar -P collections/
tar xvf collections/neuclir22-zh-en-splade.tar -C collections/

To confirm, neuclir22-zh-en-splade.tar is 4.0 GB and has MD5 checksum 3ca6540bd4312db359975b9f90fad069. With the corpus downloaded, the following command will perform the remaining steps below:

python src/main/python/run_regression.py --index --verify --search --regression neuclir22-zh-dt-splade \
  --corpus-path collections/neuclir22-zh-en-splade

Indexing

Typical indexing command:

target/appassembler/bin/IndexCollection \
  -collection JsonVectorCollection \
  -input /path/to/neuclir22-zh-en-splade \
  -index indexes/lucene-index.neuclir22-zh-en-splade \
  -generator DefaultLuceneDocumentGenerator \
  -threads 8 -impact -pretokenized -storeRaw \
  >& logs/log.neuclir22-zh-en-splade &

For additional details, see explanation of common indexing options.

Retrieval

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

target/appassembler/bin/SearchCollection \
  -index indexes/lucene-index.neuclir22-zh-en-splade \
  -topics tools/topics-and-qrels/topics.neuclir22-en.splade.original-title.txt.gz \
  -topicreader TsvInt \
  -output runs/run.neuclir22-zh-en-splade.splade.topics.neuclir22-en.splade.original-title.txt \
  -impact -pretokenized &
target/appassembler/bin/SearchCollection \
  -index indexes/lucene-index.neuclir22-zh-en-splade \
  -topics tools/topics-and-qrels/topics.neuclir22-en.splade.original-desc.txt.gz \
  -topicreader TsvInt \
  -output runs/run.neuclir22-zh-en-splade.splade.topics.neuclir22-en.splade.original-desc.txt \
  -impact -pretokenized &
target/appassembler/bin/SearchCollection \
  -index indexes/lucene-index.neuclir22-zh-en-splade \
  -topics tools/topics-and-qrels/topics.neuclir22-en.splade.original-desc_title.txt.gz \
  -topicreader TsvInt \
  -output runs/run.neuclir22-zh-en-splade.splade.topics.neuclir22-en.splade.original-desc_title.txt \
  -impact -pretokenized &

target/appassembler/bin/SearchCollection \
  -index indexes/lucene-index.neuclir22-zh-en-splade \
  -topics tools/topics-and-qrels/topics.neuclir22-en.splade.original-title.txt.gz \
  -topicreader TsvInt \
  -output runs/run.neuclir22-zh-en-splade.splade+rm3.topics.neuclir22-en.splade.original-title.txt \
  -impact -pretokenized -rm3 -collection JsonVectorCollection &
target/appassembler/bin/SearchCollection \
  -index indexes/lucene-index.neuclir22-zh-en-splade \
  -topics tools/topics-and-qrels/topics.neuclir22-en.splade.original-desc.txt.gz \
  -topicreader TsvInt \
  -output runs/run.neuclir22-zh-en-splade.splade+rm3.topics.neuclir22-en.splade.original-desc.txt \
  -impact -pretokenized -rm3 -collection JsonVectorCollection &
target/appassembler/bin/SearchCollection \
  -index indexes/lucene-index.neuclir22-zh-en-splade \
  -topics tools/topics-and-qrels/topics.neuclir22-en.splade.original-desc_title.txt.gz \
  -topicreader TsvInt \
  -output runs/run.neuclir22-zh-en-splade.splade+rm3.topics.neuclir22-en.splade.original-desc_title.txt \
  -impact -pretokenized -rm3 -collection JsonVectorCollection &

target/appassembler/bin/SearchCollection \
  -index indexes/lucene-index.neuclir22-zh-en-splade \
  -topics tools/topics-and-qrels/topics.neuclir22-en.splade.original-title.txt.gz \
  -topicreader TsvInt \
  -output runs/run.neuclir22-zh-en-splade.splade+rocchio.topics.neuclir22-en.splade.original-title.txt \
  -impact -pretokenized -rocchio -collection JsonVectorCollection &
target/appassembler/bin/SearchCollection \
  -index indexes/lucene-index.neuclir22-zh-en-splade \
  -topics tools/topics-and-qrels/topics.neuclir22-en.splade.original-desc.txt.gz \
  -topicreader TsvInt \
  -output runs/run.neuclir22-zh-en-splade.splade+rocchio.topics.neuclir22-en.splade.original-desc.txt \
  -impact -pretokenized -rocchio -collection JsonVectorCollection &
target/appassembler/bin/SearchCollection \
  -index indexes/lucene-index.neuclir22-zh-en-splade \
  -topics tools/topics-and-qrels/topics.neuclir22-en.splade.original-desc_title.txt.gz \
  -topicreader TsvInt \
  -output runs/run.neuclir22-zh-en-splade.splade+rocchio.topics.neuclir22-en.splade.original-desc_title.txt \
  -impact -pretokenized -rocchio -collection JsonVectorCollection &

Evaluation can be performed using trec_eval:

tools/eval/trec_eval.9.0.4/trec_eval -c -m ndcg_cut.20 tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade.topics.neuclir22-en.splade.original-title.txt
python -m pyserini.eval.trec_eval -c -m judged.20 tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade.topics.neuclir22-en.splade.original-title.txt
tools/eval/trec_eval.9.0.4/trec_eval -c -m recall.1000 tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade.topics.neuclir22-en.splade.original-title.txt
tools/eval/trec_eval.9.0.4/trec_eval -c -m map tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade.topics.neuclir22-en.splade.original-title.txt
tools/eval/trec_eval.9.0.4/trec_eval -c -m ndcg_cut.20 tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade.topics.neuclir22-en.splade.original-desc.txt
python -m pyserini.eval.trec_eval -c -m judged.20 tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade.topics.neuclir22-en.splade.original-desc.txt
tools/eval/trec_eval.9.0.4/trec_eval -c -m recall.1000 tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade.topics.neuclir22-en.splade.original-desc.txt
tools/eval/trec_eval.9.0.4/trec_eval -c -m map tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade.topics.neuclir22-en.splade.original-desc.txt
tools/eval/trec_eval.9.0.4/trec_eval -c -m ndcg_cut.20 tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade.topics.neuclir22-en.splade.original-desc_title.txt
python -m pyserini.eval.trec_eval -c -m judged.20 tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade.topics.neuclir22-en.splade.original-desc_title.txt
tools/eval/trec_eval.9.0.4/trec_eval -c -m recall.1000 tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade.topics.neuclir22-en.splade.original-desc_title.txt
tools/eval/trec_eval.9.0.4/trec_eval -c -m map tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade.topics.neuclir22-en.splade.original-desc_title.txt

tools/eval/trec_eval.9.0.4/trec_eval -c -m ndcg_cut.20 tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade+rm3.topics.neuclir22-en.splade.original-title.txt
python -m pyserini.eval.trec_eval -c -m judged.20 tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade+rm3.topics.neuclir22-en.splade.original-title.txt
tools/eval/trec_eval.9.0.4/trec_eval -c -m recall.1000 tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade+rm3.topics.neuclir22-en.splade.original-title.txt
tools/eval/trec_eval.9.0.4/trec_eval -c -m map tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade+rm3.topics.neuclir22-en.splade.original-title.txt
tools/eval/trec_eval.9.0.4/trec_eval -c -m ndcg_cut.20 tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade+rm3.topics.neuclir22-en.splade.original-desc.txt
python -m pyserini.eval.trec_eval -c -m judged.20 tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade+rm3.topics.neuclir22-en.splade.original-desc.txt
tools/eval/trec_eval.9.0.4/trec_eval -c -m recall.1000 tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade+rm3.topics.neuclir22-en.splade.original-desc.txt
tools/eval/trec_eval.9.0.4/trec_eval -c -m map tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade+rm3.topics.neuclir22-en.splade.original-desc.txt
tools/eval/trec_eval.9.0.4/trec_eval -c -m ndcg_cut.20 tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade+rm3.topics.neuclir22-en.splade.original-desc_title.txt
python -m pyserini.eval.trec_eval -c -m judged.20 tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade+rm3.topics.neuclir22-en.splade.original-desc_title.txt
tools/eval/trec_eval.9.0.4/trec_eval -c -m recall.1000 tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade+rm3.topics.neuclir22-en.splade.original-desc_title.txt
tools/eval/trec_eval.9.0.4/trec_eval -c -m map tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade+rm3.topics.neuclir22-en.splade.original-desc_title.txt

tools/eval/trec_eval.9.0.4/trec_eval -c -m ndcg_cut.20 tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade+rocchio.topics.neuclir22-en.splade.original-title.txt
python -m pyserini.eval.trec_eval -c -m judged.20 tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade+rocchio.topics.neuclir22-en.splade.original-title.txt
tools/eval/trec_eval.9.0.4/trec_eval -c -m recall.1000 tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade+rocchio.topics.neuclir22-en.splade.original-title.txt
tools/eval/trec_eval.9.0.4/trec_eval -c -m map tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade+rocchio.topics.neuclir22-en.splade.original-title.txt
tools/eval/trec_eval.9.0.4/trec_eval -c -m ndcg_cut.20 tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade+rocchio.topics.neuclir22-en.splade.original-desc.txt
python -m pyserini.eval.trec_eval -c -m judged.20 tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade+rocchio.topics.neuclir22-en.splade.original-desc.txt
tools/eval/trec_eval.9.0.4/trec_eval -c -m recall.1000 tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade+rocchio.topics.neuclir22-en.splade.original-desc.txt
tools/eval/trec_eval.9.0.4/trec_eval -c -m map tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade+rocchio.topics.neuclir22-en.splade.original-desc.txt
tools/eval/trec_eval.9.0.4/trec_eval -c -m ndcg_cut.20 tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade+rocchio.topics.neuclir22-en.splade.original-desc_title.txt
python -m pyserini.eval.trec_eval -c -m judged.20 tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade+rocchio.topics.neuclir22-en.splade.original-desc_title.txt
tools/eval/trec_eval.9.0.4/trec_eval -c -m recall.1000 tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade+rocchio.topics.neuclir22-en.splade.original-desc_title.txt
tools/eval/trec_eval.9.0.4/trec_eval -c -m map tools/topics-and-qrels/qrels.neuclir22-zh.txt runs/run.neuclir22-zh-en-splade.splade+rocchio.topics.neuclir22-en.splade.original-desc_title.txt

Effectiveness

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

MAP SPLADE +RM3 +Rocchio
NeuCLIR 2022 (Chinese): title (original English queries) 0.3068 0.2843 0.3151
NeuCLIR 2022 (Chinese): desc (original English queries) 0.3108 0.2631 0.3108
NeuCLIR 2022 (Chinese): desc+title (original English queries) 0.3034 0.2820 0.3092
nDCG@20 SPLADE +RM3 +Rocchio
NeuCLIR 2022 (Chinese): title (original English queries) 0.4233 0.3816 0.4204
NeuCLIR 2022 (Chinese): desc (original English queries) 0.4299 0.3496 0.4142
NeuCLIR 2022 (Chinese): desc+title (original English queries) 0.4236 0.3770 0.4206
J@20 SPLADE +RM3 +Rocchio
NeuCLIR 2022 (Chinese): title (original English queries) 0.3851 0.3575 0.3860
NeuCLIR 2022 (Chinese): desc (original English queries) 0.3825 0.3531 0.3825
NeuCLIR 2022 (Chinese): desc+title (original English queries) 0.3961 0.3675 0.3969
Recall@1000 SPLADE +RM3 +Rocchio
NeuCLIR 2022 (Chinese): title (original English queries) 0.7997 0.7546 0.8038
NeuCLIR 2022 (Chinese): desc (original English queries) 0.7597 0.6969 0.7623
NeuCLIR 2022 (Chinese): desc+title (original English queries) 0.7922 0.7481 0.8067

Reproduction Log*

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