Skip to content

Latest commit

 

History

History
76 lines (60 loc) · 4.08 KB

regressions-mrtydi-v1.1-sw.md

File metadata and controls

76 lines (60 loc) · 4.08 KB
Raw

Anserini Regressions: Mr. TyDi (v1.1) — Swahili

This page documents BM25 regression experiments for Mr. TyDi (v1.1) — Swahili.

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 mrtydi-v1.1-sw

Indexing

Typical indexing command:

target/appassembler/bin/IndexCollection \
  -collection MrTyDiCollection \
  -input /path/to/mrtydi-v1.1-sw \
  -index indexes/lucene-index.mrtydi-v1.1-swahili/ \
  -generator DefaultLuceneDocumentGenerator \
  -threads 1 -storePositions -storeDocvectors -storeRaw -language sw \
  >& logs/log.mrtydi-v1.1-sw &

See this page for more details about the Mr. TyDi corpus. 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.mrtydi-v1.1-swahili/ \
  -topics tools/topics-and-qrels/topics.mrtydi-v1.1-sw.train.txt.gz \
  -topicreader TsvInt \
  -output runs/run.mrtydi-v1.1-sw.bm25.topics.mrtydi-v1.1-sw.train.txt \
  -bm25 -hits 100 -language sw &
target/appassembler/bin/SearchCollection \
  -index indexes/lucene-index.mrtydi-v1.1-swahili/ \
  -topics tools/topics-and-qrels/topics.mrtydi-v1.1-sw.dev.txt.gz \
  -topicreader TsvInt \
  -output runs/run.mrtydi-v1.1-sw.bm25.topics.mrtydi-v1.1-sw.dev.txt \
  -bm25 -hits 100 -language sw &
target/appassembler/bin/SearchCollection \
  -index indexes/lucene-index.mrtydi-v1.1-swahili/ \
  -topics tools/topics-and-qrels/topics.mrtydi-v1.1-sw.test.txt.gz \
  -topicreader TsvInt \
  -output runs/run.mrtydi-v1.1-sw.bm25.topics.mrtydi-v1.1-sw.test.txt \
  -bm25 -hits 100 -language sw &

Evaluation can be performed using trec_eval:

tools/eval/trec_eval.9.0.4/trec_eval -c -M 100 -m recip_rank -c -m recall.100 tools/topics-and-qrels/qrels.mrtydi-v1.1-sw.train.txt runs/run.mrtydi-v1.1-sw.bm25.topics.mrtydi-v1.1-sw.train.txt
tools/eval/trec_eval.9.0.4/trec_eval -c -M 100 -m recip_rank -c -m recall.100 tools/topics-and-qrels/qrels.mrtydi-v1.1-sw.dev.txt runs/run.mrtydi-v1.1-sw.bm25.topics.mrtydi-v1.1-sw.dev.txt
tools/eval/trec_eval.9.0.4/trec_eval -c -M 100 -m recip_rank -c -m recall.100 tools/topics-and-qrels/qrels.mrtydi-v1.1-sw.test.txt runs/run.mrtydi-v1.1-sw.bm25.topics.mrtydi-v1.1-sw.test.txt

Effectiveness

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

MRR@100 BM25
Mr. TyDi (Swahili): train 0.2610
Mr. TyDi (Swahili): dev 0.2693
Mr. TyDi (Swahili): test 0.3893
R@100 BM25
Mr. TyDi (Swahili): train 0.5903
Mr. TyDi (Swahili): dev 0.5789
Mr. TyDi (Swahili): test 0.7642