/pairwise-ttest

Scripts to perform pairwise t-test on TREC run files

Primary LanguageShell

Pairwise t-test

Scripts to perform pairwise t-test on TREC run files.

Requirements

Usage

There are two bash scripts to run. First run pairwise-eval.sh to evaluate the TREC run files. Then run pairwise-ttest.sh to compute statistical significance.

The bash scripts assume that rbp_eval, gdeval.pl and trec_eval can be found in your PATH environment.

To compute a pairwise t-test of all run files in the runs directory for NDCG@10 using foo.qrels (which contains the relevance judgments), run the following:

./pairwise-eval.sh ndcg 10 foo.qrels runs/*.run
./pairwise-ttest.sh runs/*.run.ndcg10 > result.txt
cat result.txt

The pairwise-eval.sh script can compute ERR, NDCG, RBP and MAP. gdeval.pl is used for ERR and NDCG, rbp_eval for RBP, and trec_eval is used for MAP.