Skip to content


Switch branches/tags

Name already in use

A tag already exists with the provided branch name. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Are you sure you want to create this branch?

Latest commit


Git stats


Failed to load latest commit information.
Latest commit message
Commit time

This repository contains the data and source code for the following paper:

A single ZIP file can be downloaded as well.

Project Structure

  • data/ Input data files.
  • output/ Generated output files.
  • src/ Source code in R.
  • scratch/ Temporary files generated in the process.

All code is written for R. You will need the following packages installed from CRAN: simIReff, VineCopula, rvinecopulib, doParallel.

How to reproduce the results in the paper

The source files in src/ need to be run in order. You can run each file individually by running Rscript src/<file>.R. They will store intermediate data in scratch/ and the final data in output/.

It is important that you always run from the base directory.

  1. src/01-margins.R fits all the marginal distributions (section 4.1).
  2. src/02-transform.R transforms all distributtions fitted in step 1 to have the same expected value as in the original data (section 4.1).
  3. src/03-margins_select.R selects the best distributions from step 1 (section 4.1).
  4. src/04-transform_select.R selects the best distributions from step 2 (section 4.2).
  5. src/05-bicop_types.R fits all possible bivariate copulas (section 4.2).
  6. src/06-cop_fit.R fits all the full gaussian and R-vine copulas (section 4.2).
  7. src/07-cop_sim.R simulates from the full copulas in step 6 and records observed means and variances (section 4.3).
  8. src/10-power.R runs the first sample application (section 5.1).
  9. src/11-type_I.R runs the second sample application (section 5.2).
  10. src/ generates all figures in the paper and stores them in output/figs/.

It takes several days to run all the code, so it is ready to run in parallel. Most of the above code parallelizes using function foreach in R's package doParallel. In particular, it will use all available cores in the machine except one. Check the individual code files to modify this behavior.

Custom test collections

You can easily run the code with your own initial test collection. Add the matrix of topic-by-system scores in data/ using the name <collection>_<measure>.csv (see for instance file data/web2012_ap.csv). Then, edit file src/common.R to add the new data:

.COLLECTIONS <- c("web2010", "web2011", "web2012", "web2013", "web2014")
.MEASURES <- c("ap", "ndcg20", "err20", "p10", "p20", "rr")

Note that the code will run for all combinations of collection and measure. For more specific modifications, edit the corresponding source file in src/ (see above). Note also that the script src/99-paper.R is only intended to generate the figures in the paper. If you customize something and want a similar analysis, you will need to extend this script yourself.

How to simulate your own data

A full R package has been developed for this: simIReff. Please refer to it's documentation for details. The simplest use case would be something like:

d <- read.csv("data/web2012_p10.csv") # read original data
effs <- effDiscFitAndSelect(d, support("p10")) # fit and select margins
cop <- effcopFit(d, effs) # fit copula

y <- reffcop(1000, cop) # simulate new 1000 topics
# compare observed vs. expected mean
expected <- sapply(effs, function(e) e$mean)
observed <- colMeans(y)
plot(expected, observed)


When using this archive, please cite the above paper:

  author = {Urbano, Juli\'{a}n and Nagler, Thomas},
  booktitle = {International ACM SIGIR Conference on Research and Development in Information Retrieval},
  title = {{Stochastic Simulation of Test Collections: Evaluation Scores}},
  pages = {695--704},
  year = {2018}


Reproduce results from the SIGIR 2018 paper "Stochastic Simulation of Test Collections: Evaluation Scores"




CC-BY-SA-4.0, MIT licenses found

Licenses found






No packages published