Code to reproduce experiments in paper "The Weighted Kendall and High-order Kernels for Permutations"
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
R
results
src
.gitignore
README.md

README.md

weightedkendall

This repo contains R code for reproducing results from the paper:

Yunlong Jiao, Jean-Philippe Vert. "The Weighted Kendall and High-order Kernels for Permutations." arXiv preprint arXiv:1802.08526, 2018. arXiv:1802.08526

Quick Start

See the compiled results/notebook.md for data, code and results for the numerical experiments of this study.

Directories

The top level structure is as follows:

  • data/ - static RData to be studied, including
    • fulldat_eubm.RData the full Eurobarometer 55.2 data (provided upon request, or freely accessible via DOI:10.3886/ICPSR03341.v3).
    • dat_eubm.RData the anonymized, randomized, and subsampled rank data to run the binary classification (provided upon request).
  • results/ - notebook of experimental results, including
    • notebook.[md|Rmd] the project notebook of experiments, with code in notebook.Rmd, compiled version in notebook.md and figures saved in notebook_files/.
    • scores.txt table of performance scores per experiment
    • weights.txt matrix of learned weights in a weighted kernel (see paper and notebook for details)
  • R/ - R code and general purpose scripts, including
    • func.R implements utile functions and classifiers for rank data.
  • src/ - C++ code and general purpose scripts, including
    • dots.cpp implements some dot (inner product) function for rank data.

Hands-on

Note: If you would like to reproduce the experiments, first request data access to the Eurobarometer 55.2 survey through the website DOI:10.3886/ICPSR03341.v3, then you should be able to process the data locally. Or alternatively send me an email (with data access confirmation attached), I will be happy to provide you the processed dataset.

In order to build the project notebook results/notebook.md, make sure your local machine has the following R packages installed (or run the corresponding commands to install them in R console):

> require(kernrank)   # devtools::install_github("YunlongJiao/kernrank")
> require(kernlab)    # install.packages("kernlab")
> require(caret)      # install.packages("caret")
> require(ggplot2)    # install.packages("ggplot2")
> require(corrplot)   # install.packages("corrplot")
> require(rmarkdown)  # install.packages("rmarkdown")

Then run in shell,

$ git clone git@github.com:YunlongJiao/weightedkendall.git
$ cd weightedkendall/results/
$ Rscript -e "rmarkdown::render('notebook.Rmd', output_format = 'html_document')"

Authors

  • Yunlong Jiao - main contributor