Paper

Overcoming Prior Misspecification in Online Learning to Rank

Installation

Run pip install -r reqs.txt

Usage

This module contains the synthetic experiments and algorithms source codes. For all the synthetic experiments, we use

$cd <project dir>
$python synthetic.py [ex_type] [expr_num]

For each experiment, set the variables in the main() function of synthetic.py as follows

• Non-contextual experiment

  $python synthetic --ex_type=stand_ex_type --expr_num=1

• Prior initialization experiment

  $python synthetic --ex_type=stand_ex_type --expr_num=7

• Prior misspecification experiment (Fig 3)

  $python synthetic --ex_type=stand_ex_type --expr_num=5

• Linear contextual experiments (Fig 4)

  $python synthetic --ex_type=linear_ex_type

• Logistic contextual experiments (Fig 5)

  $python synthetic --ex_type=log_ex_type

Notes:

• You can use multiprocessing by setting parr=1. Note that this might run into a deadlock due to memory issues. See examples here.
• After an experiment is run, the result is saved in a pickle file and the plot is generated in PDF format.