Synthetic Personalization Environment

This repository contains the source code for the numerical experiments presented in the paper On the Unreasonable Efficiency of State Space Clustering in Personalization Tasks.

Installation

Install the requirements via pip install -r requirements.txt
Configure the environment in config.yaml
Run the experiments via python -m run_experiments

Files Overview

synthetic_gaussian_mapping.py --- creates the Synthetic Gaussian Mapping that acts as a latent feature extractor for the simulated reward signal
bandit_environment.py --- creates the Synthetic Hyperpersonalization Environment with the simulated reward signal as an OpenAI Gym environment
online_rl.py --- trains online RL algorithms on a given environment
run_experiments.py --- sets up and runs the experiments
config.yaml --- stores the environment/training/experiment parameters
requirements.txt --- lists the required packages

License

This project is licensed under the MIT License.