A contextual bandit benchmarking package.
Project description
Bandit Benchmarking
This repository contains python modules designed to benchmark the performance of contextual bandit algorithms.
Core Modules
The package has three core modules:
- simulations -- This module contains contextual bandit simulations with context, actions and rewards for those actions.
- learners -- This module contains common contextual bandit algorithms such as epsilon-greedy and upper confidence bound.
- benchmarks -- This module uses simulations to determine how well a learner performs and compare to other learners.
Core Interfaces
In order to make the package testable and extensible it has been built around three simple interfaces.
- The simulation interface
- The learner interface
- The benchmark interface
For most users the only interface that they will ever need to implement on their own is the learner interface. Implementing the learner interface on a custom algorithm allows one to leverage all simulations and benchmarks provided by the package to collect performance metrics.
Learning More
For those who wish to learn more the source code is perhaps the best place to start. All modules and classes have detailed docstrings and unittests.
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
