KaleabTessera / Multi-Armed-Bandit

Implementation of greedy, E-greedy and Upper Confidence Bound (UCB) algorithm on the Multi-Armed-Bandit problem.

Multi-Armed-Bandit

Description

This is an implementation of $\epsilon$-Greedy, Greedy and Upper Confidence Bound algorithms to solve the Multi-Armed Bandit problem. Implementation details of these algorithms can be found in Chapter 2 of Reinforcement Learning: An Introduction - Rich Sutton

How to Install:

# In project root folder
pip install -r requirements.txt

How to Run:

# In project root folder
./run.sh

Tasks

Part 1

A plot of reward over time (averaged over 100 runs each) on the same axes, for $\epsilon$-greedy with 𝜖 = 0.1, greedy with 𝑄1 = 5, and UCB with 𝑐 = 2.

Part 2

A summary comparison plot of rewards over first 1000 steps for the three algorithms with different values of the hyperparameters.