GitHub - Larkinnjm1/Software-Agents-Reinforcement-learning-project: The purpose of the project was to generate a software agent to complete in a single and multi agent environment Auction game with Q-learning and SARSA along with the multi agent version of q learning Friend vs Foe Q-learning. Single agent performed without issue. Multi agent system did not converge due to the presence of degenerate game issues in the Q matrix due to the fact that the decision boundary for the auction game had non binary decisions.

Branches Tags

Name		Name	Last commit message	Last commit date
Latest commit History 1 Commit
Coding Scripts.zip		Coding Scripts.zip
INM426_SoftwareAgents_FinalSubmission.pdf		INM426_SoftwareAgents_FinalSubmission.pdf
ReadME.txt		ReadME.txt

Repository files navigation

File breakdown:
The folders for this project are broken down as follows:

Folder name: 
-Final Report
Contains:
-PDF file of final report on software agents

Folder name: 
Raw data and Scripts
Contains:
- Subfolders Raw Data and Scripts

Raw data and Scripts Subfolder name:
Raw Data 
Contains: 
Q learning Grid search Raw data results
Sarsa Raw Data results

Raw data and Scripts Subfolder name:
Scripts
Contains: 
Q learning,SARSA ,NASH equilibrium and convergence calculator scripts

About

The purpose of the project was to generate a software agent to complete in a single and multi agent environment Auction game with Q-learning and SARSA along with the multi agent version of q learning Friend vs Foe Q-learning. Single agent performed without issue. Multi agent system did not converge due to the presence of degenerate game issues i…

Readme