Connect-4 Game Playing Framework (C4GPF)
Connect-4 (Connect-Four) is an interesting game for several reasons. It is a non-trivial game, most humans will have difficulties to play well against a master level player (human or computer). At the same time it is of medium complexity (4.5·1012 states). It can be solved by tree-based methods (minimax, alpha-beta pruning).
Over the last years we have worked on a Java-based Connect-4 framework which makes it easily to develop, test and play with trainable Connect-4 agents. This work is now ready to be shared with other interested researchers or game-playing users. You can download it for your own research from this GitHub link.
"Why another Connect-4 program?", you might ask. There are many Connect-4 realizations, but – at least to our knowledge – no other one which incorporates a fast-playing perfect player. This player can be used to benchmark the strength of other, self-trained agents. We derived a variety of trainable agents (more details in our recent TCIAIG-paper "Online Adaptable Learning Rates for the Game Connect-4"). We have developed the framework in such a way that it is for an experienced Java programmer fairly easy to add his / her own agent. Once added, several methods for evaluation are at the disposal of the user.
So we want to encourage with this Connect-4 framework other researchers to benchmark their game-playing agents against other Connect-4 agents with well-known strength. If you are interested in having your Connect-4 agent being added to the C4GPF framework, feel free to contact us via eMail.
Features of C4GPF:
- built-in reinforcement learning agent (TD-learning)
- eligibility traces
- several adaptive step-size learning schemes: TCL, IDBD, …
- N-tuple features
- perfect-playing Minimax agent with alpha-beta pruning and opening book
- interface "Agent.java" for easy plug-in of new agents
- several benchmarking options (competitions, move inspections, …)
Getting started:
- Read the file CFour/READM.txt on GitHub
- Read the file CFour/src/doc/index.htm = CFour/src/doc/Help.pdf (help file for the GUI of C4GPF)
Authors of C4GPF:
- Markus Thill (markus.thill "at" fh-koeln.de)
- Wolfgang Konen (wolfgang.konen "at" fh-koeln.de)
Publications Games