search

druzil / nggp-base

A partial port of the ggp-base code-base to .net
BSD 3-Clause "New" or "Revised" License
1 stars 0 forks source link

Partial #1

Open tomitrescak opened 6 years ago

tomitrescak commented 6 years ago

Dear @druzil thanks for this fantastic effort, you saved us a lot of time. I was wondering how "partial" is this? That is, what is included and what is missing?

Thanks

druzil commented 6 years ago

The java ggp-base codebase is quite expansive, it includes quite a few GUI's for easy of use. I have not implemented these. Also I tried to provide the base functionality for someone who would like to implement a bot as a student to learn, without 'giving away' the prop-net (eg forward or backwards) or search methodologies (monte carlo, adversarial, etc). While I do have implementations of these, providing them would remove the learning element that I believe the ggp competition provides. I'm curious what your intention is? Incidentally I noticed you are based in Sydney, so am I :)

tomitrescak commented 6 years ago

That’s great news mate, I’d invite you for a beer as this will save us a lot of time;) I am AI researcher based in WSU. Currently we are working on a multi agent planning and coordination mechanism. I partnered up with Dave De Jonge who is a guru in this and did fair bit of it in GGP. We are applying this to virtual simulations and our virtual institutions technology that need to run in Unity 3D, so we are searching for a solution how to achieve this. With your client we can now connect to GGP server and run those simulations from unity. I still have not tested it yet though as unity is a bit particular on choice of .net. I’d love to have the whole GGP running in Unity (server as well) but that may be quite complex to achieve. What’s your interest in ggp?

druzil commented 6 years ago

Your research sounds very interesting. I'd definitely keen on meeting up for a chat. My interest in GGP is AI for modern board games. Besides an AI agent learning a good strategy, I'm interested in using it as an analysis tool for both improving play via game evaluation and as a tool in assisting game design. The recent development around AlphaZero is of particular interest to me. However, I still find that the methodology that is used in this instance, while being able to produce super elite playing agents are still opaque from a human instance in how an agent is able to evaluation particular game states that understandable to a human. As both neural nets and monte carlo search is not something that humans can employ in there own game playing.