openEtG

[andretimpa]

What graphics are needed? 2 players are needed to test it?

[serprex]

Open etg.dek.im in two tabs and you can match against yourself. CuCN's been finding lots of issues by running 12QP/6 SoSe

As for graphics: everything. Status effects, element icons, card templates, card pictures. I'm unsure about effects, that'd require the engine have a way to tell the UI to start an animation (Bolting targets, healing glow, etc)
Right now openEtG is a graphically a bunch of text and text over colored rectangles

[EmeraldTiger]

Will this be able to test card ideas in the future?

[serprex]

It can test card ideas now. Run updatedb.py or download sheets from here as csv. Add card info to respective sheet, add any new active to Actives in etg.js. Passives/statuses are implemented rather poorly, as they have to be implemented in game flow

Granted that may seem cumbersome. Ideally passives could hook to events and there'd be a scripting language to describe effects which could be encoded in base64 bytecode in a spreadsheet. Lots of options on how to make this more generic, which should be explored eventually, though currently all my effort is in getting the base Elements card set working

[Gandora]

So, I'd like to add this to my Eclipse Workspace. I already installed EGit, but I don't know how to proceed from there. Can someone help me out?

Ok, I think I somehow did it. Got the project in eclipse now. Got to find out how to collaborate and how to use github now ^^

[Zso_Zso]

In the OP you mention no AI. Both as a programmer and gamer, I have been fascinated by AI development, but never really had the time to invest in it (apart from a fun project long time ago in a 3D tic-tact-toe game).

Mind, you I am not so keen on re-programming a pre-scripted AI to behave the same / similar as the current one in EtG. However, I would be interested in exploring some rather different avenues, such as:

• user-script driven AI to be supplied per-deck, i.e. implement a scripting language in which one could describe how a specific deck should be played, this could lead to AI3/FG decks being played much smarter and also to an Arena, where the player who submits the deck would also submit how to play it.
• Dynamic learning AI, that would record PvP games, analyze patterns to determine what moves lead to victories for various deck archetypes, play-styles  depending also on the opponent decks.
• Optimum response calculation based AI. This would "cheat" in the sense that it would peak at the whole deck of the opponent and calculate what possible moves the opponent can make in subsequent rounds, building a decision-tree like in a chess-AI to determine what's the best move.

So, I'd be interested in exploring these types of AI-implementations if you don't mind me joining in on this effort.
I have a little bit of experience in javascript  (although more extensive in Java ), though I do need to dust-off my knowledge and look into any recent changes in the language as well as HTML5.

[andretimpa]

Optimum response calculation based AI. This would "cheat" in the sense that it would peak at the whole deck of the opponent and calculate what possible moves the opponent can make in subsequent rounds, building a decision-tree like in a chess-AI to determine what's the best move.

a less "cheaty" implementation would be trying to infer the deck your opponent is using, or at least the general strategy (example: mark of darkness and a bunch of time pillars, assume ghostmare, unless ghosts take too long to appear). This would lead to a fuzzy decision tree (not sure if that's the technical term ).

[Keeps]

AI should be using a field algorithms, it'll play smarter that way.

[rob77dp]

Optimum response calculation based AI. This would "cheat" in the sense that it would peak at the whole deck of the opponent and calculate what possible moves the opponent can make in subsequent rounds, building a decision-tree like in a chess-AI to determine what's the best move.

a less "cheaty" implementation would be trying to infer the deck your opponent is using, or at least the general strategy (example: mark of darkness and a bunch of time pillars, assume ghostmare, unless ghosts take too long to appear). This would lead to a fuzzy decision tree (not sure if that's the technical term ).

Your description reminds me of my days designing custom AI scripts for AOE II.  Essentially, I imagine AI's for Elements that looks at factors and uses various methods of inferring deck opposed from those factors/clues.

- Mark
- Deck size
- First turn plays (then expanding to future turn card plays)
- RECENTLY experienced opposition matching these clues
- Expanding from there into a library of past decks "not seen lately" by AI from opponents
- [own deck stuff as Zso describes about a list of "how I prefer AI to use my deck" stuff]
etc.

Edit:  The AOE II corollary is that you can 'teach' (write) your AI scripts to learn/know the map style, opponent quantity, opponent civilizations chosen, etc and then invoke certain sections of AI method from that data.

[eaglgenes101]

My idea of the AI was this:
1. Count quanta available
2. Check for one-turn game end
3. Run synergy function on every card pair possible
4. Note down all synergy chains
5. Use computationally efficient function to determine the play that does best
6. Sort this best play by which cards are best to play first
7. Play it, if a random event or new card happens, rerun search

The evaluation question I suggest is turns to lose - turns to win.

I'm eager to help with the AI.

[antiaverage]

For those looking to contribute, the beauty of github is that any of you can fork the project and then submit a pull request.

https://help.github.com/articles/fork-a-repo

https://help.github.com/articles/using-pull-requests

[Keeps]

For those interested in AI, as an old AI developer, I would suggest looking into
Potential Fields: (for general board and quanta control)
http://aigamedev.com/open/tutorials/potential-fields/

When potentials are matching, or potentials show inevitable defeat, then have it switch  all fuzzy logic algorithms to develop possible combinations, and make random 'guesses' as to what could shift the potentials.  Store the results, using some XML or MySQL in your project, and record the usage of and include these results in the potential field results, creating a Class 1 self learning brain.

Remember, null, is the third state, and means unknown.  This is a powerful tool to acknowledge.