Hello

I was probably to fast with anouncing my next generation bot. I don't have enough time at present. What I wanted to do is to add labels to each region allowing me to distinguish far more situations and giving my bot a better map understanding. For example he could tell quite precise when a stalemate situation occured.

@GreenTea:
It depends on how you define that the bot only works on this map. In my opinion there are three different map types:
- Huge expanding maps (Big Europe map, Eastern Asia)
- Balanced expanding / fighting maps (Medium Earth, Europe map)
- Fighting maps (Small Earth)

The more the map is foccused on fighting the better my bot works. The heuristic never to expand into two bonuses at the same time isn't good for all maps. Also I have a heuristic about when to use armies to expand into a bonus depending on whether I'm losing or winning and how much regions are still to take. Although this isn't hard coded this heuristic doesen't make sense on all maps. When I lose connection to the opponent then at present I walk strong into his direction. This works very fine on this small map. On a bigger map like medium earth you can't just build a big stack and march towards the opponent.

Hi, I understand and generally agree.

"The heuristic never to expand into two bonuses at the same time isn't good for all maps."
but if you implement some rule, when to expand on one, and when for more bonuses - then bot will be more map indepented.

"When I lose connection to the opponent then at present I walk strong into his direction." The same , could be possible to define whether it is needed to go to enemy with big stack or not.

But yes, probably this map is not require such checks and some strategies optimal for it could be hardcoded.

I didn't test bot on another maps, but it would be interesting. Small earth is too.. small :)

Norman your bot would do better on strat 1v1 than you think. Never expand in 2 bonuses and when you lose connection with your opponent walk strong into his direction are pieces of advice many players need to start following in their strat 1v1 games.

I agree with Norman. I tried to code my bot as map-independent as possible, but some of the rules that it follows just scale very bad with map size.
For example, my bot only considers bonuses for taking if no smaller bonuses are available (either because they are already owned or because they are disputed). This works well on small earth, but on any other map it's a recipe for disaster because I might simply not have any "smallest non disputed bonus" within reach, so the bot would not expand in this situation.

I sent Fizzer en email to see if we could maybe access the map data from some of the popular WL maps. If he agrees, it shouldn't be hard to feed them into the conquest engine and run our bots on some other maps for fun.

WOOOO!!! Thanks pedrito!

i know mine would require some changes to be more generic, but would definetely be cool to test it on other maps and keep working on it as an open source project :)

So interesting to know what generic features are implemented in your bots, and compare with mine. But lets wait untill finals. Competition should be fair.

lol, win by running away until the opponent's bot breaks down :D
http://theaigames.com/competitions/warlight-ai-challenge/games/536a4af44b5ab23518a001fb

my bot ran out of expansion targets (since all superregions had already been taken atleast once) and that made it crash with an index out of range :D

fixed on v77 :)

Yeah, once the finals are on their way we can all have a laugh and pass around bots and beer bottles. But until then... no truce! :)

Regarding map data, Fizzer has pointed me here:

http://wiki.warlight.net/index.php/Query_game_API

With the help of this tool we can easily extract all the information needed to set up games on any map!

edit:
The Small Earth map seems to be hardcoded into the conquest engine in RunGame.java
makeInitMap() would need to be modified in a way it could parse a file in this format:
http://elsocio.es/maps/europe.txt
Ideally in a way that the map to be used could be specified from the command line the way you can specify the bots.
I have no experience with java, but I can probably get it done with enough time and persistence. As said before, it's no priority though until the finals start. And if any of you guys with plenty of java knowledge (Norman, GreenTea) is up for doing it I wouldn't complain either ;)



Also the game ps posted:
http://theaigames.com/competitions/warlight-ai-challenge/games/536a4af44b5ab23518a001fb

Funniest game so far :)

Edited 2014-05-07 16:30:43

"http://theaigames.com/competitions/warlight-ai-challenge/games/536a4af44b5ab23518a001fb"
It looks something like this
https://www.youtube.com/watch?v=VUOe_hLg7Bo
xDD

Those are the best games, what do you mean?

http://theaigames.com/competitions/warlight-ai-challenge/games/5364f85f4b5ab235189f8ea1

unless it happens to you...

http://theaigames.com/competitions/warlight-ai-challenge/games/5368f2814b5ab235189fe812

Also the game ps posted:

http://theaigames.com/competitions/warlight-ai-challenge/games/536a4af44b5ab23518a001fb
Watching this game while playing "Yakety-Sax" just totally made my day :)

Edited 2014-05-07 20:28:23

http://elsocio.es/maps/europe.txt
its is json format which can be easily parsed on java.

So it should be no problem to convert it to model of the game. I see just 1 problem with some maps of warlight, that they support many bonuses per 1 region. In current game model region realated only to 1 bonus.

c# Timed Out error is back :( can't test anything else this evening :(

Timeout happens only for C# bots?

yes, afaik

Yup, it had something to do with them running two queues.

C# bots take a split second longer to load up due to the hooks into the GAC, but when some other bots that are running at the same time are requiring a lot more processing power, it caused that load-up to take more than 2 seconds. If you could see the output for the C# bots when they crashed, the preferred starting armies gets a null value set, then the first command sent to the game host for place armies is the actual string that was supposed to be the preferred starting regions move.

Frustrating when you're trying to figure out what new bugs you've introduced into your code and even the test match crashes :(

-cringe-

Yeah it's painful to watch. And it doesn't help very much when trying to determine if a loss is because the other bot played better or because he got luckier.

At least the problem with C# bots should be solved now, the time limit for picking starts is up to 10 seconds.