As I was coding an update this morning, I wrote this little gem while in code-mode:

RoundNumber = roundNumber; //(for mocking opponent)

Gave me a little chuckle :)

I am now mocking you. Every. Single. Game.

Cheers!

Edited 2014-05-11 14:57:37

Thank you all so far for your suggestions. Definitely some food for thought. I see now why FFA isn´t maybe the best idea for the next competition. But we´re going to think hard and try to come up with the best combination of ideas.

More suggestions are always welcome :)

fixed some bugs and ended up with worse performance on my bot, need to do some serious update on initial rounds battles during this week or i'm out of the top race, hope i find the time. :S

Do you guys think the suggested changes; different map, wastelands, 16% luck factor and different starting regions distribution is enough to have people really adapt their bots?

We're looking for people to really have fun again in (re)coding their bot and not just re-upload the one they have already written and perform really well with it.

Wouldn't a good bot perform well on all settings? :P

jim: like i said before, just a different map would not be enough (people would still hardcode their strategy), having random maps might be much better imho.

ps what strategy would you hardcode for a strat 1v1?

@ Jim: Speaking just for myself here.... Unless the next competition has random maps I hardly need to recode my bot at all. If it's just a different map (but always that same map) I can adapt my current version fairly quickly, and that wouldn't be much of a challenge. Different luck requires no adaptation at all, wastelands require a single check to disable expansion in that bonus and factor them into pathing.

I'd look forward to recode my bot from bottom up if there are at least 3 different maps, randomly selected each game.


@ Timinator: Coding a bot that performs well on any map and any settings requires many many times more work than a bot that performs well on a specific map under specific circumstances. So the answer to your question is yes, but the price is high for creating such a bot.

Would LD need more new coding pedrito?

I assume you mean local deployments? It depends on the map size. Yes it would require a substantial change in the way the AI thinks about expansion and exploration in order to use it efficiently on a bigger map. On the Small Earth map however it wouldn't be much of a factor because by the time any bonus aside Oceania and South America is owned by a bot the game is usually over, or will be before those reinforcements could make any impact.

Edit: Oh yeah, local deployments would also totally break my bots ability to guess how big my enemy is and where he's expanding... ouch!

On the other hand local deployments would make for much longer games on a bigger map, and therefore more server load.

Edited 2014-05-12 15:56:19

dead piggy: i'm talking about fixed map knowledge, like how for example we know SEA is a nice location to counter russia and scandinavia combos. or how greenland is more safe then australia. this info is mostly the same regardless of where the wastelands are at.

i would prefer to have coders focused on coding an algorythm that analyzes the map layout and comes to all those conclusions / decisions by itself. not hardcoded as a "if russian combos are good, deploy safe and place counter in SEA" rule.

so that algorythm could be carried to another map (battle islands v for example) and work just as good.

so my point is to have a batch of, i dont know, 10 maps, something that would be annoying enough for a coder to avoid detecting and hard code all decisions, each of the maps with slightly different settings and layout and have the coders abstract all that layer of strategic decision.

Couldn't agree more with what ps said.

The real challenge would be to create a bot that can read the map, identify weak and strong spots, plan ahead, and execute a real strategy.

Currently none of the bots even works at a strategic level (as far as I can see that is). They are quite efficient tacticians, but what little strategy is used comes mainly from the hard coded opening moves. SupremeDalek's strategy is to deploy strong in Oceania, Trogatrog's strategy is to deploy equally on both small continents and then counter enemy deploy. After the initial phase is broken, the bots play on a purely tactic level.

This has to do with the simplicity of the Small Earth map, and also with the fact that if the coder knows what map the game will be played on it's easier and more efficient to hard code strategy than to rely on more volatile interpretation methods.

Hello

How dare you announce my strategy pedrito. However I don't see why you call Trogabot's strategy hardcoded:
--> Give each bonus a value according to how many territories still to take --> Initially South America and Australia get a higher value than Africa. Perhaps mark Africa as unimportant since to much regions there.
--> First try to deploy enough armies to defend. This gives the South America spot 3 armies and the remaining two armies go to Australia since defence is also possible there.

I admit I hardcoded that my bot prefers Australia (by giving Australia extra points in the evaluation stage). Also you can call my first move hardcoded but that's just to prevent my bot from attacking first move. Apart from that it's not fair to call my bot hardcoded. I wrote quite complex algorithms telling me which bonuses the opponent has and there I didn't add any "South America" exception. If you see my bot stacking in Africa till he has 20 armies together and then marching towards South America then this is completely dynamic behavior that will work on every map.

OK, you told my secret now I tell your secret:
The main reason both SupremeDalek and pedrito play that well is because they both rip off bots that can't play South America. Also neither of our two bots is able to play South America themselves. Trogabot was the first bot to reach the 3000 mark but then he changed his strategy and he immediately dropped down in rank. That's not because the other bots improved but because him rising that high was also due to him ripping off bots not capable of playing South America.

Pedrito's last game against a SouthAmerica bot:
http://theaigames.com/competitions/warlight-ai-challenge/games/537156bd4b5ab26e84c7ec58

SupremeDalek's last game against a South America bot:
http://theaigames.com/competitions/warlight-ai-challenge/games/537137ac4b5ab26e84c7ea47
(OK, wen't a little different but if not for the Africa victory SupremeDalek would have marched towards South America)

Pedrito won this game but still this game acts as proof that pedrito can't play South America:
http://theaigames.com/competitions/warlight-ai-challenge/games/5370f46c4b5ab26e84c7e5ce

The real challenge would be to create a bot that can read the map, identify weak and strong spots, plan ahead, and execute a real strategy.

That would be very cool, do you think anyone can do it? I assumed that was way beyond your capacity as most humans cant do that, and the bots in the current competition aren't even close. It sounds like you should use the real time ladder templates.

If you wanted to push the bots to the limit you could force the same bots to play some FFA and team games as well. Im sure a lot of these games will be total clusterfucks though.

So here's my confession while we're all giving away secrets: my bot is a SA bot. I started taking picks in Australia simply to get that ever important, "Where is my opponent?" information. The Australia portion just kind of grew out of necessity.

When I hit 3000, I realized that my bot's biggest weakness was becoming the fact it sucked taking on other SA bots and was losing more and more to them, that's why I changed strategies and now take a completely more dynamic route to early game tactics. I've spent much less time coding combat and more predicting what your bot is going to do. You'll notice that Trogabot gets significantly worse when we are spread out across more than 2 continents in combat because the possibilities become more and more complex.

Every once in a while, I'll get both SA regions and just smile as my bot settles in like it's back home :)

I am seriously looking forward to the post-finals on this. I've gotta give serious props to all of you guys though:

Norman: Dalek has just been amazing to watch and learn from. I don't know how you resist not trying to code an update whenever a game is lost because you truly know strategy on this map.

Pedrito: you clearly are a better Warlight player than I am. Again, I've learned tons from watching our games and how you exploit weaknesses of other bots so naturally

PS: you give me the most frustrating matches of all simply because every time I think my bot has out-smarted yours, you still humble me and send me back to the drawing board

125ch209: I think Gadzbot is the only bot who has truly tapped into the overlooked opportunity of really analyzing the place starting armies and using it to its full potential. I swear to God, if I lose because you put troops in Siam again, I'm going to scream... Not bad for a rookie programmer :)

GreenTea: where the f*** did your bot come from??? Out of nowhere a new player shows us what dynamic bots (I can only assume) are capable of doing.

I don't think I've had as much fun coding and talking to other programmers/warlight junkies as I have these past couple months. Seriously, thanks everyone and good luck come May 25th!

Hello

I stronlgy dislike the idea of teamgames while ffa maybe could work for having some casual fun with bots. One reason is that I think teamgames are to complicated. I think most bot programmers are just casual programmers who took the starterbot and added some better heuristics to it. Here is how teamgames would have to be implemented if you take things seriously:

Teamgames:
Bot teamgames fall into the category of distributed artificial intelligence. It has to be well defined how the bots communicate with each other. Human teamgames have quite a complex communication for example:
- Send me 10 supporting armies to North Africa.
- Let's attack North Africa together.
- Move your stack before fifth order out of North Africa since I will attack you fifth order to take the bonus.
- Let me take Africa please.
- ...
Humans then might agree or not agree with the proposal. Of course you could just make one bot the puppet master telling all bots how to move but then it's not really a team game. Then it's just one bot playing who doesen't have the ability to deploy his full income everywhere.

ffa:
When I create an ffa I try to make the settings so that alliances aren't that important. The settings are:
- Huge open world
- Multiattack
- many neutrals
- light fog
The open map ensures that you can't just sit behind a chokepoint. The high amount of neutrals makes attacking the opponent often more worthwile than expanding. Multiattack makes it possible to form a huge stack and crush a far stronger opponent. You are never really safe. So sad people just don't understand. Even with super low prerequisites my ffa's take forever to fill while all those other crappy ffa's fill in no time. Just look at those awesome games. How can anyone who ever felt the touch of a real ffa play those stupid diplos any longer:
http://warlight.net/MultiPlayer?GameID=5394051
http://warlight.net/MultiPlayer?GameID=5388727
http://warlight.net/MultiPlayer?GameID=5389146
Ok, this was more about me hating all those ffa in the open games tab but what I also wanted to say is that if you implement ffa you have to be very careful about the settings. Use a map with a chokepoint in the middle and nothing will move, even without the bots forming alliances. Some ffa creators enforce an army cap so if you have to much armies on the board your income drops down. That way it's more worthwile to use your armies against an opponent.

Norman: Dalek has just been amazing to watch and learn from. I don't know how you resist not trying to code an update whenever a game is lost because you truly know strategy on this map.

Sorry, no more updates on SupremeDalek. I'm not in school anymore ;(

Since you talk about strategy: I believe on this small earth you can implement something like a "perfect play" and I think I know how the perfect play looks like. You go for South America and stack in North Africa. Even without the opponent showing up in Africa immediately you just sit there for a couple turns until you are able to safely take North America.

The reason most bots fail playing South America is because they don't defend North Africa. With you having 7 income and the opponent having 7 income from Australia there is no way for you to throw the opponent out of Africa if he just plays passively.

Since you talk about strategy: I believe on this small earth you can implement something like a "perfect play" and I think I know how the perfect play looks like. You go for South America and stack in North Africa. Even without the opponent showing up in Africa immediately you just sit there for a couple turns until you are able to safely take North America.

I had a version do that once -- on accident. It won a couple games, but I was like, "Yeah, this can't possibly be good..."