Norman, your bot is playing the same strategy as 4 other bots and each of them has been at #1 before. There's a lot of variation in (these settings of) Poker. Proper bots can only get roughly an estimated win ratio of 60-65% against an AlwaysRaise strategy in the current settings. Couple of hours later and you're on the #5 spot. Or are you just trying to encourage these guys to participate :D?

As for Blender: It's funny how Ad is now suddenly also playing Europe. I realized I removed the line which activates my SA strategy when I obtained < 2 regions in Europe. My SA strategy also includes a pretty good anti Europe measure ;). I might just add a version for that. I think if we would combine our methods (my method of defending and going for Australia, his expansion and killing off adequately), we'd get pretty close to an actual optimal Europe bot I'd say!

Have you guys ran some simulations by picking regions in Europe; I'd be really curious how your general bots would do it?

Edited 7/4/2014 05:14:59

"Have you guys ran some simulations by picking regions in Europe; I'd be really curious how your general bots would do it? "
I know for sure that it will not work for my bot, because if it is more then 1 turn to capture bonus (Europe in this case) then bot will prefere put armies to defend from enemy. And for now I can't see way to conclude in general way that capturing Europe is better then defending spot in SA for example. It should be some forecast of game on 3-4 turns ahead.

Edited 7/4/2014 08:27:48

http://theaigames.com/competitions/warlight-ai-challenge/games/53b866404b5ab20b4223b53f

lol xD

Here's the code that was used for Trogabot in the finals (I think):

https://bitbucket.org/trogatog/trogabotpublic

I made a branch from this to work on for future challenges, and I'm quite happy with 95% of the code.

Current changes include removing the stalemate code to be more dynamic (not specifically referencing North Africa) and the opponent income calculation method.

Most of what I'm happy with I hope that other players can use to benefit from:

Calculating attacks
Calculating defense
Stacking regions where opponent is about to interrupt income
Completing super region bonuses depending on income

In the GetAttackTransferMoves routine, I suspect about 50% of the code is unreachable at this point due to army attack/defense calculations taking priority, but you never know! (i.e. statements that say "if (fromRegion.AvailableArmies >= 100) {//then do something}" )

Anyways, enjoy :) I look forward to some good feedback from the other players out there!

Edited 7/6/2014 23:50:23

Hello

OK, let's try in some usefull criticism and not just in some Haha, my bot is more general than yours stuff.

I don't think you will get usefull criticism from people looking at your code because since it's lacking modularization it's to hard to read for non involved guys. Separate files like OpponenInfo.cs that have a clearly defined task are way easier to read than Trogabot.cs

In the past you topped the leaderboard for some time but at present you are struggling. Let's have a look why you are losing your games:

http://theaigames.com/competitions/warlight-ai-challenge/games/53ba57904b5ab20b4223dc8c
Africa can only come into play when the bot who is holding it blocks Brazil. Best play would be to stack in South America until you receive a stalemate and then stack enough in Australia until you have enough armies together to get a safe Australia break. As you faced elimination you let your bot get slaughtered. You bot should have tried to turn that lost game into a stalemate by running away. Some bots are very vulnerable to a run away strategy. Around round 30 the game is of course already lost but your bot is doing the mistake of not keeping his stack together, probably because he only attacks with the needed amount of armies. Just look at the terrible fight you run into round 31. I recommend to attack full force by default and when you are defending a valuable spot to add some carefully designed exceptions.

http://theaigames.com/competitions/warlight-ai-challenge/games/53b9c2124b5ab20b4223d09f
Turn one your bot uses his army inefficiently in South America. Look at
https://github.com/Norman1/SupremeDalek/blob/master/src/basicAlgorithms/IdealExpansionCalculator.java how I calculate the optimal moves to take a SuperRegion in one turn. In this game your bot should also have never stacked against me in Africa. There are some bots capable of playing South America, either because of hardcoded stuff (Herz, Gadzbot) or because they have a capable AI (AdsRiskbot). Neither Trogabot nor SupremeDalek are one of those bots. Pick 1+2 in Australia, 3+4 in South America and 5+6 in Africa. South America has following advantages over Australia:
- You can use your remaining armies form taking South America to fight in Africa. Those few remaining armies might allow you to win in Africa, especially if the opponent used to much armies to take Australia
- If the initial momentum from your remaining armies isn't enough to win in Africa you are probably closer to North Africa than your opponent and you can block North Africa. However if the opponent blocks North Africa and eliminated you from Africa then you lost.

Edited 7/7/2014 10:30:23

I don't think you will get usefull criticism from people looking at your code because since it's lacking modularization it's to hard to read for non involved guys. Separate files like OpponenInfo.cs that have a clearly defined task are way easier to read than Trogabot.cs

I completely agree Trogabot.cs is way too difficult to read. I had intended on using a similar structure to what you use in Dalek in that I was going to separate all the place army moves and attack transfer moves routines into their own strategy pattern (if you look at the place army routines, they all have the same signature) with a decider class that would determine if it's better to expand, press, or hold.

But, alas, I ended up spending more time trying to figure out a better Warlight strategy than keeping my code clean. Next version will look SIGNIFICANTLY better. The Omaha bot I'm working on, I took a lot of precaution to not get into the jumbled mess that my bot class ended up in the Warlight challenge.

So you guys think you can take on a bot?

Give it a shot!

http://elsocio.es/BP/index.php

There is no way to see any moves or deploys happening for now - only the resulting map is displayed each turn. But for the sake of fairness that information is not given to the bot either, he doesn't know any more than you do.

You'll have to excuse the raw interface too. I'll get around to improve it in time.

Ah, and luck is at 0% to avoid excessive frustration because you can't really see moves.

Edited 7/11/2014 15:21:01

rigged game, got 1,5,6 :X

new attempt with "normal" dist:



Edited 7/11/2014 16:05:32

Congratulations both of you.

Nothing rigged about the distribution though, if both the bot and the player tie for a spot, its a 50/50 chance. And if you tie for the next spot as well, its 50/50 again. So technically if you take the exact same picks as the bot, there is a tiny chance that you get 4,5,6. And the same probability of getting 1,2,3.

Therefore it's fair. But if you keep restarting the game until you get the picks you want, and then play that and win, the game is rigged in your favour.

Well I was winning my first game but I couldn't finish as your program bugged on me and locked my cursor to one spot moving the map when I tried to move around. Clicking brought up the 'territories not connected' message. Which I could then move cursor around and click away the message, but then it returned to the former locked position back to the same spot which I couldn't get it to release from. Stuck dragging the map around, can't finish game.

Edit: Nevermind, looks like it saved my game and a refresh fixed the issue. Territories in North America aren't connected properly though.

Edited 7/12/2014 06:08:09

Are you able to see a history of games people play?

Yeah it's all saved in a database.
Whenever you run into a client-side bug like the sticky map refreshing the page should fix it.

I'll have a look into those NA territories, you don't remember which one it was?
edit: As far as I can see they all connect properly.

Edited 7/12/2014 12:11:19

The error is in your picture, there's a red line between Quebec (6) and NW Territories (2). They aren't actually connected though and shouldn't be. On their map Greenland extends down into parts of Canada and connects to Ontario (5) which is not shown on your map. So two errors in your drawing. But how it runs is probably in proper accord with how the map should be. Just need to fix the picture drawing and connections.

Thanks, I saw now what you mean. Different maps indeed, but that should be fixed now.

Also, luck is up to 75%, and both the player and the bot can see enemy moves from last turn now. So that removes both handicaps that the bot was fighting with.
Handicap number 1: the bot expects 100% luck, and in certain situation with 0% luck is better to expand because you know for sure you will have a bonus next turn, while with 100% luck its a gamble.
Handicap number 2: without access to enemy moves most of the strategic brain in my bot doesn't even get called. No predictions, no bonus guessing, no info about deployment and therefore it doesn't even try to break bonuses.

So far I've only posted the link in this forum and over at the The AI Games. Which means only experienced players have had access so far, and with both of the aforementioned handicaps, yet the bot has managed to get some wins. Time to send some people with no prior Risk experience and see how long it takes them to figure the game out.

http://elsocio.es/BP/index.php

Edited 7/12/2014 20:56:05

I tried it and lost honorlessly to your bot pedrito :(

But something else: I'm working on the next challenge and particularly on generating random maps. What existing Warlight maps do you guys think are good maps to have an algorithm recreate in terms of structure and amount of regions/bonuses?
The way I'm going to do it is randomly create maps, but using a (couple of) predefined structures based on good Warlight maps, so you can't end up with worthless maps, but still always get different ones.

All suggestions are welcome

small/medium earth are the ones I'm going to start with by the way

Nothing rigged about the distribution though, if both the bot and the player tie for a spot, its a 50/50 chance. And if you tie for the next spot as well, its 50/50 again. So technically if you take the exact same picks as the bot, there is a tiny chance that you get 4,5,6. And the same probability of getting 1,2,3.

That's not how distribution in Warlight works. Not sure how it worked in this AI-Challenge.

@ jim: These maps are quite nice too I think:

Europe:
http://warlight.net/Map?ID=73

Roman Empire A (has 1 province bonuses though):
http://warlight.net/Map?ID=1173

@ It's not how it works in Warlight maybe, but it's not rigged. That was my point.

if you expect distribution to be like warlight and you gut 1,5,6 first thought is "rigged" :P