i'm talking about this:
http://theaigames.com/competitions/warlight-ai-challenge

season 2 hasn't started yet but it's rumored to be around the corner.
i'm looking forward to the challenge, don't know how much free time i'll have to dedicate to it, but regardless of that i'm looking to team up with someone on the development cycle (messing with actual code, not just throwing ideas around).

previous season i developed under c#, for season 2 i'm thinking of changing to javascript (which i'm more used to in terms of data object handling) and perhaps use the opportunity to get better acquainted with the lodash library (it's a library that somehow facilitates functional programming and data manipulation for javascript, evolution of the underscore library afaik) which i been trying to get more into but never seem to find a proper application for.

so if anyone is interested please let me know so we can start talking about it.

Edited 10/2/2014 16:45:48

I'd love to try again, the last time I tried to put one together I got sidetracked by school and couldn't finish it. I always thought it would be a little extra challenge to put together a competitive bot that would work for any map...wonder if anyone tried that?

Edited 10/3/2014 03:45:55

some of the guys said their bots were like that but i'm sure they'll find plenty of bugs when testing on new maps.

i was trying for that at some point but ended up adding specific exceptions. that's also one of the reasons i want to start fresh :)

I'd love to see a few bots fight it out in strat 1v1! I haven't really bothered with AI since I lost a leg in '97

Hello

messing with actual code, not just throwing ideas around

I read the input other WarLight players gave you for the last AI challenge and this was indeed "just throwing ideas around". However writing the actual code is usually the easiest part and this is especially true for this AI challenge.

Here some concepts for how to move your armies around:

"Move armies to closest border" --> Initial thought

"If opponent border is at least two territories further away than closest neutral border, move to opponent border instead" --> SupremeDalek

"But maybe we should move our armies to a border where we actually want to expand... maybe our armies arrive to late for helping with expansion since we take that region before they arrive... maybe we already have an extreme overkill amount of armies in one place against the opponent while he is crushing us in another place... maybe we want to not move our armies straight forward to the border but add some delay moves..." --> Here stuff gets difficult.

ps: some of the guys said their bots were like that but i'm sure they'll find plenty of bugs when testing on new maps.

True. One of reason why I want to start new season is to see how my bot will play on random maps. And im sure there will be many of bugs..

Norman: However writing the actual code is usually the easiest part and this is especially true for this AI challenge.

Not agree. Depending on quality of code. Spaghetti/script code - yes - easy to write. But to write clear, supportable code (which is important since this is long term contest), could be not so easy task. Because often some new, revolutional ideas is hard to implement with current architecture. And you need to write some hacks, breaking all architecture, to make it work. Or rewrite all to make architecture of bot more flexible.

For example Im thinking now of some pattern detection in moves of opponent during game. And using this learned patterns in next turns. But with pain understand how hard will be implement this.

Hello

Not agree. Depending on quality of code. Spaghetti/script code - yes - easy to write. But to write clear, supportable code (which is important since this is long term contest), could be not so easy task. Because often some new, revolutional ideas is hard to implement with current architecture. And you need to write some hacks, breaking all architecture, to make it work. Or rewrite all to make architecture of bot more flexible.

I'm also wondering about what components a good WarLight bot needs to have and what interactions between them are necessary / beneficial. However when talking about this stuff I use plain english text or some fancy diagrams.

I have started throwing ideas around on paper (err...okay computer memory...) for a versatile ai that could be adopted to function in any situation (for example, different maps, different luck modifiers, multi-attack, cards, etc.). If anyone wants to take a look or throw in ideas of their own let me know.

Hello Traintown

Why not explain your ideas here or upload your papers somewhere.

The second season is shaping up very nicely, with random maps and a bunch of other new features.
Can't wait!

Tired of watching on silly mistakes of my bot and started implementation of strategic moves analyzer version 3.0 :P It will be fair simulation of all (well, almost all) possible moves of army stacks in 2-4 turns ahead. Hardest porblem for now - to make it fast, because need to simulate a lot of games in real time..

Edited 10/20/2014 12:41:25

Hello

I see that your new bot version is really trying some stuff:
http://theaigames.com/competitions/warlight-ai-challenge/games/5446c3024b5ab24ab3d7a317

Is quite interesting to watch. Problem with this smart stuff is though that the map doesen't reward many aspects of good play. GreenTea wins quite often against SupremeDalek and when I watch what happened it's always one of two situations:
- SupremeDalek thinks it's a stalemate but doesen't count repeated attacks so he deploys in unimportant Africa / in the background of South America / Australia
- SupremeDalek wins in Australia and GreenTea in South America. The extra armies from South America help GreenTea to push SupremeDalek out of Africa.

Problem I have with this stuff is that normally WarLight gameplay isn't about who builds huge stacks in the smartest way. I see this every time when I improved my bot. Improving the general way my bot fights / leftover management /... gives win rates not much above 50% against the old version but improvements that effect the initial stalemate situation gave win rates above 70%.

Agree Norman. With all my last improvements new version is almost always in 50% (+-3%) win rate with very old version (for 500 test games). This is because same start picks and we got initial stalemate situation which very depends on luck and not affected by recent improvements.

Unit test a bit helps, but not much, because far not all situations are covered by tests, and often I see bugs in such places where I never expect them ;)

Today I deployed new version, lets see how it will go.

http://theaigames.com/competitions/warlight-ai-challenge/games/544fd9124b5ab24ab3d8227d
I like how it played!
After assumption of possible enemy bonus it selects direction of attack from Australia. And this completely breaks hardcoded tactics of Blender :)
The way how it's played is unexpected even for me.. Just result of new strategic moves selection algorithm.

Edited 10/28/2014 18:35:36

Had some spare time in the last couple days so I decided to claim first place for Supreme Dalek real quick ;) Didn't work on the real flaws of my bot though, but if bots like Herz decide to sit me out in North Africa then I can play this just as dirty.

Not much to proud. Hardcoded implabanced strategy of start in Europe. And even not original after Blender and AdsRiskbot. Really borring to watch such games.

As I said: Just wanted to take first place back real quick with the minimum effort required. And not every bot playing certain aspect of the game better than GreenTea have this thing hard coded. Sometimes other players have better ideas and the own super smart idea turned out to have some serious flaws.

http://theaigames.com/competitions/warlight-ai-challenge/games/544cb85f4b5ab24ab3d7f6cb
--> I don't know of any other top ranked bot besides GreenTea who wouldn't have won this game after turn 9.

Yep. I noticed this bug with not expanding in Africa. Alreay in my list to fix :P
Last days changed so many code that such bugs could appear as side effect.

But anyway current version wins near 65% of games from version which played in finals.

UPDATE: Actually it was very rarely reproduced bug with exapand AI in code which I didn't touched maby half of year :P

Edited 11/6/2014 23:45:46

Can't prove that, but I feel that strategy with Europe is more stable vs usual strategies of Aus+SA in respect of luck, compared to Aus+SA vs Aus+SA.

As general rule: luck factor is more valuable in games of same strategies.

Thats explain why even with perfect Aus+SA strategy will be amost impossible to be 1st on ladder vs perfect Europe strategy, when sandbox contains 90% of Aus+Sa strategies and 10% Europe strategies.

And vice versa, if sandbox will contain 90% Europe strategies and 10% Aus+Sa strategies, then perfect Aus+Sa strategy will be most likely 1st, for the same reason.

Luck is a factor that helps me, yes. If all goes well I should be able to take Europe after turn 3 and hit North Africa turn 4. If a bot has a very good map understanding he might know what I do and attack me before taking South America + Australia. In this case I should have the game lost. If I get unlucky and only manage to take Europe after turn 4 then I usually still win. Problem then are the 'fast bots'. However, I have quite some luck buffer against most bots while getting a little bit unlucky in the initial South America + Australia stalemate situation can result in a game loss. If I'm up against another Europe bot then I have to expand in South America / Australia so it's not much different from a strategy around those two bonuses and luck affects the outcome in the same way.

The reason my bot is number 1 at present is because I'm exploiting the lack of map understanding in most other bots. If a bot doesen't expect me in Europe and I don't get unlucky then I win against him.

Edited 11/14/2014 23:04:30