<< Back to Warzone Classic Forum   Search

Posts 341 - 360 of 565   <<Prev   1  2  3  ...  10  ...  17  18  19  ...  23  ...  28  29  Next >>   
WarLight AI Challenge: 5/22/2014 00:04:45


Trogatog
Level 52
Report
Looks like first 6 places of ladder are almost equal in game power. So if nobody untill 25 of May will not invent some 'super weapon', then it will be very tough competition.

I'm just hoping I get enough matches for you guys to help me correct behavior that doesn't function as expected.

Of course, now that those bots that kept crashing the queue are gone, hopefully we'll see a lot more matches complete
WarLight AI Challenge: 5/22/2014 02:15:32


Trogatog
Level 52
Report
Think I found the "super weapon" GreenTea was talking about:



Too. Freaking. Often.

Also, funny game of the day:

http://theaigames.com/competitions/warlight-ai-challenge/games/537d67044b5ab2583f23cea4

JUST DIE ALREADY!!!

Edited 5/22/2014 02:57:36
WarLight AI Challenge: 5/22/2014 09:19:55

GreenTea 
Level 60
Report
"http://theaigames.com/competitions/warlight-ai-challenge/games/537d67044b5ab2583f23cea4"

hahaha xD
WarLight AI Challenge: 5/22/2014 20:23:01

GreenTea 
Level 60
Report
Somebody stop pedrito. He goes too far.

Edited 5/22/2014 20:23:11
WarLight AI Challenge: 5/23/2014 22:59:39


Trogatog
Level 52
Report
Somebody stop pedrito. He goes too far.

I know! What happened to being on vacation, sir? How will we ever knock you out of this competition???
WarLight AI Challenge: 5/25/2014 11:06:17

{rp} pedrito 
Level 48
Report
My vacation start tonight, after 23:59 :)
WarLight AI Challenge: 5/25/2014 16:09:33


Trogatog
Level 52
Report
Moment of truth is fast approaching! Good luck everyone!
WarLight AI Challenge: 5/25/2014 18:45:36

GreenTea 
Level 60
Report
Well, for my bot there are no any major updates to code in last days, just some fixes. And those fixed didn't improved win rate against previous version (probably because in fixes I had added new bugs). So good luck for everyone, and let the strongest win!

Edited 5/25/2014 22:25:36
WarLight AI Challenge: 5/25/2014 19:22:02


Trogatog
Level 52
Report
Once it's all locked down, shoot me a pm and I'll invite you to Trogabot's git repo for some good laughs at my lack of cleaning up my code base :)
WarLight AI Challenge: 5/25/2014 21:56:02

Ikalgo
Level 50
Report
"Blender" here guys, good luck all :). It surely has been a lot of fun! Just wanted to invite people to chat (http://webchat.freenode.net/?channels=theaigames) now that we are all free to share our methods (some people were obviously liberal about it, but I decided I needed a bit more secrecy given my specific "europe" strategy). Since I felt the tactic sometimes comes a bit short, I realized I should still invest in the whole SA/AU business (which when I was top 40 was nice to avoid, but in the end became very interesting). Roughly 47% of the games will be played as Europe!

I am really curious to hear more about specifics of the guys around here! Especially near the end, pedrito: what did you do to have so many wins at the end?! Trogabot also made big changes and GreenTea also became very solid overall.

Similar to GreenTea, I used the approach of simulations. I basically load the information for 2 million cases during the game. You give it your own income, the enemy's income, your troop count and enemy troop count, and it will give you the probability for a successful defense or attack.
I have this exact data for lower than 400 troops and up to 24 income. Beyond these values I use some pretty accurate approximation (off at most like 3 troops). As GreenTea tried to argue, your linear method (troop_count * 1.167 or something) actually scales quite terribly! I actually posted on Stackexchange: http://stats.stackexchange.com/questions/91814/probability-enemy-territory-captured-in-x-turns might be interesting to someone.

My goal was to basically both attack and defend with at least 70% successful chance, given constant attack. I use a probability transition matrix for deciding where to defend for example.
The place where a troop is most valuable is where it would be placed then. Though, without Norman's assigning values to territories I still had a lot of strategy to write to guide this.

Furthermore, I was forced to write as good of a "One enemy region able to attack 2 of my regions in an owned super region" as possible for playing Europe. Against some people, it really makes a difference.


I really love how SA/AU games have evolved, especially GreenTea, Trogabot and pedrito have delivered some interesting games and really have made me step up my game!
The latest additions was dealing with you guys playing AUS game and attacking from Asia to USA. GreenTea: I hope you're not going to attack me from Europe when you discover a stack on the Asian border; approaching from Europe is wide and open (same for pedrito)! Furthermore, I will now "hide" the bonus from you guys so I hope you won't rush me ;)

I have always been at awe with every game of Norman/Dalek; you can't see him make stupid moves (ok, just really few)! And no matter how much we have worked at it the last few days, he's still upthere. Though lately I have to admit it holds for others as well. My bot is still pretty stupid, and I had to hardcode a lot of strategy. I do think my europe tactic is almost there :) I actually felt like the more changes I made recently, the worse it became :/

It will be very interesting to see, but I'm so afraid for silly bugs that might appear ... Again, good luck everyone, I know I'll be watching many games on the edge of my seat!

Edited 5/25/2014 22:56:26
WarLight AI Challenge: 5/26/2014 14:44:01


ps 
Level 61
Report
cool, i'll try to schedule a livestream event to chat about this and take a look at some of the games, have to figure out dates not to clash with other livestream events and bother people to join me live, if anyone is willing to join in commenting please send me a message. i think it would be fun and nice buildup for the finals :)
WarLight AI Challenge: 5/26/2014 18:02:10

GreenTea 
Level 60
Report
Thanks, Blender, for big post from you :) Now I think it is safe to share some info and ideas about our bots.

So, breafly about my bot, it consists of such parts:
Core:
* Breadth first search utils.
* Attack analyzer - for precise calculation of chance to capture, average losts etc using formula of Bernulli.
* Game simulator - for making 'virtual' turns and working with modified state of map. Very usefull to modelling game on few turns ahead. Then all changes could be rolled back.
* Strategic score analyzer - analyzing of strategic value of every region on map. For example region near enemy bonus is more important then region far from enemy bonus.
Expand:
* Expand AI - finding best bonuses to expand, and best way to expand. Also possible to calculate best way to exapnd on many bonuses simultaneously. Used to found best picks, except
that first 2 picks are hardcoted to Australia :\
Battle:
* Battle AI - most complex part.
1) Prediction of enemy income and bonuses.
2) Analyzing every pairs of near reagions, where first is my and second is enemy region. Modelling all possible interaction between those regions using my income and predicted income of enemy.
3) Generation of all possible income usage and finding combination which gives max scores after analyzing of region pairs. Also on this step for every region I found best turn for it, which could be:
- regular attack (with all forces)
- split attack (in all direction, when big advantage of armies)
- partial attack (attack with part of forces)
- defence
- strategic move on neutral region (to protect my bonus or move to enemy bonus)
- stragetic move on my region (usually to protect my bonus)
4) Moving of rests from far (not border with enemy) regions.
5) Moving to enemy strategy.
6) Expand using Expand AI.
7) Identifying best order of moves.

And about 120 unit tests for separate modules and on various game situations. Something like this :)

Edited 5/26/2014 19:16:14
WarLight AI Challenge: 5/26/2014 20:32:23

GreenTea 
Level 60
Report
http://theaigames.com/competitions/warlight-ai-challenge/games/5383a2a54b5ab2088e1b9e53
If such game will be somewhere in finals, I'll be very upset..
WarLight AI Challenge: 5/26/2014 23:00:31

GreenTea 
Level 60
Report
WarLight AI Challenge: 5/27/2014 00:14:47


Garrett
Level 56
Report
I haven't participated in this thread at all while the competition was going on, but now that everything's locked in I'd like to join in on the fun.

First off, I hope everyone had a good time programming their bots (even through all those long hours debugging), and that we all have some good games! It's easy to watch your bot lose and blame it on yourself for not changing the logic up or getting rid of that one bug, but we should really be congratulating our opponents for doing an excellent job on their own bot. So good job everyone!

Now to share my AI logic. I still have a lot to learn regarding computer science, I learned it as a hobby a few years ago, so my code is terribly organized. That said, I think the main ideas behind it are valid, so I'll be glad to hear everyone's criticism on it.

-The picking strategy is quite simple, chosen from smallest bonus up, further organized by bonus safety (ways in) and regions are chosen from distance to other bonuses.

-Each round starts with predicting the enemies income, which works only if they deployed more than the previous known income a previous turn. It then assigns the smallest bonus that would fit the income shown to the enemy, if it's possible for them to have it.

-Enemy probabilities are then calculated. This includes lost picks if it's round 1, otherwise just some basic expansion predictions which could be improved quite a bit.

-Bonus goals are then calculated. This is the defining feature of my bot, it works based on 4 main bonus goals, which include Survival (if one's about to be eliminated), Capture (if it's a good candidate for capture), Break (if an opponent has it and it needs to be broken), and Elimination (if it's possible for an enemy to be eliminated completely from a bonus). Each of these are easier to calculate then you'd think. At the moment the calculations look like this:
bonus.Goals.Capture = bonus.ArmiesReward * (1 - enemyNearby) /
    (territoriesToCapture - nearbyArmies / 30.0);
SurviveRating = Minimum Ratio Of((mostSurroundingArmies + state.PredictedEnemyIncome) * 0.6) /
    (playerRegion.Armies * 0.7);)
bonus.Goals.EnemyElimination = (playerArmies + state.StartingArmies) * 0.6 /
    totalEnemyArmies;
BreakRating = 1 if enemyIsVisible, else it's 1*enemyProb of all territories.

All of these are of course conditional on if my opponent and I are in the bonus being analyzed.

-Next comes the DeployArmies code, this is easily the most massive section of the program, at over 700 lines. That's not bragging, I'm rather disappointed by how much typing I had to do to accomplish what I wanted, but it is what it is.

-To skim over the details, the bot looks over the bonus goals and makes objectives based off of them of the different types (survival, break, etc.), doing its best to sort from most important to least important.

-It then runs a loop over the objectives, deploying the minimum armies necessary to accomplish each objective, with different logic for each type.

-If at the end there are some leftover armies, it deploys them evenly to all deployment orders made (with some backup logic in case no objectives were made).

-Then the AttackTransfer orders are made, this section looks at all the remaining useful territories and decides what to do with their remaining useful armies (transfer, add to previous attacks, etc.). When the attacks are made the Priority is included (important, delay, safe, unsafe), but this could've used some more work as no delay orders are generated if they've gotten to the most useful territory. Some transfers are odd as well.

And that's that. Sorry about the wall of text if you weren't interested in this. And I wish everyone the best of luck for the remainder of the challenge! I don't expect my bot to do too great, but I might get around that 1 euro mark. :) And I'll leave with one recent game I liked against ps, whose move into North America seemed costly but eventually won out, and had some nice anti-AI maneuvers (on purpose?) in South America. Also, I enjoyed watching my AI run away in Africa after I had lost. gg ps!
http://theaigames.com/competitions/warlight-ai-challenge/games/5383a22e4b5ab2088e1b9e48
WarLight AI Challenge: 5/27/2014 11:57:59

GreenTea 
Level 60
Report
WarLight AI Challenge: 5/27/2014 12:48:21

Ikalgo
Level 50
Report
@GreenTea: Thanks! I'm now regretting not playing a higher percentage with Europe ;) Though yea... if you would be prepared for it I think you would handle it appropriately, perhaps even you will in some other games :)

This is nasty though: http://theaigames.com/competitions/warlight-ai-challenge/games/5383e7cb4b5ab2088e1ba427 (I also have something similar happen, sadly)

Edited 5/27/2014 12:52:04
WarLight AI Challenge: 5/27/2014 13:02:27

GreenTea 
Level 60
Report

This is nasty though: http://theaigames.com/competitions/warlight-ai-challenge/games/5383e7cb4b5ab2088e1ba427 (I also have something similar happen, sadly)


Yes, strategic scores for move on my region becomes highter then regular attack.. But too late for fix)

Edited 5/27/2014 14:44:39
WarLight AI Challenge: 5/27/2014 13:11:46

Ikalgo
Level 50
Report
Trogatog: I'm so embarrassed with this game http://theaigames.com/competitions/warlight-ai-challenge/games/53848ea54b5ab2088e1bb203

Edited 5/27/2014 13:12:04
WarLight AI Challenge: 5/27/2014 17:54:05

Ikalgo
Level 50
Report
Hey GreenTea, based on this game, http://theaigames.com/competitions/warlight-ai-challenge/games/5384cb8e4b5ab2088e1bb744, what is your criteria to attack? You attacked 42 units with 43 units? As you probably know, considering our bonuses, your chance for successfully breaking through is only 10%. A little later you also attacked 22 units with 23 units. So I'm curious what rule made your bot attack?

Edited 5/27/2014 17:58:28
Posts 341 - 360 of 565   <<Prev   1  2  3  ...  10  ...  17  18  19  ...  23  ...  28  29  Next >>