Ants

[AVAVT]

Your bot looks great perceval! It probably will get higher than mine.
I think you have some problem with food gathering though, after some time into the game your spawn rate decreased significantly

does that mean that only the top 100 will be enterd in to the tournament at mif december ?

Yeah I remember reading it somewhere that only the top 100 get into the final tournament... don't remember where though.

Is anyone on 64 bit linux and interested in a binary exchange? (i.e., we trade binaries for local testing...)

I don't mind exchanging binary but I don't think you would need testing with my bot to improve yours (I use 32 bit windows but Java is system independant I think)

btw I haven't heard of Go language before, how is it? I mean comparing to other more popular languages?

[perceval]

Looking at the bots my current goal is to enter the top 200 or 300 but i don t think this version will make it

I'm a little miffed that my current entry is better with its bugs than without. It got above 50 so I don't feel too bad that I can't beat it yet, though.

saraneth, good luck, added you to the first post...

Is anyone on 64 bit linux and interested in a binary exchange? (i.e., we trade binaries for local testing...)

i am on windows but i am willing to send you my jar file "for free" (well not sure what kind of training you'll get from it. in its current version it is not so good)

[lorill]

perceval> you need to improve your priorities :

http://aichallenge.org/visualizer.php?g ... &user=5586

turn 123, (17,13) could take a hill but moved north instead to get killed
turn 146, (93,24) could also take a hill, but went to food instead.

This will cost you the game with stronger bots.

[lorill]

Your bot looks great perceval! It probably will get higher than mine.
I think you have some problem with food gathering though, after some time into the game your spawn rate decreased significantly

does that mean that only the top 100 will be enterd in to the tournament at mif december ?

Yeah I remember reading it somewhere that only the top 100 get into the final tournament... don't remember where though.

This is wrong: http://forums.aichallenge.org/viewtopic ... 279#p11279

[perceval]

as developers we know that devil and bugs are in the details.
nice illustration on my ant project: i have been trying to understand why my bot sucked so much at combat.

cur.EDist(m.getPosition().getNeighbor()[d])<Ants.getAttackRadius2()

instead of

cur.EDist(m.getPosition().getNeighbor()[d])<=Ants.getAttackRadius2()

right in the heart of the combat code where i determine if an enemy is in range if i go somewhere

[perceval]

made it into the top 100 .. lets see if it lasts.
i miss ideas for improvments though

[Euphony]

Hey, thanks for the heads-up on this contest, daniel_the_smith . I hope I'm not too late to get a good bot going. I'm not a very experienced programmer, but I'm definitely interested in improving my skills, so this contest will be fun and useful.

I'm still working on food gathering, as mazes with just one "choke point" for getting through to the other side really screw me up. Heck, I'm not even really getting around on any map. What I was thinking about doing was setting waypoints (based on the A* link you mentioned daniel_the_smith), like setting one at the choke point on a map. Then A* there from the hill and cache that search and re-use it to get around the map.

Or maybe I should start looking into things besides A*. I finally got the A* working, and I'm basically using it for everything.

btw- I'm 'Euphony' on aichallenge.org

[perceval]

Hey, thanks for the heads-up on this contest, daniel_the_smith . I hope I'm not too late to get a good bot going. I'm not a very experienced programmer, but I'm definitely interested in improving my skills, so this contest will be fun and useful.

I'm still working on food gathering, as mazes with just one "choke point" for getting through to the other side really screw me up. Heck, I'm not even really getting around on any map. What I was thinking about doing was setting waypoints (based on the A* link you mentioned daniel_the_smith), like setting one at the choke point on a map. Then A* there from the hill and cache that search and re-use it to get around the map.

Or maybe I should start looking into things besides A*. I finally got the A* working, and I'm basically using it for everything.

btw- I'm 'Euphony' on aichallenge.org

what i do is only looking for close things: ie when i found a food, i compute path (with A*) only to ants within viweing range: the chance that some ant further away can go there is slim. that way my pathfinding qre usually short, and i cache all ancestor graphs which is not too much in memory.
to explore i keep a list of all "boundary" tiles ie tile that i have nerver seen but are next to an already seen one. when an ant is not hunting for food i just direct it to the nearest boundary. each time i see a tile for the first time i pdate this list.
i also keep ants in alread discoverd territory to camp for food, but the minimum, ie the camping ants viewing range should not overlap.
and for combat i more or less just run if i have more ennemies in my vieweing range than friends, and advance otherwise.
this worked surprisingly well ( i went as far as 26 ) but i dont dare touch anything now

[Euphony]

Thanks for the response, perceval. And congrats on (at the time of this post) being #23. That's pretty freakin' awesome. You have some very impressive games.

Well, I started caching my A* searches to nearest unknown tiles, and suddenly my ants stopped bouncing back and forth (from each A* search they were doing) and started following the cache to the unknown tile. Plus, I was able to increase search depth, so suddenly my ants were exploring pretty well. I was surprised how much improvement I got from following that cache. maze_04p_02.map still gives me a bit of trouble, as that single passage to the other side, combined with a solid wall preventing access around the edge, makes it difficult. My ants try to get through the wall rather than take the single passage through. That's why I wanted to set a waypoint at the single passage to direct my ants there so they would get through. However, on most aichallenge.org maps, I seem to be doing just fine as far as exploration goes. That maze_04p_02 may be a bit unusual.

So I was excited about my improvement in getting around the map and then I played on the TCP servers and realized that I'm going to get nowhere until I implement good combat code. LOL, without combat you get annihilated. And after getting crushed by a bunch of probably top 200 players, I played in a game with GreenTea, whose last bot was in the top 20, and saw him crush all the players that were crushing me. Pretty cool to see the difference in skill.

I'm working on combat right now. Still learning data structures and such, so it's taking me longer than I thought.

[Euphony]

Is anyone still working on their bot?

So, it looks like the new cell_maze maps are "closest to what will be used for the final tournament." according to a contest organizer:

http://forums.aichallenge.org/viewtopic.php?f=27&t=1973&p=12056&hilit=cell_maze#p12056

[daniel_the_smith]

I'm still working on my bot, although I haven't had much time lately. I was in the top 50 for a bit, but the competition has gotten stiffer. Also, irritatingly, my bot was much better before I fixed the bugs. The cell maps confused my bot a bit, too. Hopefully I'll be able to get my act together before the 18th...

[lorill]

I'm not working ont it anymore, I can't see what I could improve, even when reviewing lost games

There's a bit of luck in there, I got several ties with lower ranked bots at first, and going up after that takes a very long time. I still get good results again better bots, so meh. We'll see.

[perceval]

Still working on it but i was really stuck for a while : for each enhancement i couldn't win against my older V2 bot.
Finally i found a couple of nasty bugs that crops up once in a while and kills my perf.. sometimes.
also i was lucky to find a decent,very simple combat heuristic for my V2 and my attempts to improve while staying simple failed miserably.
i finally beginning to beat my V2 version consistenlty so i'll upload soon.
improvments:
a funny bug meant that on some maps my ants would not explore a portion of it.. but they would still go there if drawn by a fight or food, so it was barely noticeable on replays but it lead to big lag in developpemt => less ant => death during middle game.
its often the case that if you miss your start and are stuck with 10 % less ants around turn 200 you ll be slowly grinded down so fuseki it very important (and unfortunaly here comes luck: sometimes you are screwed by your neighbor at the start an can only watch the other bots harvest while you defend until you are too far behind )
-the biggest improvments are less frequent stupid ant death .
- hopefully better hill defense we'll see.
-implemented dead end detection to avoid overcrowding boring places, i used it also to avoid retreating into a corner and getting killed it does save 10 ants /game whixh might make a difference, i need to think a bit on how to use those dead end more generally.
later if i have time:
- my bot is still quick so could compute more stuff but i am not sure how to use that: maybe i ll try to detect chokepoint and block them.
- need to stop sending scouts to get slaughtered in ennemy territory, while no stopping exploration.
- recognize when i am blocked and find a way around, or force through i f i am strong enough. when i tried to put code to force through chokepoints, there were always cases where it would allow slaughter of dozens without progressing
- special code for mazes /1v1 but not sure what?

final goal is to stay in the top 100 it will be hard
The competition is much tougher these days, i am no sure that my improved version with climb back to where my V2 is, even thought it is slightly better.

[perceval]

So anybody still working on that ? i ll submit my very last bugfix soon.
in the end i got some noticeable improvments but the competion got a lot harder too. I hope to be in the top 50 in the end but it will be hard. Maybe top 100 is more reasonable; Last 2 weeks were really fun with a lot of bots becoming real good
I had a blast with this contest anyway !

[hyperpape]

I'm disappointed I didn't get a bot with any actual ideas done. I was using this approach, which seemed interesting: http://benjamin-meyer.blogspot.com/2011 ... ather.html, and then got sidetracked into other things. Maybe I'll try to make it actually work in the next day and a half, but I doubt it.