EPTI guess that's why after the first attack most computers are pretty much useless.
Bot Fight: Examination of Starcraft AI - Page 2
| Forum Index > BW General |
mikeymoo
Canada7170 Posts
I guess that's why after the first attack most computers are pretty much useless. | ||
|
Luddite
United States2315 Posts
+ Show Spoiler + Assuming that your playing on hunters, with the dumbest possible strategies, in an 8 player free for all, with unlimited resources. | ||
|
fusionsdf
Canada15390 Posts
On April 15 2008 09:25 mikeymoo wrote: What you've posted is actually really interesting fusionsdf- I didn'trealize computers worked off of waves like that. The first wave always seemed the same, but I didn't realize the way it looped into two and three. I guess that's why after the first attack most computers are pretty much useless. thanks | ||
|
Vin{MBL}
5185 Posts
and it's like the comp KNOWS where you are and doesnt scout HAX | ||
|
Newbistic
China2912 Posts
On April 15 2008 13:13 Vin{MBL} wrote: I HATE that zerg 4pool it makes me want to cry and it's like the comp KNOWS where you are and doesnt scout HAX I think it's just throw in to discourage people from playing too many compstomps. But yeah, its a bitch if you want to train vs comps. For whatever reason, the worker stack + attack doesn't work as well vs the computer either, I can never trap their lings :/ | ||
|
B1nary
Canada1267 Posts
| ||
|
fusionsdf
Canada15390 Posts
| ||
|
Polyphasic
United States841 Posts
On April 15 2008 09:04 GTR-2-Go wrote: You guys should start playing vs the Ahzz AI, I have troubles with it sometimes with TvP (Spirit Toss anyone?) what is the ahzz AI? | ||
|
Physician
United States4146 Posts
On April 15 2008 13:32 Polyphasic wrote: what is the ahzz AI? http://www.broodwarai.com/pages/BWAIWarII.html jeje his experiment has been done years ago and came up with the same inevitable truth -the original toss AI script is simply the best out of the three. gone off to recruit another to our band of brothers, thanks for the find op : ) (brood war thread btw) | ||
|
jimminy_kriket
Canada5529 Posts
| ||
|
MiniRoman
Canada3953 Posts
| ||
|
evanthebouncy!
United States12796 Posts
When the 2nd cycle is done it attacks again, when those units die it moves to third cycle, and then stays on third cycle for the rest of the game." So basically horrible multitask macroing while attacking? No wonder they're nub. | ||
|
VIB
Brazil3567 Posts
I cannot comprehend how can it be "fun" to watch and analise AI created by others O.o You're just studying the person who wrote it, if you didn't notice... | ||
|
BluzMan
Russian Federation4235 Posts
| ||
Carnac
Germany / USA16648 Posts
On April 15 2008 08:32 fusionsdf wrote: Yeah, 90% sure this is true. The AI can definitely do different things. For Zerg: -4 pool -12 hatch expo *I think -9 pool, hatch in base, mass sunken For protoss: -3 gate, save up zealots till 12, attack. -cannons, and 14? nexus -DT rush For Terran: -marines and medics -Anything else???? also, the computer will pretty much always time its attacks together if there are more than one and they are allied. Pretty sure they wont delay 4 pool or dt rush, but every other attack is delayed until both are done first cycle attack. I should point out that bw ai work off 3 basic cycles. Basically they follow a preset build order (like 12 zealots) execute it, and then when those units die, it moves to the next cycle. When the 2nd cycle is done it attacks again, when those units die it moves to third cycle, and then stays on third cycle for the rest of the game. The Terran wave is: * 12 Marines, 3 Medics vs. Terrans OR 10 Marines, 4 Firebats, 3 Medics vs. Protoss or Zerg * 5 Tanks, 12 Marines, 3 Medics * 4 Battlecruisers, 6 Wraiths, 2 Valkyries, 3 Tanks, 2 Goliaths, 3 Ghosts, 10 Marines, 2 Medics, and a Partridge in a Pear Tree. Zerg: Zerg AI: The Zerg AI uses only one general strategy regardless of map and enemy race. There is a random chance of using a pool first 6 zergling rush. If used, nothing else changes except that their later attacks take longer to prepare. The attack waves are as follows: *initial lings/build * 20 Hydralisks, 2 Lurkers * 9 Mutalisks, 4 Devourers, 4 Guardians, 8 Scourge * 60 Zerglings The final attack loop is variable. The AI will pick one of the following for each attack: * 12 Mutalisks, 4 Devourers, 4 Guardians * 6 Fully Upgraded Ultralisks, 12 Hydralisks * 20 Hydralisks * 20 Hydralisks, 20 Zergling * 40 Zerglings with the Adrenal Glands upgrade Protoss: Protoss AI: The Protoss AI has 4 basic attack methods: Normal, Dark Templar Attack, Dark Templar Defense, and Double Nexus. The Dark Templar Attack method is used if the scouting unit sees the enemy may not be ready with detectors. Dark Templar Defense is used if a scouting unit sees an enemy Protoss player possibly making Dark Templars. Double Nexus is used if the map seems to provide a close, guardable expansion. Normal is used in all other cases. The attack waves on land maps are as follows: * 12 Zealots if using Normal OR 12 Zealots and 1 Dark Templar if using Dark Templar Attack * 12 Zealots, 6 Dragoons, 2 Dark Templar * 12 Zealots, 8 Scouts, 4 Corsairs * 10 Scouts, 8 Corsairs, 5 Carriers See: http://www.battle.net/scc/faq/aiscripts.shtml most of this sounds accurate, however the comp doesnt 9 pool. it either uses 4 pool or goes hatch pool (sometimes hatch in main, sometimes at exp) | ||
GrandInquisitor
New York City13113 Posts
A familiar battle: Siege Tank vs. Mutalisk. Also popular is the Valkyrie vs. Dragoon fight. The Terran AI always seems to have the wrong unit, at the wrong time, in the wrong place, doing the wrong thing, and not nearly enough of them. That made me laugh xD | ||
|
kovarex
Czech Republic65 Posts
Much more interesting would be to make some way (library) to control units with computer program, so it would have (almost) limitless posibilities (same as player) and could also micro, and if the library was standardised, it could also ensure that the code itself doesn't maphack. I would not like to do the hard work with the library to control bw, but if it was done, I would really !!ENJOY!! making bot playing bw, it is very challenging task and even small modules to play for example just micro (so bots could compete in some micro maps) would be interesting enough. It would be funny to see for example bots playing on iccup, getting ranks, so one would be pushed to be able to beat at least computer (like in go). I really got cought by the idea, so if there are any programmers around who would like to do something like that I would like to know, it is also possible that the "library" I'm talking about already exists, I will google a bit. | ||
|
Goosey
United States695 Posts
On April 22 2008 23:47 kovarex wrote: + Show Spoiler + I believe that the bot competition uses some simple and very limited scripts, so you can't define micro etc, but anyway evolution programming could be nice (and funny) way how to find very good scrips (My firend was making half life bots using evolution programming) Much more interesting would be to make some way (library) to control units with computer program, so it would have (almost) limitless posibilities (same as player) and could also micro, and if the library was standardised, it could also ensure that the code itself doesn't maphack. I would not like to do the hard work with the library to control bw, but if it was done, I would really !!ENJOY!! making bot playing bw, it is very challenging task and even small modules to play for example just micro (so bots could compete in some micro maps) would be interesting enough. It would be funny to see for example bots playing on iccup, getting ranks, so one would be pushed to be able to beat at least computer (like in go). I really got cought by the idea, so if there are any programmers around who would like to do something like that I would like to know, it is also possible that the "library" I'm talking about already exists, I will google a bit. I was thinking about this the other day, of trying to use a BWL plugin to create an AI which operates based on the information the human has and uses the human's controls (actually generating mouse moves, clicks, keyboard events). I have not looked into how much information the BWL library has access to. To do this you would need to be able to access the data for fog of war, all visible unit positions, resource count, etc etc. If this is possible through their API it sounds like a potentially interesting project to me. I am a programmer in games industry, but have never done a complex AI, so it would be a fun learning experience.. | ||
|
omnigol
United States166 Posts
| ||
|
VIB
Brazil3567 Posts
On April 23 2008 00:25 Goosey wrote: I have plenty of experience with AI and I always loved the idea of creating one for sc/wc since I ever started playing it many years ago. But now I finally got the c++ knowledge to actually make a program that could do it. Mimicking key and mouse strokes would be easy. I just don't know anything about where/how the variables unit position / fog of war are stored. Any idea of how I can get started to find that information? I'm sure many people done this before (finding those variables). I was thinking about this the other day, of trying to use a BWL plugin to create an AI which operates based on the information the human has and uses the human's controls (actually generating mouse moves, clicks, keyboard events). I have not looked into how much information the BWL library has access to. To do this you would need to be able to access the data for fog of war, all visible unit positions, resource count, etc etc. If this is possible through their API it sounds like a potentially interesting project to me. I am a programmer in games industry, but have never done a complex AI, so it would be a fun learning experience.. | ||
| ||
