VIB, Goosey, kovarex: Your interest in improving the computers behaviour is always a welcome sight for the dream that Entropy and Racine started almost a half-decade ago. Considering how dynamic Starcraft is and how limited the internal Ai engine and how it deals with scripts are, Starcraft Ai is actually quite good. The UMS scene and recent Micro grid challenges such as Star's vulture Ai and the Muta Challenge here at TL.net have proven that with some decent micro, a computer can be quite difficult... and fun!
Few ppl have experimented with controlling the Ai outside of UMS triggers and the simple ai scripts, but there have been some attempts. What kinds of things are necessary to accomplish this, I do not know, but our small family @ Broodwarai.com might be a good place to snoop around as we have some very gosu programmers lurking there and we would love to have more!
*** as for the "bot contest" in first post *** Interesting study, but a bit skewed. I would agree with omnigol in that "why isn't this guy testing 1v1 Ai matchups on normal maps?" We have been running the Ai vs. Ai War that Entropy linked to for nearly 4 years and different races fair better on different maps even as computers. Furthermore the "insane" Ai's that were used in his challenge are junk despite their cheating. In fact our non-cheating scripts blew them away on nearly every map in our BWAIWar II. The Protoss insane being the exception.
Still it's fun to see ppl interested in something as old and simple as Starcraft Ai. ... here's looking forward to Starcraft2!
Thanks a lot Bajadulce, I'll surely take a look at that and hope we can help each other. I'm really really glad to see I'm not the only one so interested in the subject. TSL #16 winner = my bot >D
Actually, I think that being able to modify the starcraft/BW AI would give the community yet another great learning tool. One would be able to practice against one particular strategy without having to rely exclusively on training partners. In modern chess, the computer has been fully integrated in the learning process and is even helping finding new counters to popular strategies. If only the people that made deepblue were starcraft enthusiasts
On April 23 2008 06:00 misterroboto wrote: Actually, I think that being able to modify the starcraft/BW AI would give the community yet another great learning tool. One would be able to practice against one particular strategy without having to rely exclusively on training partners. In modern chess, the computer has been fully integrated in the learning process and is even helping finding new counters to popular strategies. If only the people that made deepblue were starcraft enthusiasts
Select AI Race: Protoss Select AI Skill: A+ Select AI APM: 350 Select AI BO: Bisu FE Fast-Sair
I've done some 1:1 AI tests on normal AI and if I remember correctly T was the best unless it got cheesed with dt or 4pool. Zerg is ofcourse the worst. Not 100% sure on te TvP issue.
EDIT:I did this many years ago so my memory is kinda shady.
Bajace: Ok I moved the discussion to the broodwarai forums, the problem is that they concentrate 99% of the time on script programing, but I hope I can get some help there, I'm more and more decided to work on that, and after all I could use it as school work.
PS. I already made program that solves L&D go problems quite well (similar to bw micro but in different game), so bw could be another challenge.
I would also want to ask if anyone made program that will make this: You insert: what is the fastest way to get 24 speed zerglins? Answer: some BO
This should be quite easy to do, all you need to know is the effectivness table for mining and some estimations like - time to go with probe to build pylon etc.
On April 23 2008 06:54 kovarex wrote: I would also want to ask if anyone made program that will make this: You insert: what is the fastest way to get 24 speed zerglins? Answer: some BO
Doubt it, doesn't think it's as easy as you make it sound. It would depend so heavily on the map (changes mining speed, possible expo timing) and some hard-coded game mechanics (building placement, worker movement/collisions) that to estimate anything with a reasonable error margin you would need to actually simulate it many times with an AI and learn from the results. Of course you could make some very rough estimation for all of those but it's so many little errors that after multiplying them all you'll actually have a big error margin and the final result would be very very unreliable.
On April 23 2008 06:54 kovarex wrote: Bajace: Ok I moved the discussion to the broodwarai forums, the problem is that they concentrate 99% of the time on script programing, but I hope I can get some help there, I'm more and more decided to work on that, and after all I could use it as school work.
PS. I already made program that solves L&D go problems quite well (similar to bw micro but in different game), so bw could be another challenge.
I would also want to ask if anyone made program that will make this: You insert: what is the fastest way to get 24 speed zerglins? Answer: some BO
This should be quite easy to do, all you need to know is the effectivness table for mining and some estimations like - time to go with probe to build pylon etc.
Yeah it exists and i've used it some years ago. The version 2.0 was very cool, but it was slow. And there was no 'permanent SCV production' option, so the fastest BO was with minimum SCV. I guess its ok for hardcore cheese.
When I don't use the building placement into consideration and just estimate, that it takes x time to make building in main, y time to go to expansion etc it will be accurate enough.
I believe that error don't multiply, the error will be the smaller the more probes will be already mining, and building producing, and I believe it will be really small.
I believe it can get to like 95% or so of accuracy which is far good enough, the purpose would be to use it as tool when making new bo's.
I will try to make it in the next week/month and I can show the results
If only the people that made deepblue were starcraft enthusiasts
Chess is a static game on a static board and so there are limited permutations of possible scenarios. An rts such as Starcraft would be almost impossible to account for even a fraction of the possible scenarios. But DeepBlueStarcraft would be sweet!
I would also want to ask if anyone made program that will make this: You insert: what is the fastest way to get 24 speed zerglins? Answer: some BO
We have a program in our database called "evolution forge" basically you plug in what you want to build say 12 hydralisks and it calculates the fastest build order. It's an interesting program, but a bit of a novelty more than anything else. Here is the program if you are interested in checking it out regardless: EvolutionForge v1.62
I remember one memorable game a few years back I dropped a pair of tanks and four gols on a ledge on Arctic Station and somehow the tanks accumulated 50 kills each.
Zbath's original idea was a script that contained nothing but endless zerglings w/ eventual defiler/dark swarm and very little else. It was meant to be a funny little script designed for my 12yr old nephew and some of his friends to play team melee TvZ on a normal map. Here the excitement was in the first 10 minutes as you desperately fought back wave after wave of ever increasing zerglings (upwards of 100) by bunkering down/firebats/medics etc. Eventually you would manage to scrape by and work your way towards comps base.
Zbath was meant to be gay. It was always intended as somewhat of a joke, but my nephew sure loved it. The Fa$te$t community also got a hold of it and seemed to enjoy it's wackyness. ... Ya it was/is a pretty wacky Ai.
ZBATH_AIWar_v1.0.zip Oddly enough there's actually a newer version. The version you know of was heavily nerfed!. Here is the Ai vs. Ai version w/ a just as obnoxious Protoss. Basically it would wipe out about 7 Blizzard Insane Ai's without much problem. ... ya it's gay as all hell, but that's the point!.. MUST KILL ALL OTHER AI'S!
Tried some PvT 1:1 games and it was 4-2 in favour for the toss. The games terran won was maps with ramps. Most of the games were long and the protoss managed to MC terran units in two if them. I'd say that P has a slight edge and then it is some map balance ect. Anyway the ai ranks 1 : P 2 : T 3 : Z. The Insane ai conclusion is due to that they have strict build orders and apparently the terran has a very lousy one compared to the other ai scripts.
The most ideal thing one shuld do when it comes to ai scrips is to make one that does a normal "rush" build really well, like a 9pool speedling-->all lings with cut gas or a 7z-->goons so they could be used as a ok teammate when you are a friend short. Atleast thats the scenario most of the times when I have to use the ai playing bw.
Does any of those AI using scripts + triggers deal correctly with worker harass? (like when you send 1 scouting probe to attack the AI before they get anything useful and they send in all workers to defend and literally stops all mining for chasing 1 worker across the map)
On April 23 2008 07:50 kovarex wrote: I will try to make it in the next week/month and I can show the results
Hope it works ^^ make sure you post it here when it's done, I would love to know.
EPT
