I have published the 1v1 ladder maps on Battle.Net using a new AI called Fyn AI, which is essentially the Green Tea AI. The main difference between Fyn AI and Green Tea AI is I removed the computer's ability to have vision without scouting because I am not a fan of cheating AIs.
I am a Silver league player so I have no idea at what level this AI becomes useless, but it is certainly challenging for me. For those not familiar with Green Tea AI, Very Easy and Easy do not micro and have an APM cap. Medium is what i use and it adds in micro and has no APM cap. Anything harder than Medium cheats so I don't use them, but I left them in the AI. If nothing else it is much more aggressive and expands more than Blizzard AI.
I published all the 1v1 ladder maps on the NA servers and thanks to n3wb13, Robqxz and Black Gun they are also on the SEA and EU servers as well now. You can find these maps on BNet by searching for "fyn".
If anyone wants to publish this AI on other servers or publish maps like some iCCup maps or w/e, the AI script and a guide on how to publish it can be found here, a website I created for Fyn AI. Thanks in advance for helping the SC2 and TL community!
At some point I may make use of an issue log built in to Google Code, but for now if anyone has any problems with publishing the AI on another server or bugs with the AI itself or just feedback in general feel free to leave a post here. You can also contact me in game Fyndor.408
An alpha version of the next upcoming release of Fyn AI is on NA Battle.net. If you would like to help test read the spoiler below. + Show Spoiler +
If anyone on the NA servers has some spare time, I need to test all the builds in the upcoming version of the AI and make sure they are labeled properly. The version I posted is very alpha. Not all the features are enabled or working yet and there are known bugs in the interface.
If you would like to help test, what you will do is log on to NA Battle.net server and search for "faitest" in the map pool. Add 1 AI opponent of what ever race you want to test and leave it on medium. Start the game. When the game starts there will be a new dialog. Uncheck the "Random Build" checkbox and select a build you want the computer to do. Hit the start button and the game will start. Play against the computer and see if the description of the build you chose matches what the computer actually did. Do this for each build (or as many as you can manage to do). If you like it wouldn't hurt let me know what you think of the build, whether it should be in the build list, and whether it is a good build to use against your race (tell me your race of course). But consider all that stuff extra credit. I mainly need to know that it did what it was labeled as so people aren't like WTF when it is released on all the maps in its public release.
****NOTE**** As of Patch 1.3 I have to remove the faitest map to make room. When the next round of changes to the new version are ready (larva fix) I will find upload it again under a different account.
Go to http://sc2.nibbits.com/assets/green-tea-ai/ for full details, but Hard ramps up 1 to 1.5 rate over 10 min, and Very Hard does it 1 to 2.0 over 10 min. The AI I posted is essentially the AI on that page minus the part where it has vision all the time.
Will try this out later, I love giving AI's a good spin now and then to see how the community is advancing. Love the idea of a decent AI that actually uses/needs vision.
I tried this out. It's great. Doesn't need resource boosting because the AI aggressively expands and saturates. The AI still has some problems of course, mainly the way it treats it's units. One medium zerg computer sent 20 lings at my blocked ramp and did not retreat them until I killed half of them (I had scvs on my wall too). It's not smart enough to make a game interesting in terms of utilizing unit control, but it does mix and match the unit combination nicely sometimes (think ling/hydra/banelings/infestors or thor/siege tank/marine/marauder/dropship/helleon mixes).
Primarily, this is a good way to practice at platinum and below against macro imo.
On January 05 2011 14:24 CooDu wrote: I'll give this a go tonight. Currently at a Gold level so should be good fun in the late hours of the night when not a lot of players are on SEA.
I'll be tuning into GSL actually so this weekend hopefully lol D: Keep up the good work!
Well I am NA so that may mean you can't get them If someone has an NA acct that wants to publish on other servers you are more than welcome to. I am not quite sure why they want to separate the maps pools anyway. Seems a bit silly from a developers point of view.
The AI still has some problems of course, mainly the way it treats it's units. One medium zerg computer sent 20 lings at my blocked ramp and did not retreat them until I killed half of them (I had scvs on my wall too). It's not smart enough to make a game interesting in terms of utilizing unit control, but it does mix and match the unit combination nicely sometimes
Yea it can do some dumb things early game. Cannons trip it up for one. From what I can tell to retreat it has to see an imbalance in units, and I don't think it takes in to account stuff like cannons when calculating whether it should retreat. My zerg strats involve cannons so this can be a problem. Overall I think its good at training you to macro because it certainly always does and will walk all over you if you let it expand too much (well atleast me of course). And like you said it has more variety in builds than compared to Blizz AI.
As I stated above, any NA user with EU etc accounts is welcome to republish the maps on other servers. I don't have anything but a NA account so there is not much I can do on my end. Sorry.
ah, no wonder it's so tough on hard lol. didn't read the part about hard ai cheating.
but definitely a lot more early pokes and aggressions, and this AI is a macro beast, taking expansions very quickly. gonna try it on medium later and report back
AI doesn't cheat. Blizzard said that they removed the capability of the AI to know what you are doing without scouting. If it can't scout you , it will do a basic build , if it scouts you , it will try a counter.
It is said in one of blizzard's -making of sc2- video from the SC2 Collectors Edition.
I will try out your maps tho.
Edit: Can't you just put the map for download here? I am on EU.
Blizzard was saying their stock AI, aka Blizzard maps, does not cheat with vision like Brood War AI did; not that the API did not allow you to make mods that could have vision. Unless it was patched in the last 2 weeks or so it, it can have vision if you make the call in your AI to turn it on.
I removed a call the Green Tea AI, not Blizzard AI, used that gave vision to the AI because Green Tea used vision to make sure their AI had good info on what it needed to counter you. You could see that had vision before I took the call out. Without detection it would freak out following invisible units that it couldnt attack but it could see. It annoyed me that as protoss it was impossible to get vision with an obs. So it took I out and added scouting back in. It is a weaker AI than standard Green Tea, but I don't want something that cheats so you have what I published
Edit: Can't you just put the map for download here? I am on EU.
http://code.google.com/p/fyn-ai/ Here is the AI script. It's not too hard to figure out if you haven't done it before. Essentially what you do is load a map in the galaxy editor, import those files in the zip file found at that URL, then publish the map to EU. Rinse repeat for each map. Let me know if you need more specifics to get it published. If you do publish them, please use the same titles as I did so people reading this post can find them. Thanks in advance for helping the SC2 and TL community!
I just played one of these t osee what it was like on Very Hard.... Holy cow what was that? I got steam rolled by the 10 minute mark. after watching the replay the toss ai I was against had 3 bases, a templay archives and chargelots and was pumping units straight into my nat like crazy.
On January 06 2011 09:23 fyndor wrote: Oh is that possible? You can just log on to NA? So does that mean as an NA account I can log on to other servers? How is this done?
It's just a SEA thing.
Blizz gave SEA players access to NA servers as well.
Anyway, I tried out this maps and it's much better practice than the regular AI! Recommended it to my friend who is quite new to SC2 so he can get a better feel of actually playing realistic timings.
Okay seriously, is there any way to beat this way without some sort of rush? Cause it kills me easy in the medium difficulty games even. The only thing that is keeping me alive half the time is my micro abilities...
Sorry but I have no experience with the editor. However I tried to import the files and I get this error ->
"The following files are using reserved file names and cannot be imported:
TriggerLibs\GameData\GameData.galaxy"
If you got that far it means you are close. I think that folder is just there to play it offline stand alone with a launcher. What you do is when you are importing you uncheck that folder/tree. Leave everything else checked. Then I save it. Then close the import window. Then you go to publish and you have to tell it all languages, name it, and switch it to public then publish it. Each time you change things in that publish dialog it "resets" certain things which is a kind of annoying interface. Basically when you see no errors you are ready to publish.
Okay seriously, is there any way to beat this way without some sort of rush? Cause it kills me easy in the medium difficulty games even. The only thing that is keeping me alive half the time is my micro abilities...
Yea early pressure is the only way I can beat it, but I figure that is because I am silver. On medium, it starts to micro, and it expands a lot so if you sit on 1 or 2 bases it eventually overruns you. I am working on trying to beat it in a more macro style game. For one, when it sees what units you have it follows a counter chart that Blizz has in their default AI, so if you stay in one tech tree for too long and dont keep up with or stop its expanding, it will get more than enough of the right stuff to finish you off.
I wasn't aware that very hard AI on normal settings mined more minerals? R u guys sure?
very hard ai plays very very badly...
Yea I am pretty sure even the default Blizzard AI cheats after Hard. Play a game against it, write down its build timings on the hardest setting then try to mimic it. You will notice stuff won't line up which is because it mines more minerals than you so the build orders they designed it to use don't work with regular mining settings. Ultimately though they had to just ramp up mining like that because they don't exactly know how to accomplish the same thing with intelligence vs brute force. And for that matter this AI after medium follows the same model. Its much easier to just essentially handicap you than to make an AI that can outsmart you if are good player
I finally beat a protoss ai on medium. It made a forward pylon and everything on it's first attack and kept the pressure up. Somehow I managed to keep up my macro and eventually overran it with roaches.
Seems like pretty good AI for practice. The reg very hard ai never gave me trouble cause it would wait for the 7 minute mark or later to attack. This one gets 4 units and started to pressure. It'll also micro pretty decently.
Sorry to double post here, but I think there are some suggestions I have that could make this AI the "perfect" practice ai.
First off, if you could program the terran and protoss ai to somehow know to wall off their ramp like actually players do against zerg, this would be almost perfect. Why? Because as a zerg once their initial attack utterly fails I can simply counter attack and almsot disable them without much effort. A wall-off would actually be tougher to fight against becuase I may be able to keep them from expanding I won't be able to run into their base with my initial 10 or 12 lings and wreak some havoc.
On a similiar note, zerg ai seems actually perfect except that it doesn't spread creep. If the zerg ai would spread creep then I think that it would be a tough ai to beat.
Secondly, I think that the ai needs to be programmed with some "standard" built orders. I would like to see if toss will actually 4 gate instead of 2 gate cyber expanding most games and then teching to counter what they see (or not incase of the high templars to counter roach?). This would allows players to get that game feel of what they may face on the ladder or in tournaments alittle better.
These are just some of my opinions. I don't know if you can actually program all of these because I stopped trying mod back when I was 12 , but if you could then that would be a wonderful addition to an already great ai system.
Sorry but I have no experience with the editor. However I tried to import the files and I get this error ->
"The following files are using reserved file names and cannot be imported:
TriggerLibs\GameData\GameData.galaxy"
If you got that far it means you are close. I think that folder is just there to play it offline stand alone with a launcher. What you do is when you are importing you uncheck that folder/tree. Leave everything else checked. Then I save it. Then close the import window. Then you go to publish and you have to tell it all languages, name it, and switch it to public then publish it. Each time you change things in that publish dialog it "resets" certain things which is a kind of annoying interface. Basically when you see no errors you are ready to publish.
Ok thx , done that. Published it as "private" to test it but I haven't noticed any change in how the AI plays. Maybe I've done something wrong?
Did you make sure it was not set to "Locked". When I set it to locked versus unlocked it didn't work. You will know it works because when the game starts there is some mod info that pops up on the bottom left of the screen where messages are shown.
First off, if you could program the terran and protoss ai to somehow know to wall off their ramp like actually players do against zerg, this would be almost perfect. Why? Because as a zerg once their initial attack utterly fails I can simply counter attack and almsot disable them without much effort. A wall-off would actually be tougher to fight against becuase I may be able to keep them from expanding I won't be able to run into their base with my initial 10 or 12 lings and wreak some havoc.
On a similiar note, zerg ai seems actually perfect except that it doesn't spread creep. If the zerg ai would spread creep then I think that it would be a tough ai to beat.
Secondly, I think that the ai needs to be programmed with some "standard" built orders. I would like to see if toss will actually 4 gate instead of 2 gate cyber expanding most games and then teching to counter what they see (or not incase of the high templars to counter roach?). This would allows players to get that game feel of what they may face on the ladder or in tournaments alittle better.
I may try at some point to improve it, but there are going to be limits to what I can and am willing to put the time in to do. If you search for Dark Green Commander there actually was an effort to adapt the original Green Tea AI to use more standard builds. I don't think there is a version out that works with the current game though, and I certainly would like at some point to make the different "commanders" in this AI use standard builds. I didn't do it from the start because I have not spent enough time looking at the code to get the full picture of what they were doing so I didn't make it weaker than the original. For one I believe it has this system where if they have too much air it will switch to a more air heavy build at some point and go to ground when necessary also. I would have to make sure what ever I did kept that model intact so that it could still react correctly. Plus the builds we all see online really is about the first few minutes of the fight. But how many gates / robos / stargates etc and how do you transition to mid and late game matter depending on the build you started with and what kind of economy you have and so on. As I said I am a silver player. I play Protoss, but know enough about the other races that I am not clueless. That being said though am I really the best person to be designing builds yet? I still have a rough time beating the Medium AI unless I put on more early pressure than I am normally comfortable with. I feel that at this time if I attempted to rework the builds I could change it in a way that would probably still do well against myself, but I would not be skilled enough to really make an accurate judgement whether the changes were good.
As for the first point these two examples are probably a bit tricky. As it stands there is no handbook for using this mod API, and as far as I know what the community has is a list of functions and parameter names, and the default AI script that blizzard uses on BNet. I have only grazed the surface, but I suspect both of those could be solved.
I could have sworn I read something about creep spread in the code, though for what ever reason I never noticed it wasn't doing it. One thing I see in this AI code is that usually if not all of the time it tells the AI to build things near other things. Its seems to always be a relative location kind of system. So it will say build a pylon near nexus and the system will try and find a spot not occupied to build it near where you wanted. It does this because exactness is a problem when stuff can be blocked. How do you tell it to accomplish a wall if a drone is blocking it. Its easy for a human but they may not have the tools available to make that easy, I will have to see. Also unless they have a function that just says block your ramp with these buildings (dont think that exists), you would have to give exact coordinates for positioning which I dont think I have ever seen this AI do. That is probably because it is generic AI that works for all maps. If possible it would likely require figuring out all the placements on every map, for all the spawning positions, for the different races, for the different wall types that may change depending on builds. The creep thing could be even crazier in some ways but easier in others. Again I don't know if they have to tools available to easily say build another creep tumor "towards" their base or expo or to your expo etc. But I really haven't dug in enough to really know how possible this is. It may not be that hard if you just had certain positions you precoded in where you wanted them to be built, but it would probably be time consuming to figure out from all the spawn positions and make it work in this "fuzzy" world where stuff can be blocking your predefined positions.
Like I said I may play with using standard builds, but I would certainly need help from the community in testing it to make sure it was an improvement. The creep tumor and walling I don't really care to tackle because of the time likely required to solve, but someone else is more than welcome to solve it for me/us and I will add your patch in.
On a similiar note, zerg ai seems actually perfect except that it doesn't spread creep. If the zerg ai would spread creep then I think that it would be a tough ai to beat.
Ok so I just checked and there is certainly code for spreading creep. I will have to figure out if its something I did to break it or if it just doesnt work. Hopefully I can fix that pretty soon.
On a similiar note, zerg ai seems actually perfect except that it doesn't spread creep. If the zerg ai would spread creep then I think that it would be a tough ai to beat.
Ok so I just checked and there is certainly code for spreading creep. I will have to figure out if its something I did to break it or if it just doesnt work. Hopefully I can fix that pretty soon.
Actually it does spread creep. You could certainly start doing it faster but it will spread creep. I saw it first start doing it around 16 min mark. It appears though that 2 of the commanders may not have the code set to do it. Not sure why they would omit it, will have to check it out in offline testing. One thing I did notice when clicking on his overlords in the replay is that it did have them all set to spread creep, but I think it uses all of them once it gets speed as scouts. This is presumably because I turned scouting back on, so they are never sitting still long enough to really drop it. It may have worked better with original Green Tea with vision enabled and no scouting. Hopefully I can figure out how to improve it and I will see if it needs to be added to those 2 builds it was omitted from.
Side note: It actually did drop creep in my naturals mineral line which was kind of funny. Not sure what the point was, maybe to improve a ling drive-by.
I wasn't aware that very hard AI on normal settings mined more minerals? R u guys sure?
very hard ai plays very very badly...
Yea I am pretty sure even the default Blizzard AI cheats after Hard. Play a game against it, write down its build timings on the hardest setting then try to mimic it. You will notice stuff won't line up which is because it mines more minerals than you so the build orders they designed it to use don't work with regular mining settings. Ultimately though they had to just ramp up mining like that because they don't exactly know how to accomplish the same thing with intelligence vs brute force. And for that matter this AI after medium follows the same model. Its much easier to just essentially handicap you than to make an AI that can outsmart you if are good player
I've played this on medium, it isn't too bad, certainly better than the blizz ai. But I feel as though it just constantly sends in small clumps of units too frequently which easily get slaughtered by my units. I understand that it does retreat sometimes, but it just doesn't want to macro up to a sizeable army before pushing...Gets really annoying as zerg, when you have to make barely enough units to survive, and you have to make a few lings every 30 secs just to stop the 3 marines and 1 marauder on the way to your base...
Is it just me or does this AI seem a bit too hard on medium on the jungle basin map? The ai immediately expands to the safer expansion while sending out it's first wave of pressure. The pressure can be held off as zerg, but the constant pressure following causes the constant production of combat units and also no way to expand outward past the safe expansion and also no apparent way to have time to break through rocks. By 9 minutes the CPU is already taking a 3rd and has 60+ workers plus keeps pelting you with decently sized armies to keep you inside your base. Suddenly by 12 minutes there was a BC with a large amount of marines and marauders on the map including some hellions and I was getting wrecked.
This was on medium ai (The apm of the ai was upwards of 600!? by this point and was multitasking beyond what anyone I have every seen could do).
Is it just me or does this AI seem a bit too hard on medium on the jungle basin map? The ai immediately expands to the safer expansion while sending out it's first wave of pressure. The pressure can be held off as zerg, but the constant pressure following causes the constant production of combat units and also no way to expand outward past the safe expansion and also no apparent way to have time to break through rocks. By 9 minutes the CPU is already taking a 3rd and has 60+ workers plus keeps pelting you with decently sized armies to keep you inside your base. Suddenly by 12 minutes there was a BC with a large amount of marines and marauders on the map including some hellions and I was getting wrecked.
This was on medium ai (The apm of the ai was upwards of 600!? by this point and was multitasking beyond what anyone I have every seen could do).
Well for one I can tell you there is not difference between the maps codewise. As far as I can tell the AI acts generically no matter the map. Certainly some maps will be more or less conducive to the AI. The original AI never shipped with the pure 1v1 maps (Steppes, Xel'Naga, Jungle) and I assumed this was because its weakness to early pressure. One thing you could do would be try and prevent it from taking a 3rd and eventually try get your own up and try and "starve" it to death. I would assume that is kind of the strat for that map vs computer or human, though I am Siilver so what do I know : /
As far as APM goes, the Very Easy and Easy difficulties have an APM cap and no micro. Easy has more APM than Very Easy. On Medium the APM cap is removed and micro is added. I am not 100% sure, but I would imagine the difficulties I have with medium are more due to it micro'ing than APM. Plus I am not sure what would happen if I turned the APM cap on for Medium, I would imagine its micro would suffer. I don't know if I really want to mess with that, but I have been considering adding "commands" in to the mods so you could do things like reset, and tell it to use a specific commander (aka build). If I do that I may also add in a way to set the APM to your desired setting if that is something the community would like.
Edit: Does anyone by chance know of a custom map that has user interaction using "chat commands" who's source code is publically available. Looking for some example code so I can start adding in commands which would help alot when I start modifying the builds to more closely match what the pro's are doing and need the ability to select which commander I want to play.
I wasn't aware that very hard AI on normal settings mined more minerals? R u guys sure? very hard ai plays very very badly...
Yea I am pretty sure even the default Blizzard AI cheats after Hard.
I am pretty sure you're wrong. (only insane mineral cheats)
Anyway, when it comes to Green tea AI, I haven't played it recently (I played it like 2 months ago maybe, and it was very similar to it's beta level), but plan to play it again (hence my problem with importing the AI to a map).
The issues I had with the old ones, and I think is still an issue (especially with one that doesn't have revealed vision) is: 1. there are too many strategies used, by that I mean too many bad ones. The builds should be a bit less bad overall. Variety is nice, but some builds are far superior to others I'd say.
2. Armies (small ones oftentimes I guess) will be sent out, but if they can't win they will run away. This is a problem because it doesn't factor in the possibility of not being able to run away due to movement speeds. If a AI is facing a zerg with speedlings, they should not send out their army at all unless they know they can stim, medivac, blink, or micro their way out, or can just outright crush the enemy.
Personally, I like the maphacking for AI, since it's far too difficult (or too much work), to give the AI good enough "intuition" and senses to know where to scan, how to scout, where to scout, how to keep updated on units, and builds that can deal with various opponent strategies. The way I see it is when practicing, one should be practicing for worst-case scenarios, such as if the opponent gets a lucky scan, gets a lucky scout, or flat out correctly-assumes what strategy you're executing. I do not like mineral cheating at all though (I dislike how green tea has 2 modes of SEVERE mineral cheating), because even in worst-case scenarios, your opponent will NEVER have such a massive advantage like that, ever, assuming you play right. Yes, they might not provide a challenge then, but that just means the AI needs to be improved. I think mineral cheating also gives a disadvantage to the players playing them who are zerg, because they don't have units that can be super-efficient vs enemy units like siege tank, thor, missile turret, colossus, force field, phoenix, even marine (microed), which make winning a fight with a resource disadvantage an impossible feat even with perfect strategy execution and micro-management.
So I just played the AI a few times, and like I thought there was an issue with the builds.
One game in a TvZ, the terran goes 1 barracks marauder ( attack with 2+ marauder) into siege tanks vs my 11 pool expand into speedlings. It's absolutely terrible. 1 rax marauder is maybe plausible if the opponent is going roach rush, but I wasn't. Builds like this should simply never be called up in a TvZ.
Of course this is more green tea AI's problem than yours, but I guess if you wanted to branch off to another AI then it could be considered.
I do plan to make changes to the builds for sure, I just don't want to be too hasty in doing it so I just published as is for now until I have time to fully understand the code I am looking at. I do not believe it takes in to account the opposing race at all when choosing a starting build. It does a random build from a list. It will switch builds at some point if it determines it will not be able to handle your army with its production buildings. I think its mainly an air vs ground thing but that equation is a bit confusing. For one it would do much better if the build lists were tailored by race and even better if it then weeded others out because of map/spawn locations. That alone probably would help a lot. That technically shouldn't be too hard to do, but research in to the builds that it should be using will take some time unless someone has a good resource beyond telling me to go watch a ton of replays which is probably what I will end up doing. To the comment about too many strategies I think one of the pluses of Grean Tea and Fyn Ai is that the computer won't us the same build over and over. My goal is to keep a good number of builds per race, but make sure they are builds you would see the pro's using against that race. I want to probably add some crazy all ins, maybe even cheese because it would be good practice but that will be last on my list.
The engagement AI certainly needs to be improved, especially since I removed vision. Before I removed vision, you would see the CPU outside your base about to attack then "flinch" back because you just spawned guys inside your base and its counter chart said it didn't have enough units any more to counter. Then sec later it would turn back to you because units just came out for it that then make its engagement calculation say that it had enough to take you on. It basically would dance down the field, and you knew when it walked in your base you were screwed, because the math said you shouldn't be able to handle what he has. It worked really effectively and I would like to try to get as close to its original ability to size up your army and engage when it is safe. But for my AI, it will never have vision, unless you could possibly show me a way to have it see what you have in order to count, without actually knowing where those units are until it scouts. I need the AI not to know that I have a proxy stargate in the 2nd expo behind the rocks in Xel'Naga for instance. I need it not to see my observer before it even enters the base and follow it freaking out because it has nothing it can use to REALLY see it. I am really just not going to budge on that because at that point the AI quits being fun to me. What I do need to do is make the Terran make use of scans, and the protoss use observers. I dont think they do that because the original AI had no need for it. But even so, I don't think it is reacting properly when it does engage your units and you out number it. It seemed to be more cautious before when it had vision. But I think that is just a matter of tweaking some numbers which should not be too hard.
I don't like the extra minerals either above medium. But I am sure they did it for the same reason I am leaving it, which is because it is the only way to have something harder than Medium without just making the AI better and I am sure the AI is about as good as they knew how to make it so cheating is the only option. My one thought on that subject is maybe I could scale the APM limit where instead of going unlimited at Medium, it is more gradual and the cap is finally removed at Insane. I could also add back in vision somewhere in the difficulties instead of minerals, but I would probably reserve that for Insane since its something I never want to play against The only thing is that if I remove the mineral cheat I would imagine it would hold no entertainment value for really good players since that is the only way it has a chance against an excellent player. I am certain open to good ideas about modifications to the difficulty levels so let me know if you have any ideas or if you like any of my suggestions.
Crap I was worried about that happening. The short of it is, yes, hopefully tonight, maybe later if I can't figure it out in a night. When I see the bases not showing up that means that there was an error in the script, usually syntax error. The way I deal with it right now when it happens is to look over whatever code changes I just did and try to spot the problem. I am really new to map/AI writing and have not learned what debugging tools are out there and for the most part I just try and write flawless code so I don't have to worry about it . For instance there may be a compiler out there that will show me where I am messing up syntax or have a wrong function name (likely issue, probably change or removed a function), but if there is I don't know about it. If anyone knows what tools are available to debug starcraft I would appreciate a point in the right direction.
I'm not really much of a ladder player and usually just spend my time 1v1ing/FFA against AI (usually very hard but I can play on weaknesses in some match ups in insane). However, very hard AI ALWAYS does the same build order (or very similar) and it gets tiring after awhile when I know exactly what is coming and can easily crush it, even if I'm doing something silly. However, whenever I play insane half the time I get sheer outmatched just because of the mineral advantage, especially early on if I can't compensate such as building bunkers as terran, etc.
However, your AI provides a brand new challenge and I even struggle at lower difficulties, so thank you for this! :D
The only thing I ask is if theres any potential/chance/etc of being able to see custom maps like the iCCup maps or MotM maps with the AI added (maybe even GSL haha)? I assume you would need the map file from the creator but any chance of seeing something like this happen? I don't mind the Blizzard maps but I love love love the custom maps, haha.
I think it was "dark tea AI" that had too many strategies, since I only see 3 builds from the terran I played.
One of the builds is terrible vs zerg (1 rax marauder into siege tank), another is decent (Marine marauder + random stuff later), last is best (marine-hellion into banshee and other inc. battlecruiser)
Because of this I'd say one of the biggest issues is strategy selection option, and/or automatic strategy chosen based on opponent race.
On January 12 2011 09:49 Graham wrote: Just wanted to say that I love this AI!
Thanks! Much of the credit is owed to the Green Tea AI creators, thus far I have merely made minor changes and published for everyone to access easily, but i fully intend to improve upon what is here and put my stamp on it.
However, very hard AI ALWAYS does the same build order (or very similar) and it gets tiring after awhile when I know exactly what is coming and can easily crush it, even if I'm doing something silly. However, whenever I play insane half the time I get sheer outmatched just because of the mineral advantage, especially early on if I can't compensate such as building bunkers as terran, etc.
As far as I understand from reading code (haven't really played above Medium for various reasons including it would just own me), the only difficulty that has 1 build is Insane. I would assume that this is because instead of ramping up mineral rate over time, Insane is full blast from the start. This would change timings completely. For instance the Protoss insane build starts with pylon on 7 then warps in 2 gates when that finishes. You would likely never do this with the normal mineral rate. You maybe could say they decided when making Green Tea that it was too much work to rework all the builds for insane.
Here is the code for Protoss initialization code for the Protoss AI. Even if you don't understand the C programming language you can kind of see what it is doing though this forum messes up the spacing.
if (PlayerDifficulty(player) == c_skirCheater) { //I believe c_skirCheater means insane, all the other the other difficulties are c_skirVeryHard etc
rnd = 8; } else { //This code is for all not insane mode AIs //It says that if the game has been running for less than 100 game seconds //then pick a random number between 1 and 7 and that number will be which build it does //If this code is called more than once, it is because it is after 100 seconds and the //AI decided it needed to change builds because of having too many or few air forces. //If the enemy has more air than ground, it will pick between the two air builds. Likewise if the //enemy is mostly a ground army it will choose between 5 ground builds.
GTChat("Vernam7: It's time to die !!!"); AISetMainState(player, e_mainState_OpenAir1, e_mainSubState_Unset); } }
In words what that code does is if its insane it does one build the whole game. If not it randomly chooses between 5 ground builds and 2 air builds. After 100 seconds in the game, it may decide at any point that it either had too much air or too much ground, so it will choose another build depending on which is the case from these same set of builds it started with. All builds are full start to finish builds (not sure if i will have it set up like this in the future), so when choosing another build if say the new build says build 2 gates and it has 2 gates already then it just skips that step. So the end result is it will skip ahead to the later part of the build where its making a stargate etc and just go from there. That is the build system in a nutshell.
You can think of difficulties Hard and Very Hard as soft enrages (term I am pulling from WoW). It slowly ramps up to fully enraged at I think 10 min. Insane mode starts enraged, so good luck. In the future you guys can help me decide on a better difficulty system if that is wanted and possible. In an ideal world you would choose between Bronze and whatever that greater than diamond league is now and it would play at those difficulty levels. We will never achieve that (by the time its possible this game won't be played anymore :p), but my goal is to get as close to that kind of setup as possible. I have a long way to go. If you think about it, very easy would maybe need to be dumbed down alot, certainly less expanding to be a bronze level player, though I am sure it does stupid things even bronze don't do at this point. Back to the point though, if there is an "enrage" and you are beating it, it is probably because you beat it before it fully enraged or shortly after it did. I wouldn't imagine too many people could handle 40 min game against this AI unless cheesed it and just kept it contained for 40 min. So if you are beating it before or shortly after 10 min, its build hasn't necessarily fully shown itself yet, so it may feel like the same build over and over.
Really what needs to happen is I need to improve it so it scouts alot better, because it uses that information to form the proper counter and right now it is not getting nearly enough scouting information if you are not engaging it often. If you looked at the code I posted 2 lines are about scouting. I am pretty sure, unless I have just not found it yet, that the AI uses out of the box Blizzard scouting. It is good for finding the enemy base and it probably does better than many players at late game expansion finding, but it doesn't do enough to keep tabs on what the enemy is making so it goes in to many engagements mostly blind. Even before updating all the builds, my main focus initially will be to watch what good players are doing in replays as far as scouting and see if I can mimic that. I think that would be the single biggest change I could make that would have a giant impact. After that it would be the build orders.
The only thing I ask is if theres any potential/chance/etc of being able to see custom maps like the iCCup maps or MotM maps with the AI added (maybe even GSL haha)? I assume you would need the map file from the creator but any chance of seeing something like this happen? I don't mind the Blizzard maps but I love love love the custom maps, haha.
I am pretty sure you can just download those maps using the map editor. The problem for me would be I need more accounts really to do more than I am doing now as far as map selection. It takes 9 slots to do the maps that are on the ladder (is Shakuras back yet?). That leaves me 1 map slot for testing or I have to pull a map down to test a change and noone wants that But... now that I think of it my girlfriend has an account too so that is 10 more slots I have. Beyond that I have distributed the source code, the link is in the first post and you can publish your own maps (as in iCCup etc) using it. The only thing would be making sure you have permission to republish whatever map you are adding the AI to. It is possible they wouldn't appreciate the clutter and confusion when seaching for iCCup, or they may think it is awesome, not sure. Ultimately I need to post a nice guide on how to publish maps using this AI. I need to do that like yesterday, because I can only publish on NA servers so I am relying on the community for these to be published on other servers. For instance, as soon as I fix what the patch did, hopefully n3wb13 will publish them to SEA servers
Edit: Btw, sorry I am so damn wordy. Not sure if anyone is reading my posts they are so long
Personally, I think you should let the AI maphack. As long as it's not mineral hacking, and it only has access to the same set of commands as any human player, then the AI will probably be more useful for a wider range skill levels and be much easier for you to program.
I understand that this is the main change you have made from green tea's version of the AI, but still, removing the map hacking is not the way to go IMO.
If you look at good players, they scout so well that they know what you are doing and know what the makeup of your army is 95 % of the time anyways. So really, allowing the AI to map hack is simulating an expert player's scouting at a much reduced complexity of the AI.
While the green tea ai is very good, what keeps it from beating very skilled players is its macro and micro.
Now, if there comes a day when the green tea ai routinely beats very skilled players, then maybe that would be the time to turn off map hacking.
If you look at good players, they scout so well that they know what you are doing and know what the makeup of your army is 95 % of the time anyways. So really, allowing the AI to map hack is simulating an expert player's scouting at a much reduced complexity of the AI.
That is 100% true, but my problem is the side effects that come with that. It is so hard to get any information with an observer. I haven't tried it by try burrowing roaches and see how well you can sneak up on Green Tea. And I think I remember a proxy stargate being found immediately at on Metalopolis that I put at the starting base base next to my enemy. I want an AI that I can practice with that will help me with my 1v1 against humans. All these side effects make you play so differently from what you would do against a human that in my opinion it makes it not so useful of a practice tool. As it stands, I question how useful anything greater than Medium AI would be for practice really since your opponents in 1v1 ladder won't have that so anything you see against Hard or greater shouldn't even be possible by a human (probably is possible though, not sure how finely tuned they made it, but it shouldn't be). But I left the harder difficulties in purely for entertainment value since not everyone is using this mod like me as a practice tool. After all this is a game so people do tend to like to play for fun
Now that I think of it, I had already planned to at some point introduce a dialog window that pops up before your game starts so you could change the APM settings to whatever you wanted. I could make the AI vision hack an option you select in that same dialog. That way I could get what I want and you could get the AI you like to play against. I can see the AI is much easier without the hack, but even so as a Silver player I still lose to it a lot depending on the build unless I use an extreme early pressure strat which it doesn't handle very well. Because of this, for me atleast, it is entertaining and useful as it stands. Due to the level I am and the feelings I get playing it I would say medium feels like maybe a Goldish player? Maybe silver? It for sure at times makes bronze mistakes I don't think I would have been able to beat it at all in my early bronze days though. Regardless, I wouldn't be surprised to hear golds or above saying medium feels like they are playing against a bronze player.
At this point I need to make the maps work again with the patch. I will release another version asap to fix that. And then maybe next I will release a version that has the intro dialog so I can make the people happy that thought its APM was too unrealistic and I can give you the ability to play with AI vision. I just have to figure out how dialogs work. The rest is nothing. I had planned to work on scouting next, then after that revamp every build so that it is using builds you need to learn to play against on the ladder, but if I revamp the builds first after adding the dialog, then I can make my camp and your camp happy while scouting only would help the non-vision people. So I guess thats the plan.
It's time to stop typing novels on TL and actually get to fixing these maps so people can start playing them again. Wish me luck
It took me all night to figure out how to find and extract the AI files out of the new and older version, but now that I have this patch and the previous patches stock AI I should be able to figure out what changed and what is tripping it up. I can't say for sure how much was affected, but looking at my text file comparison tool, not too terribly much of the stock AI was changed, so hopefully that means an equivalently small portion of Fyn AI needs to changed. I am pretty optimistic at this point. Really would have been nice for blizzard to just tell us the changes instead of having to do all this source code investigation work If I didn't have this Beyond Compare tool this would be a REALLY slow process. It is 4AM at this point so time for bed. Maybe tomorrow I will actually get a chance to see the new patch
On January 12 2011 19:04 fyndor wrote: Really would have been nice for blizzard to just tell us the changes instead of having to do all this source code investigation work If I didn't have this Beyond Compare tool this would be a REALLY slow process. It is 4AM at this point so time for bed. Maybe tomorrow I will actually get a chance to see the new patch
damn... you figure they would have learned from WoW after WoW launched, every patch broke all the UI addons people wrote. people complained for a long time until Blizzard finally release notes on what changes they made i can't believe they don't do that for SC2
It's not a huge problem, but deducing what changes have been made by doing a file comparison of 30 files is a bit silly when it probably would have just taken a few lines mention in the patch notes. Hell even a patch notes for mod makers specifically would be good. Take what they had about mods/custom maps in the patch notes, make a secondary patch notes for that and just expand bit. There weren't that many changes to mention, but if you didn't have the tools I do because programming is my profession, you would be doing line by line file comparisons manually on all these files to hopefully glean what changes are breaking your mod. It's just a bit silly when a little help on their end would fix all that : / For one, part of the problem may be that so few people do what I am doing (the AI side) that Blizzard is probably just not worried about it. I get a feeling most custom map makers for the most part have never have seen the AI code or know where it is or how you have to use 3rd party tools to extract it from Blizzard's inhouse "zip file" format. I think there are just so few people in my camp that Blizzard just wasn't worried about breaking our stuff and leaving us to figure out what went wrong : / Oh well, I think I should have this fixed by tonight or maybe tomorrow if all goes well, doesn't look too horribly bad.
Would it be possible to leave maphacking on, but program some special rules for dealing with invisible and burrowed units so the AI seems more human-like?
WOOOOOOOT! The Fyn AI loads atleast now in stand alone mode. Need to add the modified counter tables in real quick and do some quick testing to make sure I don't give you guys something that doesn't work at all and I will republish to NA and post the source code up. Likely within a couple of hours, hopefully worst case scenario early tomorrow evening. Was overall a minor change. I will post some info on TL about what was necessary to add to make it work and what blizzard thinks the counter changes should be (at least according to their AI) in case anyone fines it useful or interesting. The main thing I think that made my AI work was 2 lines in the end
Ok well damn, I had problems publishing all the maps. It is late, but I think I have the problem either narrowed down in a bug in modifying map info or naming a map with part of a Blizzard maps name. Those are my only 2 guesses. But hopefully I can get them all up real soon. For now, you can play Metalopolis on my test map called "FAITest". It is a test map so don't expect it to exist forever when the real maps are published, but for all purposes this will be what I publish for Metalopolis when I can figure out what is going wrong.
Thanks for pumping out 1.0.1 so quickly is 1.0.1 for this patch?
so far noticed that the medium AI seems to be behaving like the default blizzard AI in that it'd just be retarded at macro and float hundreds of minerals unlike the 1.1.3 version. also when i did a drop on the AI it didn't build a single probe after I harassed the mineral line.
I tried to import the script into the map and all seems fine, but how do I choose the difficulty level? There are no options to do is without using the tool? I went into map properties and it has under the AI setting just default and computer. Any advice?
Or where can I find the maps published for the SEA server by someone?
On January 14 2011 13:04 ballasdontcry wrote: Thanks for pumping out 1.0.1 so quickly is 1.0.1 for this patch?
so far noticed that the medium AI seems to be behaving like the default blizzard AI in that it'd just be retarded at macro and float hundreds of minerals unlike the 1.1.3 version. also when i did a drop on the AI it didn't build a single probe after I harassed the mineral line.
I just now updated them to a working state. You know you are seeing the current version when you see v1.0.1 on the screen when the game starts which should be right now Sorry for all the issues with the patch, I will try and get it ready ahead of time next time. Too bad they don't patch on the weekends when I don't have to sleep.
I tried to import the script into the map and all seems fine, but how do I choose the difficulty level? There are no options to do is without using the tool? I went into map properties and it has under the AI setting just default and computer. Any advice?
Or where can I find the maps published for the SEA server by someone?
Thanks in advance.
Um for one I need to upload the new script! Thanks for reminding me.
Edit: Ok the script is uploaded at http://code.google.com/p/fyn-ai/downloads/list for anyone that is going to publish another map or publish the AI on other server other than NA. The file is called FynAI_v1.0.1.zip.
The current version will not be on the SEA versions quite yet I would not believe because I just posted the the new code a few minutes ago. n3wb13 published the last set on SEA and I would prefer he published the maps that I initially published (feel free to publish other maps) just because I want the maps to retain the name "Metalopolis (Fyn AI)" etc so its standardized. Of course if he quits publishing them then someone else will have to under a different name. I just say this so there isn't clutter in the map pool.
Now if you can't wait for him, go to the code.google.com site i listed and get the v1.0.1 zip file. Then search google for SC2ALLin1 which is a tool to run stand alone. You have to the either find the maps somewhere to download or you can go in to the Starcraft 2 editor, open up the map you want (you have to log in), then save it to disk. After you install SC2ALLin1, copy the trigger libs folder in the zip I have on my site to the C:\Program Files\StarCraft II\SC2ALLin1\ai folder and run SC2ALLin1. The first time you I think you have to go in to the options tab and select the folder where it is, but maybe not it might know. Select you map from the maps tab. Click the game tab. Turn off players 3 and 4, and set player 2's difficulty. Last thing is check the Use Custom A.I. check box or it will not work. Make sure you don't have Starcraft already running, then hit the Launch button in SC2ALLin1 and it will open up SC2 for you straight in to the game like a replay does when you play old replays. I think when you are done with the game you have to remember to close starcraft again before starting the next game with SC2ALLin1.
Or... you could wait for n3wb13
Or... as someone mentioned previous, SEA users can log on to NA to play them.
On January 14 2011 20:20 n3wb13 wrote: yo~! I've republish the maps up in SEA.
Gonna outstation this weekend... hope everything is fine...
Nice thanks.
This weekend I will start working on an intro dialog that will let you adjust APM and vision settings so people who thought it was too easy or too hard can make adjustments.
I am going to publish a guide tonight on how to do it, which is a really simple task. Maybe you could, or if you don't want the responsibility of maintaining it, maybe you will luck out and one of your EU cohorts will help you out and publish.
On January 15 2011 02:42 fyndor wrote: I am going to publish a guide tonight on how to do it, which is a really simple task. Maybe you could, or if you don't want the responsibility of maintaining it, maybe you will luck out and one of your EU cohorts will help you out and publish.
i could very well do it if the guide is good enough
On January 15 2011 02:42 fyndor wrote: I am going to publish a guide tonight on how to do it, which is a really simple task. Maybe you could, or if you don't want the responsibility of maintaining it, maybe you will luck out and one of your EU cohorts will help you out and publish.
i could very well do it if the guide is good enough
It's super simple. You could publish a map in maybe 60 seconds. I will post here later tonight when I have a guide available somewhere. It will be a screenshot style guide.
Perhaps this should be it's own topic, but as far as the Blizzard AI. I don't feel like they mineral hack even at very hard. Perhaps they are but how do you know this? I actually feel as tho they macro up and then pump units, and if you watch their money they sit on a lot of minerals for a lot of the game, maybe it's BECAUSE they're getting more resources than they should but this seems like a stupid way for blizzard to do it because the computer could just have a perfect build if it wants to, and literally never miss a probe..
I'm curious what people think about the AI from differ ladder levels, because I'm only bronze, but I can beat the very hard AI with any matchup doing a straight up gameplay, no need to cheese or heavily harass...
EDIT: I"m very intersted to test out this AI when I get home from work though!
On January 15 2011 05:05 thurst0n wrote: Perhaps this should be it's own topic, but as far as the Blizzard AI. I don't feel like they mineral hack even at very hard. Perhaps they are but how do you know this? I actually feel as tho they macro up and then pump units, and if you watch their money they sit on a lot of minerals for a lot of the game, maybe it's BECAUSE they're getting more resources than they should but this seems like a stupid way for blizzard to do it because the computer could just have a perfect build if it wants to, and literally never miss a probe..
I think you are referring to a misconception I had earlier. You are right, Very Hard does not mineral hack on Blizzard AI, but Insane does at a 1.5 rate. My AI currently mineral hacks at Hard and greater. It does it over a 10 minute ramp up unless you are playing insane and then it is always 2.0 rate.
I'm curious what people think about the AI from differ ladder levels, because I'm only bronze, but I can beat the very hard AI with any matchup doing a straight up gameplay, no need to cheese or heavily harass...
EDIT: I"m very intersted to test out this AI when I get home from work though!
I thought you were talking about Fyn AI for a sec until I read the edit and was feeling a bit fail seeing as how Medium difficulty (which is what I play) gives me trouble and I am silver :X I personally am interested which ladder levels people consider the different difficulties of Fyn AI. And yea the Blizzard AI is not all that hard. It has been a bit since I have played it, but I would bet Blizzard Very Hard AI is worse than Fyn AI on Very Easy.
On January 15 2011 07:05 bahl sofs tiil wrote: Good on you, with the update. I like this and I was a little bummed when it stopped working. Nice to see you are still staying with it.
Yea I will be sticking with it. I am just getting started! In fact, I just added a wiki to the website where I host the source which I will use for various reasons. For one, I posted an expected roadmap to what my intended development priorities will be (though they are subject to change). I may not be super speedy at times since do have a full time job developing software and I would like to spend some of my time actually playing the game. This will most likely be a weekend warrior type project for me, but regardless before I am done I will at least touch on if not successfully finish everything in that roadmap.
FynAI most definately better simulates the playstyle of a human player, no doubt about it. FynAI will actually scout and expand whereas BlizzardAI tends to blindly go 1 base all-in.
Still, even on medium setting I have to say that it is not as good as the opponents I face in the ladder (high gold to high platinum). So I guess FynAI situates itself somewhere in silver.
On January 15 2011 07:24 REM.ca wrote: Still, even on medium setting I have to say that it is not as good as the opponents I face in the ladder (high gold to high platinum). So I guess FynAI situates itself somewhere in silver.
That is what I was thinking, but I couldn't really tell since I am Silver myself. Thanks for the heads up.
I will do my best to make it so the AI can start competing in the Gold and above arenas without cheating. It is going to take better scouting and reactions to what it sees for one. After I lay some initial groundwork I will start reaching out to you guys, probably in some structured way, getting feedback about what needs to be done to improve the AI. I am kind of hoping as a community we can mold this in to a really good tool for practicing.
On January 15 2011 07:44 slith wrote: Pity the Zerg AI loves to waste Mutas instead of harassing :/
Yea doesn't it just use them as part of its army? That rarely has worked for me when I play Zerg unless I harass first with them so my opponent is already behind. When I rework zerg I will at least try to implement some basic muta harass (mineral harass mostly). Following that at some point I will of course have to make it so it will do overall base harassment like taking out key pylons and tech structures and prevent turret building etc. That is a little more advanced though as far as the coding involved so first things first. Good point though, this is certainly a problem.
Yea, if you did it right, when a game loads up on the bottom left side of the screen there will text saying that this is Fyn AI written by me, giving credit to Green Tea and its creator, you will see something about difficulty level your are playing, and if you are playing Very Easy or Easy you see some greek mythological name or something like that which is actually a way to tell what build they are using it.
I will post a screenshot when I get home from work in a bit, and right after that I will be developing and posting a guide on how to publish to the wiki for this AI which should be done by about 2 hrs from now give or take. So if you can't figure it out just wait a few hours and I can get you a guide that should do the trick. There are a few places where you can easily do something that would cause it not to post right and I believe I have found a bug (if not a design decision) in the editor that prevented me from publishing patch 1.2 maps for a bit. That will all be covered in my guide.
Ok I have published a Publishing Guide on the Fyn AI Wiki. This guide should be able to show you all you need to know to publish maps with Fyn AI. If you have any input about how to make the guide clearer or any issues publishing let me know. Please read the Publishing Guidelines in the guide about how you should name maps so we don't get a bunch of clutter.
On January 15 2011 18:09 Black Gun wrote: nvm, got it.
published the first map on EU. i will add the popular ladder maps now.
these ais can be found on EU by searching for "fyn"
I already uploaded Meto at least, was gonna do the rest but had to leave!
Anyway, really great AI. I'm glad I finally get to play it on EU , thanks a lot!
would u mind deleting it again so ppl dont get confused because there are 2 versions of metalopolis (fyn ai) on EU?
i now uploaded it for all 1v1 ladder maps plus crossfire. but im missing shakuras plateau atm as it is removed from the ladder and i cant find it nowhere in the editor->open menu.
anyway, they are up on EU now. u can find them by searching for "fyn".
on this url http://www.mediafire.com/?ca5dxk7qykoyywz u can find a zip folder with all these maps. if someone wants to publish them on other servers he can use these and only has to publish them.
btw could any1 send me a blank version of shakuras plateau plz so that i can add it? its kind of the best map atm, so its a pity if its missing.
edit: @fyndor: in the future, plz pm me whenever u update this AI so i can keep them updated.
On January 15 2011 18:56 Black Gun wrote: im missing shakuras plateau atm as it is removed from the ladder and i cant find it nowhere in the editor->open menu.
Try the Shakuras in here, it should work. I had to use a version that was distributed from the start because it was already not showing up in the list. Even though its in the custom game list when you go in to the actual game....
edit: @fyndor: in the future, plz pm me whenever u update this AI so i can keep them updated.
On January 15 2011 18:09 Black Gun wrote: nvm, got it.
published the first map on EU. i will add the popular ladder maps now.
these ais can be found on EU by searching for "fyn"
I already uploaded Meto at least, was gonna do the rest but had to leave!
Anyway, really great AI. I'm glad I finally get to play it on EU , thanks a lot!
would u mind deleting it again so ppl dont get confused because there are 2 versions of metalopolis (fyn ai) on EU?
i now uploaded it for all 1v1 ladder maps plus crossfire. but im missing shakuras plateau atm as it is removed from the ladder and i cant find it nowhere in the editor->open menu.
anyway, they are up on EU now. u can find them by searching for "fyn".
on this url http://www.mediafire.com/?jz3y089buvnmxfp u can find a zip folder with all these maps. if someone wants to publish them on other servers he can use these and only has to publish them.
btw could any1 send me a blank version of shakuras plateau plz so that i can add it? its kind of the best map atm, so its a pity if its missing.
edit: @fyndor: in the future, plz pm me whenever u update this AI so i can keep them updated.
ill try to defeat the AI in a macro game on shakuras plateau now, starting with medium and then working my way up to insane.
although i doubt that its possible to beat the insane ai in a straight up macro game, having the increased income right from the start instead of ramping up over the first 10 minutes is just too much of an andvantage.
On January 15 2011 20:16 Black Gun wrote: although i doubt that its possible to beat the insane ai in a straight up macro game, having the increased income right from the start instead of ramping up over the first 10 minutes is just too much of an andvantage.
Any way to tell the AI to do a custom build and when finished select a random commander? Or getting rid of the resource bonus at higher levels of difficulty?
On January 15 2011 21:02 slith wrote: Any way to tell the AI to do a custom build and when finished select a random commander? Or getting rid of the resource bonus at higher levels of difficulty?
Custom builds are a possibility down the line, but I don't want to do something with supply based triggers like many people think of builds in. Builds based on supply numbers can be tripped up easily. If you look at the sc2builds and yabot thing that is really meant to give instructions to a player in game on how to do a build, not instruct an AI, which I don't think would work too well. I am trying to think of how I could design it so that the average starcraft player could understand it and design their own build for the AI. I would probably have to have an offline graphical build designer which would let you export something in to starcraft. It could be done but that is not going to be high on my priorities for a while. What I can say is that in the near future I am going to be redesigning all the builds that this mod does, and I will likely increase the number of builds it has since I want it to take in to account race and map at the very least when choosing builds. When I do that I will be reaching out to you guys to figure out what builds need to be added.
Unless I have to patch a bug first instead, the next release will have a dialog that pops up at the beginning of a game and lets you customize difficulty settings. I could at that point let you choose which build your opponent will start with, but it would have to be one of the builds built in to the AI. Speaking of the dialog though that gets to your second question. Right now the only thing that makes Hard, Very Hard, and Insane different than Medium is the resource cheat. Without it you have medium difficulty. When I add the dialog you could set it up so you add back in the vision hack (which no difficulty uses currently) but take out the resource hack. You should get something that is more realistic to what a player could pull off and would be harder than medium. Beyond that there is not much to be done without just making the base AI smarter. Medium is not a dumbed down AI compared to the Hard and above, but rather it is the best that the AI can do without cheating. Ultimately I need to make the base AI (aka what you see with Medium right now) smarter to the point where you guys are telling me medium is getting too hard to be called medium. At that point I could add in "imperfections" to lower it back again and call that medium, and rename the base AI hard or very hard or w/e. If you find medium too easy it is possible that the vision hack is what you are looking for, or it may not. You may find really the AI just needs to be a little smarter about choosing builds, or scouting you and reacting better. That is all in the works.
On January 16 2011 23:21 REM.ca wrote: I'd like to come back and nuance my comment that "The Medium Ai is probably at the silver level."
When I first tried, I played TvP only (about 9 games.....3 per AI difficulty from VE to Medium).
I tried again playing TvZ and found the Medium AI much more difficult than in TvP. It actually beat me a few times.
I may be wrong, but what you might be noticing is the better scouting Z does because it overlords become scouts when it gets speed and it tends to get better information than P and T. P is not using obs or hallucinated phoenix and T is not using scans because the AI I used as a starting point used a vision hack and never developed scouting code. I have tested adding observers to the scouting group, but scouting is handled by built in Blizzard scouting functionality and that code is more worried about finding new expansions than keeping tabs on enemy army composition. So I am going to have to write custom scouting code for observers and scans and the like.
Also, I am cannot remember if Protoss ever makes forward pylons, but it needs to do that more for sure. When balancing the races, Blizzard certainly considers the ability of Protoss to warp in close to the enemy and if the AI is not doing that they are effectively gimping themselves. Show me the Protoss pro that warps everything in at their base and runs it across the map
You may have also noticed Z loves to use burrow micro which noone at my level really micros with burrow, so that tends to throw me off a bit. You are probably going to find you need a raven at some point in the game with Z because I would think otherwise you are going to have to waste a lot of scans.
On January 17 2011 08:49 asin wrote: Thanks Fyndor,
I downloaded the maps with the script imported but still I can't seem to choose the AI difficulty level?
you have to search for them on battlenet, search for fyn. then select the map and open a custom game on it. then u can select the difficulty as usual. if u want to use the maps offline, download the mappack i uploaded some posts above, open the map in the editor. then go to file -> basic settings/settings (dunno exactly, im using the german version.) alternatively press ctrl + shift + alt + P. in the popup menu, select the field "test map". there on u can select the difficulty to use while testing files. then u just have to press ctrl + F9 to play the map.
On January 17 2011 08:49 asin wrote: Thanks Fyndor,
I downloaded the maps with the script imported but still I can't seem to choose the AI difficulty level?
If I remember correctly you play on SEA. I believe our good friend n3wb13 as already posted them to SEA. Go in to a custom game and search for "fyn" and you should find the maps. You are more than welcome to play them offline, but it would be much easier to play them on battle.net.
On January 17 2011 09:23 dunc wrote: I don't know what people are saying about it not being better in higher leagues, this AI definitely plays better than most Platinum players on Medium.
Having lots of fun with it so far!
actually no, im a platinum player myself and pretty easily defeat the medium AI most of the time.
On January 17 2011 09:23 dunc wrote: I don't know what people are saying about it not being better in higher leagues, this AI definitely plays better than most Platinum players on Medium.
Having lots of fun with it so far!
actually no, im a platinum player myself and pretty easily defeat the medium AI most of the time.
Different opinions. I have no problem beating the medium AI, but it plays better than most people from Platinum league I meet at least.
On January 17 2011 09:23 dunc wrote: I don't know what people are saying about it not being better in higher leagues, this AI definitely plays better than most Platinum players on Medium.
Having lots of fun with it so far!
actually no, im a platinum player myself and pretty easily defeat the medium AI most of the time.
Different opinions. I have no problem beating the medium AI, but it plays better than most people from Platinum league I meet at least.
well, it depends. it has good macro and expanding patterns, but its unit composition and micro sometimes are lacking. in particular, it is very bad at dealing with unknown situations. if i mass up a terran deathball of tank/thor/viking with some marines or hellions mixed in, it just doesnt know how to deal with it. it doesnt abuse mobility issues against a slowpush.
but obviously stuff like that is very hard to implement in an AI, no criticism for that. the sad fact is that research on "true" artificial intelligence is stagnating since many years.
This AI have a few commander which randomly executes different strategies most the time. So it force me to scout the AI most the time while macroing/defending harass so that i dun get caught with some annoying strategies (2port cloak banshee/fast bc etc). It also a good warm up before i went for laddering so i just dun waste my 1st few ladder game trying to get into mood .
In short mostly i treat this AI as "practice partner" (pity me... fren list is 1-hand-counted...) which at the same time i give myself some objectives like 10mins 3 base/ 200/200 in 13 mins / no-attack-consume-map, something like what funday monday does I dun actually play this AI to win, but to train myself to become a better player... (lol... quote day[9])
Thanks this is a great practice tool. Am currently a Gold player and have no troubles beating the medium AI, but the hard AI give me a hard time, haven't been able to win against it yet (macro style play with zerg)
mind you only starting playing it yesterday
whats everyones experience with the hard ai? is it equivalent to platinum players u think?
well i just tried the FYN ai on blistering sands on very hard. and it started of getting 5 minerals like me. but it was raising all the time. a few min into the game he was getting 6 then 7 and at the end of our 11 min long game it was getting 10 mineral while i was getting 5. ive never had trouble beating the insane ai but this one was really hard to beat even on very hard in a macro game. however without its mineral cheats it wouldnt been any better than a silver player. the ai unit loses was almost twice as high as mine so was its income on the same amount of workers. but yea its micro is definantly better than the blizzard ai. and its probably better practice for bronze-silver players. but be aware that it starts getting twice the income of you at 10 minuttes. if you dont believe me check your replays
On January 19 2011 08:42 khazgore wrote: well i just tried the FYN ai on blistering sands on very hard. and it started of getting 5 minerals like me. but it was raising all the time. a few min into the game he was getting 6 then 7 and at the end of our 11 min long game it was getting 10 mineral while i was getting 5. ive never had trouble beating the insane ai but this one was really hard to beat even on very hard in a macro game. however without its mineral cheats it wouldnt been any better than a silver player. the ai unit loses was almost twice as high as mine so was its income on the same amount of workers. but yea its micro is definantly better than the blizzard ai. and its probably better practice for bronze-silver players. but be aware that it starts getting twice the income of you at 10 minuttes. if you dont believe me check your replays
I believe you, that is exactly what happens The Insane AI starts at a rate of 2, hard and very hard ramp up to 2 over a 10 minute period. I would not think a bronze player would have any business playing even hard AI, that should be way beyond them Hard on Fyn has no correlation to Hard on Blizz AI. In fact I would bet you would probably find Very Easy Fyn AI is probably better than Very Hard Blizz AI (thats not saying much). There are plenty of things it does that can be improved for sure. In some situations it gives away units completely for free. Sometimes it does stuff that is just plain silly. For instance yesterday I did a gas steal against toss AI and as I expected it took its zealot and attacked the gas, but then because the build said it needed a second gas at that time it added the second gas...at the expansion that had no nexus at it. Not sure if it ever made use of it until it expanded but it was totally retarded. It will also waste units running in to cannons sometimes as well and for other reasons I can't think of at the moment. Luckily I can see ways of fixing a lot of the dumb things it does to waste units. First though I am going to focus redoing the openers so that they are more standardized to what most people do. After that I am going to focus on improving scouting so it has the right info to make good decisions. That is probably the stage in development where I will see if I can knock out some of the worst blunders it makes.
Also I forgot to mention, much of what makes it challenging for a player like me is that it expands more than Blizzards AI, so eventually it will have good enough macro to counter what you make (in my experience) as long as it survives past its initial decision of a randomly chosen build order. One of the things it could really use is the ability to make early game build decisions based on what your opponent does. I would image that would improve its play dramatically. But as I said before, first things first, I am going to try to make it play more "standard" using starting build orders based on map, race matchup, and possibly starting location. Then from there I think I will let it be more open ended like it is now and have a wide variety of late game build that work with the starting builds and maybe based on what the enemy army makeup is.
Terran Plat player here (though with my losing streak from last week, maybe I'm just a high gold...bleh)
This AI is a great tool. I remember reading somewhere that the difference between a platinum player and a diamond player is that the diamond player knows the timing of when to drop the third base. While the unit comps of this AI might be lacking, it is a beast at macroing. It forces you to learn when to drop your third AND fourth, or you WILL lose even with the terrible unit comps. I have attempted medium at first and have lost every time(only twice, both times barely.)
If your biggest problem is macroing and timing your third base, this is a great AI tool for you. If you want to learn how to micro, play a map geared towards microing. Of course, it's always better to play in 1v1s against capable opponents, but this is a great tool. This AI plays on the level of a very high platinum player not doing any ridiculous cheese.
Fyndor, you should be paid money for this AI. Excellent, excellent work!
This might be a stupid question, but is there any risk of getting banned using this or the regular 'green tea' ai? The green tea website ( http://sc2.nibbits.com/assets/green-tea-ai/ ) talks about importing things into SCII and I thought anything that messed with game files could get you banned.
On January 19 2011 16:01 Bizarro252 wrote: This might be a stupid question, but is there any risk of getting banned using this or the regular 'green tea' ai? The green tea website ( http://sc2.nibbits.com/assets/green-tea-ai/ ) talks about importing things into SCII and I thought anything that messed with game files could get you banned.
Just want to check before installing anything
Thanks
First off you are being a little bit too paranoid, but I guess normally that is healthy. You are safe to use this mod/map. For one this, is published to Battle.net (atleast thats how i use it, it can be run standalone). If you think about it, since the only way to get stuff on to Battle.net is through Blizzards own Map Editor tool, they have the ability to restrict w/e they want. In fact there are certain reserved files that are not allowed to be overwritten and if I try to import a new version of them it will not let me. You may be a bit confused with what you do with this AI. It is an AI that you can add to a map to replace the original AI. But to do this with Blizzard maps or any map that I didn't create myself, what I do is get a copy of the map then republish the under a different name with the AI added. I get these copies from the Map Editor itself, not through some hack, and I can only download maps that were marked by the publisher as "unlocked". All the Blizzard maps are unlocked so I can copy them. These modified maps are only used in custom games. It has no effect on ladder or any other maps is the game. The only thing affected are the custom maps I publish to Battle.net. All changes are only made to COPIES of maps, not the original maps, not any game files. Trust me you are safe.
Granted, we very recently saw where Blizzard took action against a map called World of Starcraft that was being developed (). This is the only issue I personally have heard of so far, and though Blizzard has not elaborated yet as far as I know as to what the problem is, the issue they are likely taking to that "map" is that they probably consider the name a trademark infringement. There is speculation that they may not want it published, because their next MMO may be starcraft related or other similar conspiracy theories, and their license allows them free will to do whatever they please with Battle.net. So it is technically possible that if for whatever reason Blizzard decided they didn't like people using custom AIs in custom games that they could remove the maps. That is likely the extent to what would happen though if they should choose eliminate custom AIs beyond patching the editor to not allow stuff like that to be published, but at that point a lot of custom maps will be gone and I don't see that happening. Custom AI in custom maps poses no threat to Blizzard or the community and I am pretty sure they know it and agree. If anything I can see them taking some of the code in good AIs and incorporating it in to their stock AI. At least that is what I would do if I were them
On January 19 2011 12:28 boredrex wrote:While the unit comps of this AI might be lacking, it is a beast at macroing.
Hopefully in the next few months I will have totally reworked starting builds and overall unit comp choices, though part of the changes that will need to be made will have to be improved scouting so that it can make good decision about unit comp and those changes likely come after the build redesign.
It forces you to learn when to drop your third AND fourth, or you WILL lose even with the terrible unit comps.
At this point that is what I am trying to get from the AI as well. I am not skilled enough to know whether these are reasonable expansion timings it uses yet, but I know it works against me : / So far the only problems I can certainly see for sure with the way it expands is its choice of expansion location is sometimes not the best and it apparently doesn't know that it should cancel building an expo if is about to die before completing. Do players normally try and take the shared 3rd on Metalopolis?
Fyndor, you should be paid money for this AI. Excellent, excellent work!
My current contributions to this codebase are actually minimal thus far, but soon I hope to earn that statement.
Speaking of the future, after reading this article about a Starcraft 1 AI competition and what the winners did to make a good AI, my mind is now blown and racing with ideas. I don't know what I can accomplish as a single AI hobbiest vs the small team of grad students and PhD professors that wrote that AI, but I am certainly inspired to try and reproduce some of the concepts they used to build a robust AI. Doubly so after reading that the specs for the machines running the bot software for the competition were not super computers, but rather mediocre laptops. + Show Spoiler +
Dell XPS laptop Windows XP Service Pack 3 Intel Core 2x @ 2.0 GHz 2G RAM NVidea GeForce Go 7950 GTX, 512 RAM (No CUDA)
In other words, if a strong Starcraft 1 AI can be run on a crappy laptop, then it is not out of the question to make an equally good Starcraft 2 AI run on standard end user hardware. Projects like this always worry me, because you never quite know when you get started whether you can pull it off because of the complicated nature of the problem. Now I know for sure a small team was able to build a fairly robust AI for a similar game within a 1 year time frame. I have no illusion I can match what they did with the time and resources available to me, but I do think I can do better than what I shown you so far and implementing some of the concepts from that team's AI would certainly be a great improvement.
On January 19 2011 16:01 Bizarro252 wrote: This might be a stupid question, but is there any risk of getting banned using this or the regular 'green tea' ai? The green tea website ( http://sc2.nibbits.com/assets/green-tea-ai/ ) talks about importing things into SCII and I thought anything that messed with game files could get you banned.
Just want to check before installing anything
Thanks
First off you are being a little bit too paranoid, but I guess normally that is healthy. You are safe to use this mod/map. For one this, is published to Battle.net (atleast thats how i use it, it can be run standalone). If you think about it, since the only way to get stuff on to Battle.net is through Blizzards own Map Editor tool, they have the ability to restrict w/e they want. In fact there are certain reserved files that are not allowed to be overwritten and if I try to import a new version of them it will not let me. You may be a bit confused with what you do with this AI. It is an AI that you can add to a map to replace the original AI. But to do this with Blizzard maps or any map that I didn't create myself, what I do is get a copy of the map then republish the under a different name with the AI added. I get these copies from the Map Editor itself, not through some hack, and I can only download maps that were marked by the publisher as "unlocked". All the Blizzard maps are unlocked so I can copy them. These modified maps are only used in custom games. It has no effect on ladder or any other maps is the game. The only thing affected are the custom maps I publish to Battle.net. All changes are only made to COPIES of maps, not the original maps, not any game files. Trust me you are safe.
Granted, we very recently saw where Blizzard took action against a map called World of Starcraft that was being developed (http://www.youtube.com/watch?v=roIOpHq0vUc). This is the only issue I personally have heard of so far, and though Blizzard has not elaborated yet as far as I know as to what the problem is, the issue they are likely taking to that "map" is that they probably consider the name a trademark infringement. There is speculation that they may not want it published, because their next MMO may be starcraft related or other similar conspiracy theories, and their license allows them free will to do whatever they please with Battle.net. So it is technically possible that if for whatever reason Blizzard decided they didn't like people using custom AIs in custom games that they could remove the maps. That is likely the extent to what would happen though if they should choose eliminate custom AIs beyond patching the editor to not allow stuff like that to be published, but at that point a lot of custom maps will be gone and I don't see that happening. Custom AI in custom maps poses no threat to Blizzard or the community and I am pretty sure they know it and agree. If anything I can see them taking some of the code in good AIs and incorporating it in to their stock AI. At least that is what I would do if I were them
Thanks for the informative post, can't wait to try it. One of the reasons I am so paranoid is I use a G15 keyboard, I DO NOT macro with it or use it in anyway in sc other than as a keyboard, but I was always worried/doing research on if just simply having the keyboard and the various programs that go along with it would be a threat of getting banned. Luckily this is not the case but I was worried about it for a while back when they were banning a lot
There's some small things I have started to notice about the AI. First off, the zerg ai in a zvz is tough as heck. The multitasking makes it play almost as good as a top level zerg (Early pressure with expanding). I have only beat it once in a straight up game (without my 10 pool tactice I use but with a 14 pool). Normally I have no problem against the other two races unless I try to be greedy and lose a ton of stuff at the start.
Anyway, the biggest problem I notice is it seems like the ai can still see everything going on. As in it doesn't need to scout. I've noticed every single time that i send my drone scout to the enemy base they are sending out a worker to attack my scout before it is in view. This also made me notice other little things happening (Such as pulling back before it reaches view of my army directly after I reinforce it). So it seems like the computer can still "see" some things a normal player wouldn't.
Thank you OP for reviving this great AI. I'm having a blast playing against the Medium AI and polishing up the weak areas of my macro. Keep up the good work
On January 20 2011 14:54 Atlas_550 wrote: There's some small things I have started to notice about the AI. First off, the zerg ai in a zvz is tough as heck. The multitasking makes it play almost as good as a top level zerg (Early pressure with expanding). I have only beat it once in a straight up game (without my 10 pool tactice I use but with a 14 pool). Normally I have no problem against the other two races unless I try to be greedy and lose a ton of stuff at the start.
Anyway, the biggest problem I notice is it seems like the ai can still see everything going on. As in it doesn't need to scout. I've noticed every single time that i send my drone scout to the enemy base they are sending out a worker to attack my scout before it is in view. This also made me notice other little things happening (Such as pulling back before it reaches view of my army directly after I reinforce it). So it seems like the computer can still "see" some things a normal player wouldn't.
I was surprised that you would say the ZvZ was hard so I tried it out. There are probably other builds the computer does that I have yet to face, but the first build I played against - 1base speedlings into muta + expo - got dominated by a pretty standard 1 base roach build.
EDIT: well I played the other 2 (I assume there's only 3), fast expand into speedling double expand got owned just as bad as the first build, but the roach into somewhat fast expand build was pretty good, except I still won pretty easily.
I think the biggest reason why the AI would loose is because it wouldn't threat assess me in either economy or military power, and would not try to balance his own military or economy to match mine. He'd have like 18 lings, then just POWER drones like mad - when my roaches get there he has a great economy but is completely defenseless.
Also I think I noticed the computer reacting to things he couldn't see as well.
On January 21 2011 06:08 Xapti wrote: I was surprised that you would say the ZvZ was hard so I tried it out. There are probably other builds the computer does that I have yet to face, but the first build I played against - 1base speedlings into muta + expo - got dominated by a pretty standard 1 base roach build.
EDIT: well I played the other 2 (I assume there's only 3), fast expand into speedling double expand got owned just as bad as the first build, but the roach into somewhat fast expand build was pretty good, except I still won pretty easily.
Here are the builds as annotated in the code: 1 Base Mutalisk, 6 pool, Overpool, Baneling, Roach, 2 Base Multalisk, Hydralisk
I cannot guarantee that those are actually the builds that are in there because I have seen instances in the code where a comment was not changed after a build was modified, but those are likely correct.
I think the biggest reason why the AI would loose is because it wouldn't threat assess me in either economy or military power, and would not try to balance his own military or economy to match mine. He'd have like 18 lings, then just POWER drones like mad - when my roaches get there he has a great economy but is completely defenseless.
I am not sure if your economy comes in to account at all in the AI as it stands currently. I would have to check, but it is it possible that it might take in to account enemy expansion timing, but I have a feeling it likely doesn't. I have heard from I believe Day9 that the rule is that there are two options when you see a new enemy expansion go down: make your own or push, not both. I believe this statement was in regards to early expansions. Beyond that though, do good players pay attention to stuff like mineral saturation, or is it basically about making sure you either keep up with expansion counts or attack when they expand? This kind of stuff is not one of my strengths as a player yet and could use some advice so I can make the appropriate changes to the AI.
It does attempt to do threat assessments though it certainly needs to be improved. As I have said a bunch, the scouting it does is stock Blizzard code and is not cutting it. That factors in a lot when you are talking about threat assessment obviously. It is doing a threat assessment constantly the entire game and at any point it may decide to attack. Depending on how late in the game and what race the AI is, I can generally say it asks the question "IF MYARMY > YOURARMY + THRESHOLD THEN ATTACK". The threshold I believe is probably there to make sure you account for how fast the AI can reinforce after army losses based on race and game time. Two things though. One, if you are good about not showing the scout what you have, or two, if the scout hasn't checked in while, the information it is using to make this assessment is "outdated" and therefor flawed. My thoughts on this are that the biggest improvement will come from writing my own scouting code so it does it better and more often. Beyond that, I think it may be good to come up with an estimation algorithm that will estimate how many of each unit type you probably have. This would be based on the rate you have been getting them in the past and maybe based on what is possible from the number of unit producing structures it knows you have and how fast they can make units. You could totally stop making a unit, make a different unit type than it has seen you make yet, or really ramp up the production on a certain unit all of a sudden and the rates it used to guess your army composition would be flawed, but as it stands right now, it always considers your army size to be the exact same as the last time it saw it. Right now, if you deny its scout from seeing most of your army for any significant period of time, it would assume your army is the same size as it was earlier, but unless you are capped at 200/200 this is likely a bad assumption. Also I have always felt that it many times did not turn back and retreat when it should. The code for this should be more simple since many times the decision is as simple as "is the army in front of me bigger than mine" and it may just need a bit of tuning. Looking at what it uses currently to determine whether to retreat, it probably is just oversimplifying that decision.
Atlas_550, I wrote a response your comment about it still vision hacking at home last night and apparently forgot to hit send. Because I am lazy, I am going to wait until I get home and then go ahead and hit that send button. Don't worry, I am not ignoring you
Ok, I LOVE this AI! It's got the only thing that blizzard ai's are missing entirely: Good eco. I played a PvP with it, it had 2 fully saturated bases when I had one + beginning expansion. I still beat it with mass stalkers + proxy pylon though It was doing an odd stalker+collosus build, very well played. (BTW, is it supposed to say stuff? It said "Nice game, don't think we've ever played before") Second game was PvT (I was P). I fast expanded (lol) at like 3 minutes, did one gate expand into void rays, pressured him with a stalker and 3 zealots early on. Got steamrolled by thor+marine+ghost. Third game was PvZ (I was P) The computer went with a 17 hatchery 18 pool (???) I did a 10 gate zealot pressure. He barely held me off with constant zerglings+queens, doesn't rebuild ANY drones AT ALL, and masses zerglings to kill rocks at the gold (LT) He then attacks me with 20 zerglings vs my 8 zealots, gets obliterated and retreats, and I go kill him. In summery, the AI is very good with building an economy, but will not rebuild workers AT ALL (at least on medium). The terran is sooooo powerful with its mules. Good work!
On January 21 2011 08:49 TheAwesomeTemplar wrote:It was doing an odd stalker+collosus build, very well played.
I personally find a number of builds in this AI a bit odd and not very standard. I wouldn't mind it doing odd builds from time to time to spice things up and be less predictable, but I would think it doing standard starters 90% of the time would probably be appreciated as far as practicing goes and I think that is the route I am going to go pretty soon here. I may keep some of the builds I find more strange in, but just rarely used, or I may take them out. When I make these types of changes I will consult with you guys first so the decision I make is what people want.
BTW, is it supposed to say stuff? It said "Nice game, don't think we've ever played before"
Yep. It starts doing that later in the game. It actually will BM you with stuff like offensive GGs when it thinks you should give up. This was put in by the original author of the AI. I don't want to ruin the fun, so don't ask for details because I won't give them, but I will say I have plans for some entertaining modifications to this for my next update since its pretty effortless to do
Third game was PvZ (I was P) The computer went with a 17 hatchery 18 pool (???) I did a 10 gate zealot pressure. He barely held me off with constant zerglings+queens, doesn't rebuild ANY drones AT ALL, and masses zerglings to kill rocks at the gold (LT) He then attacks me with 20 zerglings vs my 8 zealots, gets obliterated and retreats, and I go kill him. In summery, the AI is very good with building an economy, but will not rebuild workers AT ALL (at least on medium).
I will test this and see if I can see what it was thinking and get back to you. It may be a priority issue. I know for sure that all the builds I have looked at have a "set in stone" number of workers it will make in the beginning, and the way its coded, in the beginning at least it would remake those drones. But once it passes a certain point in its build, it will then run this generic routine that I have not looked into detail at, but appears to make sure it makes sure it builds workers to match how many bases and gases it has up to a max number of workers. The max for insane is 60 because it cheats and needs less workers, and other difficulties is 70. I have heard some people say always be build workers (Zerg being the exception). Is this a "rule" people tell to noobs so they don't bother asking themselves when to quit making workers and make bad assumptions, or is there really some kind max you don't want to go over so you don't have too much supply in workers? I guess 60 would be fairly saturated. Would it ever be logical late game to have say 4 saturated bases running or is that just beyond what is necessary or detrimental because of wasted supply?
On January 21 2011 06:08 Xapti wrote: I was surprised that you would say the ZvZ was hard so I tried it out. There are probably other builds the computer does that I have yet to face, but the first build I played against - 1base speedlings into muta + expo - got dominated by a pretty standard 1 base roach build.
EDIT: well I played the other 2 (I assume there's only 3), fast expand into speedling double expand got owned just as bad as the first build, but the roach into somewhat fast expand build was pretty good, except I still won pretty easily.
I think the biggest reason why the AI would loose is because it wouldn't threat assess me in either economy or military power, and would not try to balance his own military or economy to match mine. He'd have like 18 lings, then just POWER drones like mad - when my roaches get there he has a great economy but is completely defenseless.
Also I think I noticed the computer reacting to things he couldn't see as well.
Well I think my biggest problem is I am only very proficient at one zvz build and that is the overpool build. But when I play against the ai I put it on random so that I can work on scouting and being very reactive to what is going on in the game. Once I see that it is z though I already have gone to 12 drones so I try to do some sort of 14 pool or 1 base roach, but my ability with these builds isn't the greatest.
I don't know how much exactly the harder ai's cheat but its definately there. He didn't even start his second CC till 5:04 yet in less then 10 minutes he has 2x the resources i do. http://img535.imageshack.us/i/screenshot2011012019372.jpg/
replayed against medium level and rolled over it when it did the 1-rax marauder expo. He tried to poke in with 2 marauders and got pushed all the way to his main by 1 zealot+3 stalkers. Also for some reason repeatedly targeted my zealot as opposed to my stalkers when engaging. Over-all it is a decent AI for lower levels and with some minor tweaking may even be useful to people in the higher ups. Like maybe lowering the extra resources it gets ?
On January 20 2011 14:54 Atlas_550 wrote: There's some small things I have started to notice about the AI. First off, the zerg ai in a zvz is tough as heck. The multitasking makes it play almost as good as a top level zerg (Early pressure with expanding). I have only beat it once in a straight up game (without my 10 pool tactice I use but with a 14 pool). Normally I have no problem against the other two races unless I try to be greedy and lose a ton of stuff at the start.
Anyway, the biggest problem I notice is it seems like the ai can still see everything going on. As in it doesn't need to scout. I've noticed every single time that i send my drone scout to the enemy base they are sending out a worker to attack my scout before it is in view. This also made me notice other little things happening (Such as pulling back before it reaches view of my army directly after I reinforce it). So it seems like the computer can still "see" some things a normal player wouldn't.
Thats the line in the code that decides whether there is a fog of war (lack of total vision) or not.
Notice it does set normal = true, meaning there is a fog of war so any information the AI tries to get, will only be information it can truly see just a like a player could. As far as I know at least. Everything I have read (not a lot of info) and the code leads me to believe this is the case. When I had this set to false, it was a lot harder and I certainly did notice weird things it would do because it had vision. I am pretty sure this is working correctly. There could be something added i guess (if it even is possible) that extends its vision, but I don't know why they would have added this in unless it is just what the Blizz AI does as well, because the code base this AI started from had a vision hack in it so there would be no reason to increase vision range.
Pre-Post Edit: I have done some testing on the scouting worker meeting you at its ramp issue. I can not say for certain, but if I were to guess, I would say the AI can see about a probe's length past the fog of war. There are times where it looks like the probe was already there waiting, but this appears actually to be because of the timing of you approaching the ramp happened to be when it was about to send out a scout or make something near ramp and it just repurposes that worker to run after you and grabs another worker to scout or make a pylon. It may also pay attention to you "hi fiving" its scouting worker and expect you to be there at a certain time. Its a bit hard to tell, because I can't ask it how it did it, I just kind of have to figure it out through testing. Either way though, I don't think I want to turn this off at the moment because it would then let you do whatever in its base and that doesn't sound like a good idea to me. I do not believe though that it can see you EVERYWHERE even if its fog of war. At the very least it doesn't know your unit counts of each unit type, I know that for sure. Positioning though is another question and I will have to test to prove it. I think if anything, it may have just a slight bit more range of sight than a player (like I said, about the length of a probe).
A very interesting AI, I got completely taken by surprise when playing medium protoss. I'm pretty bad, lol. I tried again, got a zergy AI... and completely baffled it. I pylon blocked the natural and it sat there and didn't make another unit or building the entire game (I can send you the replay if you want).
On January 22 2011 11:03 theit8514 wrote: A very interesting AI, I got completely taken by surprise when playing medium protoss. I'm pretty bad, lol. I tried again, got a zergy AI... and completely baffled it. I pylon blocked the natural and it sat there and didn't make another unit or building the entire game (I can send you the replay if you want).
Yea it might help to send a replay. There are many factors which can lead to the AI doing different things so it will help to see what you did, what its build was, what map you used, and anything else that might give me a sign what happened.
I tested this just now myself. The first game I didn't play long and didn't finish it out like i should have. In that game I put a 2 pylons where the hatch would go and ran back in to its base. I think i waited there until its lings spawned then I ran to the pylons. They followed me there and killed them, but I didn't wait to see if it would build a hatch there. I assume it would have. The next 2 games it did the same 1 base muta build (this should be taken out shouldn't it guys, does anyone do this?) so the hatch comes later and I think this may have affected how it reacted. It didnt kill the pylons and built its expo somewhere else instead (/facepalm). So regardless, it is safe to say it doesn't react correctly to it. I have yet to see it lock up as you say and not make any more units or structures. As far as I can tell its design is robust enough that it is never does that, though you may show me a way to completely shut it down. What I have seen is it make dumb decisions when you throw it a curve ball which lead to it getting farther and farther behind, and of course I will try to weed as many of these things out as possible.
There are things just so expected like pylon blocking a hatch, or bunker rushing the ramp etc that it really needs be able to handle. Once the AI is a bit more solid, I envision it probably will even have a few "builds" that it will switch to in the case of specific kinds of early cheese or aggression so that it will exhibit the appropriate response you expect from it. For instance, not that I want to promote cheese, but lets say you wanted to practice cannon rushing on Xel'Naga by putting a pylon outside the main, then warping a cannon up top etc. To properly practice that so you can be as affective as possible, it needs to make it as hard as realistically possible to pull off. Cannon rush would probably be the last thing of that sort I would have on my list to do since it seems to be not all that viable when the opponent scouts and reacts correctly, but just as an FYI I do intend to implement specific responses to common early game tactics when I can.
Here's the replay. If you look at his base you can see a drone twitching out trying to do something. This was the Metalopolis map published on NA. metalopolis replay
On January 23 2011 00:58 theit8514 wrote: Here's the replay. If you look at his base you can see a drone twitching out trying to do something. This was the Metalopolis map published on NA. metalopolis replay
Well you did it. You truly broke it It didn't get very far so there is not much to go off of, but there are 3 builds it could have been doing based on the extractor trick. The first build I checked very well could have broken because of this. It needed more supply but it was trying to wait until it had a queen to build the next overlord. It couldn't build the queen because it was waiting for the supply to drop from the drone being lost when the expo was started. It couldn't build the expo because you blocked it and it doesn't know how to deal with it. And it is one of a few builds that uses special expo code which if I am understanding right, it will only try to put it at its closest expansion so it even refused to just put it somewhere else which is different from the few builds I happened to get when I tested.
There are a number of things I could do even before I add in proper code to deal with pylon blocking in general that should fix this build. I will check the other zerg builds as well to see if they have the same total melt down problems. At the very least I will try to get all the zerg builds from doing this complete lock down early game. If it is not too hard to figure out I will try and add some basic code to have some drones kill off the pylon, but I really want to get this dialog patch out so people can tweak difficulty settings and choose specific builds to go against.
Love it as far as helping macro, but army composition doesn't really work too well in TvZ. Ive never seen it go muta's which is disappointing because i really wanted to practice timings and turret positioning. Well worth the time to play for practice though.
On January 23 2011 10:52 Freezy3 wrote: Love it as far as helping macro, but army composition doesn't really work too well in TvZ. Ive never seen it go muta's which is disappointing because i really wanted to practice timings and turret positioning. Well worth the time to play for practice though.
The next update I make to it will allow you to select which build it does if you like or keep it random as it is now. It does have 2 muta builds, though 1 of them is 1 base muta which i am not a fan of.
On January 23 2011 11:16 REM.ca wrote: I noticed the Terran AI on metalopolis bugging when scouting with his scv. the worker gets stuck in my mineral line.
Weird haven't seen that one yet. Does it happen a lot or have you just seen it once? That might have to wait for me to write custom scouting AI (it is using blizzard scouting AI now), but maybe not if I can detect it getting stuck and figure out some "unstick" procedure. Have other people seen this as well?
OK Fyn, I noticed 2 questions you asked (not sure if you asked more).
When it comes to assessing economy (as worker count) it is of light-moderate importance in non-mirror matchups, and more than moderate importance in mirror-matchups. Particularly in the ZvZs I played it was an issue with the AI which is why I mentioned it.
As a general rule it is good to say to always build workers as terran or protoss (to a limit of about 60-80 supply), but at higher levels, it's almost always more complicated than that, because timing attacks are extremely prevalent, where after a 1 base or 2 base economy of certain level is established, the player will cut worker production and dedicate to making an army for a while before attacking..
Oh and I just played a protoss, and when I sent my scout out (steppes of war, and a bit early, like 9 supply time), he sent 4 workers to chase after it. It's a pretty serious problem, although I've only seen it this once.
In the same game, the protoss was also being terrible with his army and probes in general when I was attacking him with speedlings. I was at his ramp/expo area (he had no expo) fighting some of his units, and my lings would continually kill probes that would pass-by during the fight. When I got up into his main he would still keep throwing probes away (don't know what they were trying to do, scout I think?) in a really stupid way - I seen other races do this too of course. Obviously a much harder problem to fix, but pretty big issue. Even military units will be wandering around alone sometimes, which makes them way too vulnerable.
On January 23 2011 11:16 REM.ca wrote: I noticed the Terran AI on metalopolis bugging when scouting with his scv. the worker gets stuck in my mineral line.
Weird haven't seen that one yet. Does it happen a lot or have you just seen it once? That might have to wait for me to write custom scouting AI (it is using blizzard scouting AI now), but maybe not if I can detect it getting stuck and figure out some "unstick" procedure. Have other people seen this as well?
Happened at least 3 times so far. I think I was in top left position each time.
I've played using this AI atleast a few hours a day over the past week and I love it but as always I'll give feedback on any issues -
- I've also experienced scouting worker getting stuck in the mineral line.. - In PvT (AI as Terran) they seem to have a habit of using Helions along with the marines, marauders early-mid game, which imo isnt the wisest of choices. - When we spawn in cross-map positions on a map such as Metalopolis AI's 3rd sometimes goes for the shared base in between both our mains example - If I spawn top left position and he spawns top right, I've seen him expo after natural to the shared northern base whereas any player would go for the southern (or even southern gold) expo.
More feedback to come after I've played more games.
Does anyone know if the fyn/greentea AI gets 'harder' in 2v2, I can STOMP very easy and easy in 1v1 and win most of the time vs med 1v1, but am having issues against med whooping my ass every time in 2v2 and even easy giving us a run for our money in 2v2. Both bronze level players but still should not be having such a harder time in 2v2 than in 1v1
I want to publish the green tea ai onto iccup maps. For the life of me I can not figure out/find where to have the SC2 editor open maps that I have downloaded. Using the battle.net tab is not helpful either, searching popular finds none, searching map/mods only finds one. Someone enlighten me please.
On January 24 2011 14:17 Dvalen wrote: With a little practice on Blizzard maps I am noticing the option to lock published maps so I am assuming all the iccup maps are locked
: I have played vs fyn ki ( P ) but in one game i save me on a isle and the ai doesent comes !! 30 min ??? he see that i was there he has a 200/200 army 50% air Units ,but he dont attack me ?? Bug?
I'm currently messing around with a crazy balance mod and I LOVE this ai because it is much stronger than normal AI, so I can test out my crazy balance changes easier by either playing against AI, or putting them against each other... +100000 bonus points to fyndor for being AWESOME.
It has been fairly painful with many hours searching for typos in my code since this has no compiler, but I am getting pretty close to the first update. More than likely it will include the newer Green Tea as a base hopefully fix some of the bugs you guys have been seeing. I am going to have to replace edit boxes for apm and harvest rate with sliders, not sure how high I should make the maximums. Thinking it might be cool to have the ability to reduce the harvest rate assuming that is possible so you can make it easier for people who aren't quite to the level of what Very Easy is. You guys reminded me that some people are using this for more than 1v1 so I am going to have to add a player selector as well so you can specify a build for each computer AI. If it is not too hard I am going to add the ability to save your custom settings, but it will have to be per map due to how battle.net works. Not sure how it would work as far as the selected builds go though since every time you load the map you could have selected different race combos. I might have to exclude build selection from saving.
Last time I used the Fyn AI, I opened your Xel'Naga caverns map in the editor, then privately published it with custom unit dependencies... Will similar use of the updated maps cause problems in AI or script coding? Thanks for your hard work.
This sounds excellent. I've moved house leaving me without Internet for the moment. I'm hoping to use the sc2allin1 to load up some 4v4 AI matches and see how I fare. Thanks a bunch for all your hard work.
On January 27 2011 14:47 XerrolAvengerII wrote: Last time I used the Fyn AI, I opened your Xel'Naga caverns map in the editor, then privately published it with custom unit dependencies... Will similar use of the updated maps cause problems in AI or script coding? Thanks for your hard work.
I have no idea. I wouldn't think so. Likely your procedure will change though. There is an extra step to preparing the map file now which involves inserting a script file via a mpq editor, but other than that everything is the same. I am not all that familiar with map customizations though so I don't know what affects what. Just curious what do you mean by unit dependencies? You added extra units or abilities to the game?
Does anyone use this map with 2 or more humans at a time? Meaning 1v1 vs AI with obs or 2v2 vs AI with 2 players and 2 AI. I am still torn on how to handle multiple users interacting with the starting dialog at the same time. When someone checks a checkbox for instance, an event happens that lets me know it was checked or unchecked. But as far as I can tell it doesn't tell me who's copy of the checkbox was checked, so when multiple users are in a map at once I would have to check everyones checkbox and try to figure out which one is the one that changed, then update everyone else's checkbox to reflect the correct current value. It would be the kind of there where you could "fight" over the settings because you both have control of the dialog at one time. It is do able, but it really is a pain in the ass. Yet I can't see any other solution other than to only let the first person in the list of users to interact with it which seems kind awkward as well. You could also get in to scenarios where someone wants to do 2v2 pitting human and ai vs human and ai where they want to select the build their computer partner uses. Not sure if anyone would ever do this, but they may have a completely different expectation of the dialog hiding their build order selection. Again I am not sure if anyone, or enough people I should say, would want to do this to make it worth the effort of coding. I am a bit torn. Its not like I can't revisit the dialog code again later and make it better, but it is such a minor part of the AI that I don't want to put much effort in to it on an ongoing basis so I can actually get to improving the AI itself. Anyway I am just trying to get a feel for who would be sad if the dialog only had 1 user able to interact with it to see if this is a place I can cut corners a bit in order to get an update out.
If anyone on the NA servers has some spare time, I need to test all the builds in the upcoming version of the AI and make sure they are labeled properly. I have include the newest builds from Green Tea AI and I do not believe they are all labeled properly. The version I posted is very alpha. Not all the features are enabled or working yet and there are known bugs in the interface.
If you would like to help test, what you will do is log on to NA Battle.net server and search for "faitest" in the map pool. Add 1 AI opponent of what ever race you want to test and leave it on medium. Start the game. When the game starts there will be a new dialog. Uncheck the "Random Build" checkbox and select a build you want the computer to do. Hit the start button and the game will start. Play against the computer and see if the description of the build you chose matches what the computer actually did. Do this for each build (or as many as you can manage to do). If you like it wouldn't hurt let me know what you think of the build, whether it should be in the build list, and whether it is a good build to use against your race (tell me your race of course). But consider all that stuff extra credit. I mainly need to know that it did what it was labeled as so people aren't like WTF when it is released on all the maps in its public release.
On January 31 2011 11:39 Mackus wrote: If you tell the AI to do a specific build, will they transition after if the build you told them to do fails?
Oh you just made me realize the code in the test version will not allow protoss and terran to transition because it will always go right back to that chosen build. Normally after 10 minutes it can choose to change builds if it's army is smaller within a certain threshold which normally means it didn't do too well against your army in an engagement. It can choose to change builds once every 10 minutes, but if it is fine at the 10 minute mark it can for instance choose to transition at the 15 minute mark if things go downhill at that point instead. In that example it has a change to change again at the 25 minute mark. Zerg never did transition, it just makes adjustments in places. I will fix it so protoss and terran can transition as designed before I publish the final version.
The problem though is that even if it does transition it is a random transition in to what could be a build that will work just as poorly or if it gets unlucky it will randomly pick the same build. Beyond fixing the bug I ultimately want to add in a level of intelligence in build selection and transition. For instance, who says that only after 10 minutes can you decide the build you are doing is not going to work against your opponents build. That may be an easy way to code it and I can't fault the Green Tea author for that, but it just seems like you are throwing away a lot of chances to make good decisions. I want make a mechanism that will choose a random plan from the start that should work well as far as race matchup and map goes. But beyond minor deviations, I also want it to be able to quickly change course if it sees an opportunity to take advantage of some weakness its opponent has or if it needs to go in to emergency mode to fend off rush, cheese, or a devastating tech switch.
Edit: I may allow you to choose to force it to stick to a build, but that wouldn't be the default.
On January 31 2011 14:13 Howl67 wrote: That would be truly fantastic if you can manage to get it to do all of those things. Great job so far anyway though.
Thanks. I will try to advance this AI as much as I possibly can. There are some things I would like to do that I can not realistically do with my knowledge and available time of course. At this point I try to restrain myself from playing too much until I can release the next update, but of course I can't always help myself
I can't remember if it was mentioned in the thread, but I have noted some odd behaviour from the AI as well in my matches over the weekend.
In a ZvZ on metalopolis close positions, it kept attempting to take the 3rd between us. I forced a hatch cancel, then later destroyed a completed expansion. Then forced another 2 hatch cancels, army traded, and by the time I'd rebuilt it had taken the expansion again. It never tried to take a 3rd anywhere else and kept sending its overlords over, even though I had a bunch of mutas sniping them.
Anyway it eventually mined out and I rolled over it. I've noticed the AI take expansions in odd places (like taking 3rds across the map in 2v2s). What sort of thinking goes into its expansion choices?
I just noticed 2 people posted maps on NA with the fyn ai tag.
XrollAvenger posted a Xel'naga Caverns map. What is the difference from between the one you posted and what I posted Xroll? If you had any changes you wanted made, let me know. If I agree with what you did I might add it to the base code. If you didn't change it im not too sure why you chose to post a second Xel'Naga : /.
Seion posted the AI on War Zone (2v2 map). Seion, I don't think you got it published correctly. I made a guide to show you how to publish a map. If you have any problems following that guide let me know.
There will be a few more added steps next version for people that publish on maps I have not done yet, but it won't be too much different. Based on these changes though I will also be making the maps available for download with the AI already loaded ready to publish as well as the code by itself for people who want to look at it or do whatever. This will help the people on other servers publish the maps easily and quickly. I would be more than happy package other maps that you guys would like to see published with the AI on my site as well if you let me know which ones you want to see. I don't have any more slots to use to upload maps unless I use my girlfriends account (which I may do), but I can get the maps ready for you to publish yourself or maybe others will want to help publish some.
On January 31 2011 15:08 Fugue wrote: I can't remember if it was mentioned in the thread, but I have noted some odd behaviour from the AI as well in my matches over the weekend.
In a ZvZ on metalopolis close positions, it kept attempting to take the 3rd between us. I forced a hatch cancel, then later destroyed a completed expansion. Then forced another 2 hatch cancels, army traded, and by the time I'd rebuilt it had taken the expansion again. It never tried to take a 3rd anywhere else and kept sending its overlords over, even though I had a bunch of mutas sniping them.
Anyway it eventually mined out and I rolled over it. I've noticed the AI take expansions in odd places (like taking 3rds across the map in 2v2s). What sort of thinking goes into its expansion choices?
I will have to get back to you as to how exactly it currently chooses expos at a later time. I will say that if the newest green tea ai didn't fix some of the things like the metal close position expansions (you can test the "faitest" map, since it happens to be metal), I will have to address that fairly soon if not with this upcoming build. I would think it generally should stick to expansions that are closer to its base than its opponents before choosing other expansions. I don't think that should always be the case though. For instance on metal close positions, I wouldn't mind if it on rare occasions took a risk and tried to sneak in a expo at far gold, close gold, close shared expo, far shared expo, probably in that order of likelihood. But I would think many times you would choose the main and expo of the close air starting location farthest from your opponent because it is closer ground distance and air from your base to that expo than it is from your opponents base and a big attack there would leave his main open for a counter or likely much easier to defend than an expo at your opponents front door. And it seems there should be code to try other expansions instead of failing to keep an expansion at a certain location over and over. And even farther down the road I plan to add code that will help it try and get out of situations where you are pinned in on 1 base unable to expand. Stuff like medivacs, warp prism, ovys, nydus worms etc and either move army out of base using "elevators" or sneak worker to far away expo so it can get the econ to bust out, but that is specific to being pinned on 1 base.
On January 25 2011 14:12 XerrolAvengerII wrote: I'm currently messing around with a crazy balance mod and I LOVE this ai because it is much stronger than normal AI, so I can test out my crazy balance changes easier by either playing against AI, or putting them against each other... +100000 bonus points to fyndor for being AWESOME.
The repost of Xel'naga has my own experimental unit stats. etc. I wanted to be able to test my experiments on AI, and by placing AI against each other. I hadn't realised that I'd made a Public of that Xel'naga Caverns. Thanks for asking though, that's important.
On January 25 2011 14:12 XerrolAvengerII wrote: I'm currently messing around with a crazy balance mod and I LOVE this ai because it is much stronger than normal AI, so I can test out my crazy balance changes easier by either playing against AI, or putting them against each other... +100000 bonus points to fyndor for being AWESOME.
The repost of Xel'naga has my own experimental unit stats. etc. I wanted to be able to test my experiments on AI, and by placing AI against each other. I hadn't realised that I'd made a Public of that Xel'naga Caverns. Thanks for asking though, that's important.
In your Faitest, playing as a Zerg, you don't spawn with 3 Larvae on your Hatch. When I was playing PvT, the AI never SCV scouted. He rallied his buildings to his army which was just walking from position to position trying to find my base. It also built its first expansion in between our bases on Metal instead of their Nat and put an OC on it. (We spawned close air/ground). Only issues I've noticed so far.
On January 31 2011 15:18 fyndor wrote: I just noticed 2 people posted maps on NA with the fyn ai tag.
Seion posted the AI on War Zone (2v2 map). Seion, I don't think you got it published correctly. I made a guide to show you how to publish a map. If you have any problems following that guide let me know.
Sorry about that, I must have chosen the wrong folder when importing. It should be fixed now. Thanks for all your work on this, by the way.
Edit: Also made a Scorched Haven version just to be sure I was doing it properly.
First game I played ZvP was excellent, despite some small flaws in the AI it was a relief playing the new AI after getting used to blizz's. Im going to keep checking up on this.
On January 31 2011 17:24 Mufaa wrote: In your Faitest, playing as a Zerg, you don't spawn with 3 Larvae on your Hatch.
@#$% I thought I noticed that while testing, but I was hoping it was just in my head. That makes me a bit nervous because I am not quite sure how to fix that Thanks for the heads up. Does anyone know if the YABOT mod does that? It has a similar dialog at the start and I assume the delayed start because of the dialog must be the reason for the bug.
Well I think my suspicions are correct. There is a "bug" that happens when a map does not initialize everything at the moment the map loads. It causes the larva to not spawn. I have seen people talk about it in forums so I know its not just me. I know YABOT deals with it, but YABOT is under GPL license and as I understand it if I use any YABOT code then I would have to GPL Fyn AI. I don't feel like I have the right to apply a new license to a code-base that I used from someone else's project. I may talk to the Green Tea AI dev just to see what he thinks so I can maybe just move on instead of getting bogged down by this bug since I absolutely can't release with this bug. If anyone for whatever reason knows a method to get these larva to spawn correctly after a delayed game start please let me know.
Edit: Now that I think about it YABOT has Green Tea in it. Maybe the Green Tea dev doesn't mind. I will for sure just ask him. Copying YABOT code would be the easiest way and I believe it has reset functionality as well which would be cool to copy.
Apparently Texas was doing forced blackouts yesterday. One of the times the power came back on must have destroyed my motherboard. Fyn AI development is going to have to pause for a bit until I can get the updated code off that harddrive
On February 06 2011 12:18 SlapMySalami wrote: What the heck on medium ZvT the guy went and battlecruiser rushed me
The real question is, did it win? The current version of the AI does not take in to account matchups when choosing builds by the way. This will be something it will take in take in to account once I am able to implement it.
On February 04 2011 00:14 fyndor wrote: Apparently Texas was doing forced blackouts yesterday. One of the times the power came back on must have destroyed my motherboard. Fyn AI development is going to have to pause for a bit until I can get the updated code off that harddrive
Do you know what would make me laugh? If I could take 1 Fyn AI and put it against 2 or 3 blizzard AI for fun.... Wish I could see that, but selecting AI customization for each player would require fancier scripted menues and what not.... still, maybe a zillion years in the future we may be able to see something like this xD
On February 09 2011 04:10 XerrolAvengerII wrote: Do you know what would make me laugh? If I could take 1 Fyn AI and put it against 2 or 3 blizzard AI for fun.... Wish I could see that, but selecting AI customization for each player would require fancier scripted menues and what not.... still, maybe a zillion years in the future we may be able to see something like this xD
Um its possible but I don't know if it would be worth the effort to keep the new and old AI in parallel
congrats on this AI I really enjoy playing it, if only because it is more calming than playing a person and sometimes I want that. I'm still bronze, but the blizzard AI is hopeless by comparison. Medium AI will mostly roll me unless I save and load the game one or two times to fix big mistakes like supply block or positioning mistakes.
It always expands before me, and it always streams in an attack, it always macros almost perfectly. If my macro is not good, I am on the back foot and then my macro slips further, then I lose. If the AI could play battle.net I am sure it would be up to diamond league purely based on its insistent early attack and perfect one then two base macro.
Wishes for this AI:
I wish it would identify my units that are racking up kills, and focus on how to dislodge them. The AI wastes itself on in repeated patterns on siege tanks and bunkers, to take just one example. How about when it loses a unit making a guess about what opposing static defense was responsible then avoiding static defenses that are racking up kills, at least until it builds double the force? For example take an island and build a lot of turrets then watch it burn through its minerals failing to avoid repeated mistakes.
When it makes a decision, (attack / retreat), don't keep re-evaluating the decision every 1/2 second. Or do internally, but be more human about sticking to a recent previously made decision for at least a few seconds. All the time I see the AI wiggle back and forward as it changes its mind about whether it can win or not. How about predicting my invisible army size based on observing how fast the visible one was growing? If AI army < predicted army (rather than last observed army), don't attack without counter unit advantages.
An ability for me to choose in advance what units it is going to focus on. Obviously this makes it easy to counter, however it is useful for gaining micro skills against (for example) "mass carriers". (sorry if these are options somewhere)
A way to set at the start whether it should always attack, or turtle up and build defenses. I don't get any practice at assaulting a defended base. AI seems to always attack as much as it can.. always.. so since I know it wastes resources on static defenses, if I can hold against the attack, it isn't long before I'm accumulating a bigger army.
Would it be possible to buff the AI's decision making as far as attacking goes? It tends to just keep sending waves and waves of relatively small numbers of units, at least on "Hard".
Also, can you add the GSL maps? Terminus RE, Taldarim Altar, Crossfire SE, etc.
On February 09 2011 04:10 XerrolAvengerII wrote: Do you know what would make me laugh? If I could take 1 Fyn AI and put it against 2 or 3 blizzard AI for fun.... Wish I could see that, but selecting AI customization for each player would require fancier scripted menues and what not.... still, maybe a zillion years in the future we may be able to see something like this xD
Um its possible but I don't know if it would be worth the effort to keep the new and old AI in parallel
Well I'd like to see it. Definitely not top priority, but would be pretty cool to have a choice.
As Fyndor posted a few days ago, the development has to pause because of hardware issues. Anyway, I really like this AI and enjoy playing against it. Thanks for the good work!
Oh and it would be really cool to have the new GSL maps, as Buddhist already requested. I read you might have a problem uploading them on your account, as it is limited to 10 maps. I could lend you my 10 map spaces on the EU Server, if that helps.
I don't know how much work it takes to make the AI work on different maps though, so don't feel rushed. Just wanted to show my appreciation and offer my map spaces.
The only map i have changed since the Starcraft 1.2 patch has been the FAITest map on NA which I used to test online functionality in the next version since the offline version works slightly differently (annoying!). I intend on helping out publishing a few extra maps to NA myself with my girlfriends account, but I am waiting until I release the next version of the AI. My computer is still broken and I don't have the money atm to really fix it. When I get off my butt and decide to quit playing the game and instead develop the AI again on my girlfriends free laptop she got a few months ago (THANKS SONY!), I will go buy a USB harddrive enclosure so I can get the data off the drive and into the laptop. Sorry I just wanted to take a week or so to actually play again
As far as EU is concerned any maps would always have to come from someone other than me. There is code and directions here on how to publish maps. These directions will change in the future, but you can use them to publish the current version right now with different maps that are not posted on your server. In the future I think what I will do is have you guys request specific maps, and I will package the AI in to map files for the requested maps and have those files downloadable. I will always keep the code open source, but I think it would just be easier for everyone if I just packaged the AI in to the maps and just gave you the map files to publish on your servers, especially with the new changes to the AI. Once you get to the point of having the map file, publishing become super simple to learn.
On February 14 2011 22:33 Hopelessnoob wrote: hey are any of the maps published on NA? i somehow cannot find anything...
They were still published on NA last night. I will check over lunch time but they should still be there. Go to Create Game. Select popular from the drop down and search for "fyn". You may have to select Melee game type to find them, I can't remember what other category they are published under.
I was playing around with maps with this AI somewhat and I noticed that just like Blizzard AI this one also does not wall off on its ramp with makes my games as Zerg vs AI very easy.
Basically whenever I win a fight I need to stop myself from attacking his base or the game ends there. Also testing any early speedling play kills the AI
On February 15 2011 01:52 -Archangel- wrote: I was playing around with maps with this AI somewhat and I noticed that just like Blizzard AI this one also does not wall off on its ramp with makes my games as Zerg vs AI very easy.
Basically whenever I win a fight I need to stop myself from attacking his base or the game ends there. Also testing any early speedling play kills the AI
At some point I would like to fix this but I am still trying to figure out a way to do it that doesn't require me hard-coding positions for each map for each spawn point. I would like this AI to remain publishable on maps other than just the 1v1 Blizzard published maps and it may be hard to really keep that manageable if I had to hard code building positions. I may ultimately have to do that, but I am not going to rush to that solution and would rather get better acquainted with the API and see if I can figure out a better solution.
I am high Silver/Gold player and I can beat Blizzard Insane AI by macro game no problem. But Terran FynAI with Hard+ difficulty completely steamrolls me. MMM, Raven push, Tanks, Mech, it even nuked my pylons! I think I am doing some terrible mistake :D
Can any GOLD player upload a replay of beating the Fyn AI on Xel Naga caverns, TvZ, without rushing the AI when it still on one base, and without abusing the AI weakness to drops .. ?? (on medium)
I'm kind of curious to know if beating it via expanding to two then three bases requires more than gold level skills..
I played this awhile ago and recently lost it on my recently played tab, it no longer shows up on the custom game list (in melee or custom tab), anyway to play it?
I downloaded the Shakuras Plateau version from the link the OP posted on the first page (I used to play on the Metalopolis version), is there a way to add that to my single player list or do I have to publish it? Because I tried publishing it, but I have no idea what to do, it said I can't publish it because of the name (that was appearing on the left side) and I changed the name to heaps of random stuff but still didn't let me publish it so I am guessing the name wasn't the problem =/
As an example here is a replay of my defeat of the AI on Xel Naga in TvP, from three bases.
I am not so proud, I never scouted and had no idea where his army was, but find Z much harder, hence my question I'd be really curious to see what ladder level can comfortably defeat Z Fyn AI as TvZ ..
On March 02 2011 07:34 jpaugh78 wrote: Is there any chance we can get the Fyn AI onto the new ladder maps they just released?
Yea I should be able to do that. FYI though, because of the number of maps I can upload, this will mean removing all maps that blizzard removed as well.
On March 02 2011 17:40 nyc863 wrote: As an example here is a replay of my defeat of the AI on Xel Naga in TvP, from three bases.
I am not so proud, I never scouted and had no idea where his army was, but find Z much harder, hence my question I'd be really curious to see what ladder level can comfortably defeat Z Fyn AI as TvZ ..
On March 03 2011 05:45 ArcticVanguard wrote: ^ Or post them here and someone will upload them. I could take two, though I can't guarantee regular updates
I could, but actually now I am thinking that few would care. If you want to play against someone on the old maps (blizzards normal version) they are still there of course. But if you want to play against an AI, I am not sure why someone would want to do it on a map they will never see in the ladder anymore.
On March 02 2011 07:34 jpaugh78 wrote: Is there any chance we can get the Fyn AI onto the new ladder maps they just released?
Yea I should be able to do that. FYI though, because of the number of maps I can upload, this will mean removing all maps that blizzard removed as well.
Yeah that's no problem IMO. As others have said, who's going to practice 1v1 on a map they can't play 1v1 on?
Or you could let another user do it. For example, if you can port them to GSL maps, I'd be happy to upload them on your behalf.
I'm loving your AI for offline practice and the new upgrade looks awesome. If you need someone else to upload any maps for you I can totally do it. No uploaded maps NA server. :D
I really like this AI. I play on medium AI TvP and as a silver player it's tough.
Does anyone have a replay of beating the medium AI as Terran (vs Protoss) with a 1/1/1 opening?
I play on scrap station and he does a 3 gate opening and I can't seem to hold that with 1/1/1. Im trying to get a cloaked banshee.
Now ofc I am making mistakes but I just would like to see a good player do a 1/1/1 opening vs 3 gate that the AI is doing on scrap station just to see how it should be done.
A friend and me have taken the effort to follow the guide from the OP and have published the new 1v1 and 2v2 ladder maps on the EU Server.
namely they are:
Backwater Gulch (Fyn AI) Gutterhulk (Fyn AI) Khaydarin Depths (Fyn AI) Omega Sector (Fyn AI) Red Stone Gulch (Fyn AI) Slag Pits (Fyn AI) The Shattered Temple (Fyn AI) Typhon Peaks (Fyn AI)
I recommend searching for "fyn", that'll show you all Fyn AI maps. If you search for "fyn AI", however, you find lots of other maps.
Thanks a lot to Fyndor again, for both the AI and the very well made guide. Once you understand how it's done, it is really not that much work to implement new maps with the AI.
On March 04 2011 19:59 Robqxz wrote: A friend and me have taken the effort to follow the guide from the OP and have published the new 1v1 and 2v2 ladder maps on the EU Server.
namely they are:
Backwater Gulch (Fyn AI) Gutterhulk (Fyn AI) Khaydarin Depths (Fyn AI) Omega Sector (Fyn AI) Red Stone Gulch (Fyn AI) Slag Pits (Fyn AI) The Shattered Temple (Fyn AI) Typhon Peaks (Fyn AI)
I recommend searching for "fyn", that'll show you all Fyn AI maps. If you search for "fyn AI", however, you find lots of other maps.
Thanks a lot to Fyndor again, for both the AI and the very well made guide. Once you understand how it's done, it is really not that much work to implement new maps with the AI.
Any chance you could add a -reset command to these maps a la YABOT where it resets everything so that you do not need to recreate the game every time you finish playing a round?
On March 04 2011 19:59 Robqxz wrote: A friend and me have taken the effort to follow the guide from the OP and have published the new 1v1 and 2v2 ladder maps on the EU Server.
namely they are:
Backwater Gulch (Fyn AI) Gutterhulk (Fyn AI) Khaydarin Depths (Fyn AI) Omega Sector (Fyn AI) Red Stone Gulch (Fyn AI) Slag Pits (Fyn AI) The Shattered Temple (Fyn AI) Typhon Peaks (Fyn AI)
I recommend searching for "fyn", that'll show you all Fyn AI maps. If you search for "fyn AI", however, you find lots of other maps.
Thanks a lot to Fyndor again, for both the AI and the very well made guide. Once you understand how it's done, it is really not that much work to implement new maps with the AI.
what ver of green tea AI is in this maps?
I simply downloaded the recent version, as it was said in the guide. Don't know about the exact version number though, sorry. But if I recall correctly there was only one version to download where I looked, anyway.
On March 04 2011 18:21 papaz wrote: I really like this AI. I play on medium AI TvP and as a silver player it's tough.
Does anyone have a replay of beating the medium AI as Terran (vs Protoss) with a 1/1/1 opening?
I play on scrap station and he does a 3 gate opening and I can't seem to hold that with 1/1/1. Im trying to get a cloaked banshee.
Now ofc I am making mistakes but I just would like to see a good player do a 1/1/1 opening vs 3 gate that the AI is doing on scrap station just to see how it should be done.
This isn't explicit scrap station advice, last time I did this (today) was Xel Naga caverns.
You need to bunker up, and have a repair SCV or two.
the AI likes to do a fairly big push at 7 minutes or so, with a mixture of sentries (which never force field) stalkers and zealots. You need to have some marauders with concussive, marines and a bunker or two. It is easy if you have not fast expanded to defend, but then harder as AI has fast expanded and you are behind and falling further behind. If you have your 2nd cc built, you need siege tanks and siege asap, then you slowly expand your domain of control with new bunkers and keep macro up. Eventually he attacks into your siege and loses most (ghost helps). Then you can go destroy one of his expansions. And go on from there..
On March 04 2011 19:59 Robqxz wrote: A friend and me have taken the effort to follow the guide from the OP and have published the new 1v1 and 2v2 ladder maps on the EU Server.
namely they are:
Backwater Gulch (Fyn AI) Gutterhulk (Fyn AI) Khaydarin Depths (Fyn AI) Omega Sector (Fyn AI) Red Stone Gulch (Fyn AI) Slag Pits (Fyn AI) The Shattered Temple (Fyn AI) Typhon Peaks (Fyn AI)
I recommend searching for "fyn", that'll show you all Fyn AI maps. If you search for "fyn AI", however, you find lots of other maps.
Thanks a lot to Fyndor again, for both the AI and the very well made guide. Once you understand how it's done, it is really not that much work to implement new maps with the AI.
Now somebody just needs to do this for the US server.....
Well, why don't you? It's really not that hard to figure out. The Guide actually tells you every single step in great detail, even with pictures. Nothing you could possibly do wrong there. Once you know what you need to do, it only takes about two minutes a map.
On March 04 2011 19:59 Robqxz wrote: A friend and me have taken the effort to follow the guide from the OP and have published the new 1v1 and 2v2 ladder maps on the EU Server.
namely they are:
Backwater Gulch (Fyn AI) Gutterhulk (Fyn AI) Khaydarin Depths (Fyn AI) Omega Sector (Fyn AI) Red Stone Gulch (Fyn AI) Slag Pits (Fyn AI) The Shattered Temple (Fyn AI) Typhon Peaks (Fyn AI)
I recommend searching for "fyn", that'll show you all Fyn AI maps. If you search for "fyn AI", however, you find lots of other maps.
Thanks a lot to Fyndor again, for both the AI and the very well made guide. Once you understand how it's done, it is really not that much work to implement new maps with the AI.
Now somebody just needs to do this for the US server.....
Well actually I didn't post them up because I misread his post and didn't realize he did it for EU not NA. I will try and get them posted tomorrow or the next day.
On March 04 2011 19:59 Robqxz wrote: A friend and me have taken the effort to follow the guide from the OP and have published the new 1v1 and 2v2 ladder maps on the EU Server.
namely they are:
Backwater Gulch (Fyn AI) Gutterhulk (Fyn AI) Khaydarin Depths (Fyn AI) Omega Sector (Fyn AI) Red Stone Gulch (Fyn AI) Slag Pits (Fyn AI) The Shattered Temple (Fyn AI) Typhon Peaks (Fyn AI)
I recommend searching for "fyn", that'll show you all Fyn AI maps. If you search for "fyn AI", however, you find lots of other maps.
Thanks a lot to Fyndor again, for both the AI and the very well made guide. Once you understand how it's done, it is really not that much work to implement new maps with the AI.
what ver of green tea AI is in this maps?
Sorry I can't remember the version offhand (I'm at work), unless it has changed I believe it is a version behind Green Tea. I plan on updating it, but I was hoping to wait to introduce that change with other changes I am planning. There are a number of things that have prevented me from updating the maps yet, but if it takes me too much longer to get all my changes finished I can just update the green tea part in the interim.
On March 05 2011 07:21 Wihl wrote: Any chance you could add a -reset command to these maps a la YABOT where it resets everything so that you do not need to recreate the game every time you finish playing a round?
Yea I am pretty sure it is possible. The process is a bit beyond my current expertise, but YABOT is open source and the code seems straight forward. I think I likely will take that code from YABOT, I just haven't done it yet because it is licensed under GPL and I have not seen a stated license for Green Tea and was hoping to chat with the Green Tea guy before GPLing my code. I am pretty sure he wouldn't mind, after all YABOT has some or all of Green Tea in it so the YABOT guys already have GPLed Green Tea. Something just doesn't sit right with me about changing the license of someone else's source code without talking to them. Dunno, I'm probably just being silly.
Regardless, this reset code is actually the same code I will likely need to fix the larva spawn issue that the FAITest map on NA has (my test code for the next version), so unless I can find a creative way around that larva bug I will have no choice but to implement "-reset". Well atleast the code behind it, but of course I would add chat command as well since it would be useful.
I really like this program and do in fact find it preferable to the regular Green Tea AI which cheats.
Some questions:
Is the only difference in terms of code between Green Tea and Fyn that the maphack feature of the latter is turned off or is it the case that Fyn also has more code for scouting?
Also, will the new maps have the latest version of Green Tea incorporated and, if so, what about the old maps that are currently on Battlenet (U.S)...will they be removed/updated?
I tried the FAItest map on NA and found that when I told the medium zerg AI to do a 6 pool and 8 pool, it ended up going 16 hatch into 2 base roach/hydra or muta. I was playing Protoss if that helps.
any tips on TvP very hard? No matter what I do I can't seem to beat the ai, its always 5-10 probes ahead of me due to chrono boost and gets a shit ton more minerals, which lets it safely expand which lets it get more minerals... : (
I would never recommend using the computer AI to practice, because you won't get any better against an actual player for the soul purpose that the builds the computer does are not even remotely similar to what actual players do. However, using the computer to practice a build order or building placement is very use full.
Did something happen to the AI? Just played a 1v1 on xel naga vs med zerg and he made like 8 lings and 1 roach, never expanded just sent drones to the expansions and did nothing.... wasnt like that last time I used it...
On March 17 2011 04:38 Bizarro252 wrote: Did something happen to the AI? Just played a 1v1 on xel naga vs med zerg and he made like 8 lings and 1 roach, never expanded just sent drones to the expansions and did nothing.... wasnt like that last time I used it...
Fyn AI is based on Green Tea AI, which has 17 commanders for terran, 17 for protoss, and 9 for zerg. Each commander does different build, obviously. However, when transitioning into the mid-game, I think the AI will then respond to your army composition. Can Master Fyn comfirm?
On March 16 2011 13:36 fyndor wrote: Sorry I took so long, but the maps I published to NA have been updated to reflect the current 1v1 map pool. Let me know if there are any issues.
I dont see any new maps when i search for fyn ai. Do they have different names this time ??
On March 16 2011 13:36 fyndor wrote: Sorry I took so long, but the maps I published to NA have been updated to reflect the current 1v1 map pool. Let me know if there are any issues.
I dont see any new maps when i search for fyn ai. Do they have different names this time ??
As you apparently realized and so did jpaugh78 who sent me a PM, I was dumb and forgot to ensure the maps were marked as public. I have just republished them as public. Sorry its been a while and a forgot a quirk in the software the marks maps a private when changes are made. Let me know if you have any problems.
On March 17 2011 04:38 Bizarro252 wrote: Did something happen to the AI? Just played a 1v1 on xel naga vs med zerg and he made like 8 lings and 1 roach, never expanded just sent drones to the expansions and did nothing.... wasnt like that last time I used it...
Fyn AI is based on Green Tea AI, which has 17 commanders for terran, 17 for protoss, and 9 for zerg. Each commander does different build, obviously. However, when transitioning into the mid-game, I think the AI will then respond to your army composition. Can Master Fyn comfirm?
I am pretty sure you are talking about a Green Tea version ahead of what I have published on the servers. In short the Terran and Protoss in Fyn have fewer builds to choose from than the next Green Tea version. I am trying to decide right now if I want to go ahead and publish that version or wait until I can add in the next round of modifications I intend to make.
On that front I just found out my computer is working again which has all the source code on it so I intended to take up the AI project again starting this weekend. How the hell a computer can completely refuse to start one day then work another? And even weirder was that it was almost like starting a car yesterday when I randomly decided to press the ON button while I was in that room just for kicks. I pressed the button and let it go and it seemed like the fans went a split second longer before stopping than I had remembered it doing when I first realized it wasnt working. So I pressed the button and held it even longer (like 2 sec) and the computer seemed to almost "crank over". So the next time I pressed it and held for like 5 sec and it started. If that was the behavior previously I wouldn't think anything of it but I have never seen this or any other computer do that. Seems pretty weird.
On March 17 2011 04:38 Bizarro252 wrote: Did something happen to the AI? Just played a 1v1 on xel naga vs med zerg and he made like 8 lings and 1 roach, never expanded just sent drones to the expansions and did nothing.... wasnt like that last time I used it...
Fyn AI is based on Green Tea AI, which has 17 commanders for terran, 17 for protoss, and 9 for zerg. Each commander does different build, obviously. However, when transitioning into the mid-game, I think the AI will then respond to your army composition. Can Master Fyn comfirm?
I am pretty sure you are talking about a Green Tea version ahead of what I have published on the servers. In short the Terran and Protoss in Fyn have fewer builds to choose from than the next Green Tea version. I am trying to decide right now if I want to go ahead and publish that version or wait until I can add in the next round of modifications I intend to make.
On that front I just found out my computer is working again which has all the source code on it so I intended to take up the AI project again starting this weekend. How the hell a computer can completely refuse to start one day then work another? And even weirder was that it was almost like starting a car yesterday when I randomly decided to press the ON button while I was in that room just for kicks. I pressed the button and let it go and it seemed like the fans went a split second longer before stopping than I had remembered it doing when I first realized it wasnt working. So I pressed the button and held it even longer (like 2 sec) and the computer seemed to almost "crank over". So the next time I pressed it and held for like 5 sec and it started. If that was the behavior previously I wouldn't think anything of it but I have never seen this or any other computer do that. Seems pretty weird.
You don't have to crankstart your computer every time you turn it on? How else do you get the engine running?
after around 7 attempts I finally beat very hard TvZ... the key was Griffith 4OC --> aggressively expo --> early double upgrades --> mid-game siege tank crawl to try to take down some hatches --> thor switch after too many ultras and mutas --> stop making SCV's after you have ~100
On March 15 2011 08:49 frozt_ wrote: I would never recommend using the computer AI to practice, because you won't get any better against an actual player for the soul purpose that the builds the computer does are not even remotely similar to what actual players do. However, using the computer to practice a build order or building placement is very use full.
Fyn AI is actually pretty challenging. It's a good way to test something like... "Well if the AI can do that and kill me, even if it's unlikely for a player to do it, he can too, so what I did isn't safe".
the AI is good for practicing overall things like macro and generalized production capability of various builds, especially on levels of difficulty where the AI receives income multipliers as the detrimental effects of any slip-up such as getting supply blocked or simply forgetting to make units is amplified.
Edit: I think I will also publish the maps files this time so others can more easily spread the updates to the other non-NA servers. The change will be that to publish a map you open up the file with the Galaxy editor, then go to the publish screen and publish it. Basically cuts out a few steps. If I do end up doing this (currently the plan) I will amend the guide to reflect the changes to the process.
I have updated the maps to reflect the newest patches changes on the NA servers. If you are one of the people that have been publishing maps on other servers I have added the maps with the script already imported in to them to the download section of the Fyn AI site. I have not updated the guide yet to reflect how to use the map files (its pretty late in Texas :X), but its pretty easy. Just open them up and hit the publish button and do all the steps from there per usual.
Let me know if there are any problems.
Edit: I have had to remove faitest, the alpha version of the next Fyn AI, to make room for the new map. I will add it back when I am ready with the next round of changes (larva fix) under a different account.
My buddy Phil and I have uploaded and updated the maps from the current 1v1 map pool to the EU Server.
We are going to test them in a minute, just to go sure everything went well. If any problems occur, I'll update this post to let you know. Else, it should all work fine.
The maps we updated are: Slag Pits (Fyn AI) Backwater Gulch (Fyn AI) Typhon Peaks (Fyn AI) The Shattered Temple (Fyn AI)
Furthermore we added: Tal'Darim Altar LE (Fyn AI) Shakuras Plateau 1.3 (Fyn AI)
Don't be confused, there was an older version of Shakuras Plateau Fyn AI uploaded by BlackGun, so I had to add "1.3" to the name.
These six maps were all taken from the *.zip that Fyndor provided to download. We just published them without any further changes.
Thus, we have all maps from the latest 1v1 mappool on the EU server, since XNC and Scrap Station have already been up, as well as Delta Quadrant and Metalopolis.
On March 23 2011 22:16 n3wb13 wrote: I've update SEA with all the maps also except delta quadrant... which i make slot for 3v3 arakan citadel that i used to play with my friends...
I'm not sure why, but on Typhon Peaks the AI did something weird.
They basically stayed on one base forever, building so many different tech paths. This is unlike everything I have yet seen from the AI, even though I played a lot of games, so I don't think it was just some special tactic. Would've been a giant coincidence if they both did some funky build that I never happened to see before.
This was one of the maps we uploaded a couple hours ago, be sure to check page 12 for further information. All other maps worked well though, only Typhon Peaks in this one match had some weird stuff happening.
On March 23 2011 23:30 Robqxz wrote: I'm not sure why, but on Typhon Peaks the AI did something weird.
They basically stayed on one base forever, building so many different tech paths. This is unlike everything I have yet seen from the AI, even though I played a lot of games, so I don't think it was just some special tactic. Would've been a giant coincidence if they both did some funky build that I never happened to see before.
This was one of the maps we uploaded a couple hours ago, be sure to check page 12 for further information. All other maps worked well though, only Typhon Peaks in this one match had some weird stuff happening.
I had the exact same thing happen over here on the NA server. I wanted some more 4 player maps with fyn ai so I added Typhon Peaks.
The exact same thing happened, they just stayed on one base forever.
I just assumed that I had done something wrong in the process of adding the AI.
On March 24 2011 12:29 Eleaven wrote: shame it maphacks.
? The current main difference between Fyn AI and Green Tea is actually the fact that Fyn AI does not map hack. What are you talking about? Is there bug you are seeing?
I noticed a weird thing about the AI... whenever I do a drop on the min line and end up killing a bunch of workers (as in crippling them to <10) then fly away, the AI just doesn't like to re-drone for whatever reason. It'll rebuild a few buildings, pop a few infantry units here and there, but it just won't try to reup the worker count.
I'd love to see a little more variety to the AI, I know the terran AI does cloak banshees sometimes but I'd like to see protoss do DT rushes instead of just 1 base colossus or warpgate expand all the time. Some all-ins (ie marine/scv) would be nice too.
On March 25 2011 00:09 ballasdontcry wrote: I noticed a weird thing about the AI... whenever I do a drop on the min line and end up killing a bunch of workers (as in crippling them to <10) then fly away, the AI just doesn't like to re-drone for whatever reason. It'll rebuild a few buildings, pop a few infantry units here and there, but it just won't try to reup the worker count.
I'd love to see a little more variety to the AI, I know the terran AI does cloak banshees sometimes but I'd like to see protoss do DT rushes instead of just 1 base colossus or warpgate expand all the time. Some all-ins (ie marine/scv) would be nice too.
And thanks for updating the AI to the new maps!
The builds it does are picked randomly from a list. Depending on how many times you have played it you may have not seen all its builds. Granted though, I do intend to make significant changes to the builds and I intend to change it from random builds to random builds selected from race/map appropriate lists.
On March 24 2011 12:29 Eleaven wrote: shame it maphacks.
? The current main difference between Fyn AI and Green Tea is actually the fact that Fyn AI does not map hack. What are you talking about? Is there bug you are seeing?
I have noticed that it does, or at least might be, on medium setting anyways (I thought it was not supposed to maphack until hard and very hard settings).
Whenever I have an obs in the zerg base the queen will always go to it and stand under it, if I move it, the queen will move to stand under it again, later it will get a overseer and kill it.
Also a worker always seems to be coming at me when I scout at the beginning of the game.
On March 24 2011 12:29 Eleaven wrote: shame it maphacks.
? The current main difference between Fyn AI and Green Tea is actually the fact that Fyn AI does not map hack. What are you talking about? Is there bug you are seeing?
I have noticed that it does, or at least might be, on medium setting anyways (I thought it was not supposed to maphack until hard and very hard settings).
Whenever I have an obs in the zerg base the queen will always go to it and stand under it, if I move it, the queen will move to stand under it again, later it will get a overseer and kill it.
Also a worker always seems to be coming at me when I scout at the beginning of the game.
Anyone else notice these things?
EDIT: thanks for your all great work btw!
What you are seeing is a setting that is turned on that tells the AI it is allowed to notice invisible. I figured I would leave it on since players as well can see invisible units if they pay enough attention. A map hack would be letting it see an observer in the "black" areas (forgot what thats called lol) when it clearly doesn't have vision over that area. Try putting an observer out of visible range of the AI yet near its base and see if it decides after it has a way to see invisible units (ie overseer) to go after it. It would be the same kind of thing you would see if someone was using a map hacking program. If someone is map hacking you may see them dart across the map to some location previously invisible area and magically find and snipe your obs. You should not see the AI do this, it just does as players do and it can "see" the light being bent by obs that are in visible range.
The worker coming to prevent scouting is also a setting a chose to turn on until I have a custom logic to do so. Without it, it would be very easy to cheese it (not that its hard) since it would let you do whatever until it got military units. It seems to give the AI a bit of extra visible range outside its base (only during early game I believe). I am not a fan of this but for now the alternative would be to let you roam around unchallenged in its base early game which I also do not like.
The TLDR is that it doesn't map hack, but it does have slightly increased vision range around its base early game.
I've noticed that the AI responds particularly badly to things such as gas steals ( pulls all of its workers to kill it), perhaps it could adjust its intended build and just kill it with army units.
On March 24 2011 12:29 Eleaven wrote: shame it maphacks.
? The current main difference between Fyn AI and Green Tea is actually the fact that Fyn AI does not map hack. What are you talking about? Is there bug you are seeing?
No bug, just maphack.
I let the AI scout me, and then i drop another base in an opposite main.
Before the AI comes to attack me, it will go to the opposite main and kill it, and then come and attack me.
If i have the xel naga tower, but the AI has no vision of my units, he will still maintain a specific distance from my units. If i push forward very slightly, he will move back very slightly, without any vision of me.
I noticed both of these things the very first time i played vs this AI, and subsequently asked a friend to obs various different tests like expanding cross map, or proxying a factory etc, each time the AI would "scout" the area in question with his entire army... yet no other place on the map.
On March 24 2011 12:29 Eleaven wrote: shame it maphacks.
? The current main difference between Fyn AI and Green Tea is actually the fact that Fyn AI does not map hack. What are you talking about? Is there bug you are seeing?
No bug, just maphack.
I let the AI scout me, and then i drop another base in an opposite main.
Before the AI comes to attack me, it will go to the opposite main and kill it, and then come and attack me.
If i have the xel naga tower, but the AI has no vision of my units, he will still maintain a specific distance from my units. If i push forward very slightly, he will move back very slightly, without any vision of me.
I noticed both of these things the very first time i played vs this AI, and subsequently asked a friend to obs various different tests like expanding cross map, or proxying a factory etc, each time the AI would "scout" the area in question with his entire army... yet no other place on the map.
Can you provide a replay please. Are you sure your second base wasn't a part of the extended scouting path? It does scout the entire map through the entire game. Actually it does this more than I like because it focuses I believe too much on scouting for bases it has not found yet rather than keeping tabs on army makeup. Admittedly though I have not played the AI really since the latest patch so for all I know stuff may have changed internally.
After playing a game on this, it feels like the AI has nothing going for it past the early game. I did a ZvP against it on medium difficulty, and while it did a cool "zealot shark-mode" to force me into building extra lings early, it then seemed to not build units, over-extend itself, and get crushed.
The Typhon Peaks map uploaded by you to bnet, as well as the one in the .rar on your Google Code site, does not seem to have had the AI imported at all, and thus defaults to the Blizzard one.
Hey fyndor, I love using your AI! As a gold player, it is extremely helpful to work on my mechanics with it...
One little glitch i noticed was on xelnaga caverns when i was protoss to the north and the AI was terran I scouted the AI rushing me with a few marauders and i probably would have been killed but for whatever reason it didnt come when it should have, when i checked the replay the AI stopped to destroy the rocks at MY gold base
if you want the replay to check it out for yourself ill upload it for you just let me know if you need it
On April 02 2011 13:41 genghis999 wrote: The Typhon Peaks map uploaded by you to bnet, as well as the one in the .rar on your Google Code site, does not seem to have had the AI imported at all, and thus defaults to the Blizzard one.
Thanks for your great work on this.
Hmm oops : / Ok I will reapply the AI tonight to that map.
On April 02 2011 16:06 Gov wrote: Hey fyndor, I love using your AI! As a gold player, it is extremely helpful to work on my mechanics with it...
One little glitch i noticed was on xelnaga caverns when i was protoss to the north and the AI was terran I scouted the AI rushing me with a few marauders and i probably would have been killed but for whatever reason it didnt come when it should have, when i checked the replay the AI stopped to destroy the rocks at MY gold base
if you want the replay to check it out for yourself ill upload it for you just let me know if you need it
No need for replay I have known about this since I published it. I believe it has code that says that if it encounters rocks on its way to go somewhere then stop to break them. I haven't tried to mess with it yet because it just fell under the priorities of other stuff, but on Xel'naga it is pretty silly and can really slow down any initial push, likely way more than...just going around them.... : /
Ok so to start with, all hail to The Fyndor. Seriously awesome job mate, I'm stuck with a poor internet connection and was getting bored of destroying poor blizzard ai, then I discovered yours while wandering on these forums. Quite an incredible macro/micro, I hardly see high level players being able to split/pull individually a whole bunch of MM while producing out of a dozen of buildings, expoing and upgrading. Pretty funny to see the APM going up to 600 during the battles... However, there is one big issue, it's against speedlings (there may be others, but I only play zvx). The AI will always eventually have a continuous stream of units from their base to yours, which is quite nice for reinforcement, but right when you break through the first "ball", you just a-move your speedling to his base and rape (well, gang-bang in this case) every single unit slowly running away, end up in his non-walled mineral line, and basically win the game. I know nothing about coding, but I would advise it to act a bit more carefully, or stack a bit more their units (or at least implement a stim-retreat for T?) I'm ok not to abuse the AI's weaknesses, but with my speedling fast expo build, I can't really have the extra lings looking at the ramp after I won a battle, the whole point is to make them pay for themselves (otherwise you're behind in workers for nothing, as everybody knows), and this is exactly what I want to practice; but with its system, 1a-move does all the work... Otherwise funny story, I could manage to beat P and Z on very hard after a few attempts (time to catch the AI's early pressure with expo style), and then decided to try a T (my worst matchup). I humiliate his first pushes withs slings/blings super cost-efficiently, manage to kill like 15 workers at his third (11 minute mark, I was a bit surprised as my own third wasn't even finished), and suppose I'm quite ahead... Then it attacks my third, and as I trade my army against his, it starts the trash talk (something like "win ! maybe i'm a lucky AI "), I don't get it and then see a huge stream of units, eventually raping its way to my main... I ragequit and as a good zerg I start crying about imba terran and their effing mules, especially when I see the score screen and its depressing income rate. Well, I guess I now know that cheating isn't only for the insane AI, but after being so boosted by my won against Z/P, that was a damn mood breaker. To summarize, early speedling agression is too efficient, but when it starts to cheat, combinated with early third and MULE C-C-COMBO, it's just a nightmare. Maybe lower a little the extra income ?? Thanks again though, made my week
And a small edit: I read earlier in the thread that if a build doesn't seem to work, it'll pick randomly a transition... Am I pretty unlucky or did you improve this ? eg I go roaches, next push is filled with immortal; i go muta, he starts his phoenixes (and cannon for def <3 ) right away... Very nice, but SO frustrating haha. I play mainly on Shakuras for the record, excellent for "survival mode" (don't harass, just try to get your whole side of the map... really hardcore past the 10min mark. I didn't ladder since weeks, any clue on what ranking it takes to beat it "regularly" ?)
If left alone, it macros WAY better than the Blizzard AI. It regularly beats my friends who are learning to play the game, and in a totally different way than the Blizzard AI (which usually makes 1 decent 1 base push then goes into 1 or 2 base crap-mode). The FYN is really teaching them to put together a push in the early game, rather than just turtle up on 1 base- with cannons or tanks. (because this thing will take 5 bases on them and crush them lol)
I still find it challenging and am currently in Diamond. It can take about 1 in 5 games off me, usually with quick banshees or being on 4 bases while I'm on 3 and playing lazy. Really, the marine splits vs lings, its Banshee use and it's macro are all very very good. You can 1 base the piss out of it, but if you play the "gentleman's" game for the first few minutes, this thing can really put together a game.
The AI on medium was actually quite fantastic it actually played very well and acted pretty human like, and it even upgraded my cheese o meter as it likes to banshee rush..........a lot
I've been playing against this ai quite a lot and as long as you don't do cheesy all in's mass ling, 4 gate etc and don't abuse the fact that it's an ai it puts up quite a good fight. The AI seems to not understand how rocks work properly though, like on shattered temple it doesn't take it's 3rd and on xel naga f.ex it tries to kill YOUR gold rocks before actually trying to kill you, so it works best on maps with no rocks, :p so crossfire/metalopolis and just play a macro game against it it's quite excellent. Just wish it would do some small harassment attacks and it would be awesome. ^^
the problem with the AI (and I use it, a LOT) is that it doesn't do drops / harass.
When I play too much Fyn AI, and then play a real person, I get surprised by how much harass is happening.
It also does not use force fields to segment your army or do ramp defense - if it did, vP would be so much harder.
On the other hand, if you don't harass the AI, it will execute pretty good macro and expand to 2, 3 and then 5 bases, guard each new base with its army, and come try to kill you pretty regularly, including using back door routes etc.
I'm in silver now and have only recently starting beating the AI maybe 2 out of 3 times, but it is miles better at macro and resource spending than silver players so I use it as a way to learn taking safe expansions, scouting, drops and general awareness while also not slipping too much in macro.
I use this AI nonstop, 100's of games played. I'm absolutely dying to throw it on the GSL maps and some others. It would be very cool if there was some repository kept off-bnet for people who want to play on a higher variety of maps.
I used this at Bronze to refine my 4gate and 6gate pushes, then at Silver/Gold to refine my Macro. I can beat it on every matchup on Medium in a Macro game, and it's really helped a lot. I don't 4gate anymore because of the strong macro Fyn has helped with, mechanics as well.
Typhon Peaks seems to be broken. It seems to follow Blizzard AI (the one published by Fyndor). It makes 15 SCVs on medium and stops, letting minerals go through the roof every time.
On April 28 2011 23:33 dunc wrote: Any information on the new version? This custom AI is insanely good, would be cool if you could pick the builds.
Yea sry I have been slacking. On a positive note I have been playing instead of coding so now im GOLD! Lol, but seriously though, the code is so close to being able choose builds I really should just finish it :X. Basically at this point I need to find someones code (can be extracted out of existing mods/maps) that has the ability to "reset" the map and at that point it will be basically ready after some code cleanup. I have plenty example I have seen of maps with this feature (Obs maps for instance obviously have the ability to reset the players bases and the rest of the map). I just need to pick a map with a good implementation that and copy it. After seeing YABOT's implementation of this feature I know I want no part in recreating that wheel. Some code in that stuff just screamed to me that some poor soul probably spent hours tweaking it so they could get the zerg larva to be sitting there ready to be made in to drones as the game starts.
BUT keep in mind even after I add build selection that still leaves you with the builds that are built-in to it to choose from which you may not be all that happy with. The whole build selection project was really just intended as a setup to the build rework project, so that I could easily get feedback about a particular build.
On April 28 2011 23:33 dunc wrote: Any information on the new version? This custom AI is insanely good, would be cool if you could pick the builds.
Yea sry I have been slacking. On a positive note I have been playing instead of coding so now im GOLD! Lol, but seriously though, the code is so close to being able choose builds I really should just finish it :X. Basically at this point I need to find someones code (can be extracted out of existing mods/maps) that has the ability to "reset" the map and at that point it will be basically ready after some code cleanup. I have plenty example I have seen of maps with this feature (Obs maps for instance obviously have the ability to reset the players bases and the rest of the map). I just need to pick a map with a good implementation that and copy it. After seeing YABOT's implementation of this feature I know I want no part in recreating that wheel. Some code in that stuff just screamed to me that some poor soul probably spent hours tweaking it so they could get the zerg larva to be sitting there ready to be made in to drones as the game starts.
BUT keep in mind even after I add build selection that still leaves you with the builds that are built-in to it to choose from which you may not be all that happy with. The whole build selection project was really just intended as a setup to the build rework project, so that I could easily get feedback about a particular build.
<3 It's fine, I'm happy with the current build it would just be nice to practice my Terran without it trying to carrier rush me every game(GODDAMN IT!).
Grats on the promotion at least and keep up the good work!
One of my "WoW friends" got into SC2 and was totally lost, told him to try this A.I and from Johnny Bronze he is now up at 20th something rank in Gold, and that is all due to this A.I which has helped him alot. It can teach new players how to macro up while at the same time constantly being in battles and having to perform micro, Blizzard A.I always attacks at 6-7th minute mark, once fended off, players are too safe and can macro.
i just tried it yesterday beat it on medium and hard on the first try. but, tried to go very hard and insane...i got owned xD really fun playing this ai :D
There were a few things I noticed while playing this AI that if fixed could help optimize it a lot. The AI killing the opponent's gold rocks at maps like xelnaga can really be detrimental to a player practicing against timings against the 10 poo, 7RR, and marauders with shells. Also Ive notice that the queens tend to wander sometimes to a different hatches. Ive seen 2-3 queens on one hatch while the other hatch have none. Finally, Ive seen that the AI limits its unit production in some instances. Sometimes when it hits around 152 food it stops production and just wastes time doing nothing. Finally, can something be done with the early lings suiciding against a walling zealot.
I really appreciate this AI and how it helped me improve my game. These were just my observations. Thanks a lot for this AI.
it seems a bit broken on the new 1v1 ladder maps (doesn't expand and doesn't seem to make a lot of units). basically the same thing that used to happen on typhon peaks.
The GT AI is fixed now with patch 1.14. Fyn Ai is based on GT AI, and GT AI is developed more frequently then Fyn AI now. You may want to try the newest GT AI while waiting Fyn AI to fix.
If I get time I will try and fix but I have a lot of non-Fyn AI stuff on my plate right now. Personally I don't even use Fyn AI myself anymore to be honest. I like the added features of the Macro or Die maps (search "+macro +die" in BNet custom games).
I miss this ;(, this mod made up 60 - 75% of all my starcraft time, online and offline! This made exploring the editor so much more rewarding... please fix the AI, I would hate to see such a good thing go to waste
(and this really is much better than getting my workers punished with some scripted nuke whenever I answer the phone or some shit... I would rather be punished by units at my door than some stupid scripted tragedy...)
On October 26 2011 14:19 XerrolAvengerII wrote: I miss this ;(, this mod made up 60 - 75% of all my starcraft time, online and offline! This made exploring the editor so much more rewarding... please fix the AI, I would hate to see such a good thing go to waste
(and this really is much better than getting my workers punished with some scripted nuke whenever I answer the phone or some shit... I would rather be punished by units at my door than some stupid scripted tragedy...)
I usually play it with that turned off and only turn it on when I think I have stuff down pat enough so that I shouldn't ever get nuked. My main complaint about Macro or Die is that it is only on a few maps.
Edit: I am not totally opposed to updating the maps. I am just really busy at the moment. I have my first kid on the way with my girlfriend trying to keep me busy every chance she can getting stuff like the baby room ready. With my free time I prefer to actually play the game which I guess has paid off, since if you look at my OP it says I am Silver even though by this point I have two accounts in Plat :D I will try to update it sometime soon, but I am not going to promise any certain date because like I said, I am really busy right now and if I make a promise I will probably just make a liar out of myself.
So, the only difference between this and GT AI is that this AI has to scout is that the only difference?
How do I download this? just go to SEA account (I play SEA) and search for fyn and that is all? I would love to play GT AI too but it is so complicated to get it to run.....and I am waiting for a reply from the GT AI maker..