DarkBlizz

I'm pretty sure this goes over most people's heads (at least the beginners), but don't forget that you guys are playing normal speed as default.

I hadn't realized, but the pace of the game is ridiculously fast when set to 'faster'.

Anyway, thought I'd bring it to your attention.

Quote from: [mrn] on March 08, 2010, 04:30:21 PM
Nope, but I know my grammar.

lol, o rly? go brag at school, kid, honestly.

a little on the zerg ai:

- base is still not fully saturated (drone count at 14)
- first 22-24 lings sent on a suicide run (i had bunker up w/ 4 rines, 1 rine and 1 marauder outside. instead of attacking the bunker, perhaps they should either sit back and wait to ambush my troops or head str8 to my mineral line and take out scvs)
- around the 5 minute mark, they have an abundance of minerals (700~) and larvae that are not being used (8 larvae)
- they then throw 2 extractors up (not neccessarily great b/cuz only one is fully saturated and the other one just sits there - also their drone count is low so extractors lower mineral intake), a roach warren, start producing some drones and lings (which is good, but by the time the warren is up, they have no larvae to make roaches asap). also, they upgrade their hatchery to lair (good idea as i was going for banshees)
- at the 8 minute mark, they are now near saturation and are consuming minerals at a good rate. they decided to expand, but didnt have defence to stop my attack (only had some lings vs my small group of marauder, hellion, rine)
- they never use spine/spore crawlers (could be good thing, but only if they are spending the resource)

that's it for now. banshees ripped apart base cuz no defense ^^

neat stuff. im having a brain fart tryin to look over it tho so maybe u can help me. what are some conclusions you've come up with in terms of best units relative to resource required, building time, food, etc. or maybe u can just compare DPS relative to resource requirements, but then u gotta consider other factors too but just to dumb it down a bit, what have u determined from ur chart?

colossus atk range is 9 w/ upgrades i believe
btw, why r colossus good? do they do AoE dmg? i thot they do from their atk animation...

Quote from: katalonec on March 04, 2010, 05:22:58 AM
is this starcrack ai really "crack"? it becomes much more than small twick/hack "tool".
is there any chance to make this ai official and maybe then rebrand the title from "crack"?

i believe it's the dev team's name lol

How do you guys come about implementing suggestions? Is it primarily through this site? Other sites/networks/groups? Private testing?

i think what could happen is analyzing reps the way bwchart did and implementing the most common builds w/ accurate timings down for buildings. but this would of course be the the first few buildings in the BO as everything else should be responsive/adaptive to opponent. then, u could expand on building placements, etc.

Quote from: ElvishSolution on March 04, 2010, 12:08:20 AM
Well the other thing is you have to know they are going to have a near unbeatable attack at about 8 minutes in, so you need to build strictly defense until that happens. 


For Terran 3-4 tanks is going to take out pretty much anything they throw at you until they get air. 


Zerg will probably have the hardest time beating this, and I've not tried with Zerg so I can't say for sure. 


For Protoss what I did to stop the attack was build a gateway and a pylon at the only choke point in my base, put cannons on the cliff faces, and by the time they attacked had 6-8 stalkers, drop expanded, got void rays and phoenix(es?) not sure the plural on that, walked into his base. 


I'm sure it would seem nearly impossible for many people and for a lot it is, but its mostly experience.  SC 1 I would do 8 player Melee 7 comps vs me, to practice my defense and fast macro, so I'm used to it.


The AI is still very much a work in progress eventually the goal I assume is to have 5 or so difficulty levels so that brand new RTS players, and original SC players who have been waiting 12 years alike can enjoy playing vs AI.

i think it would be more challenging to build normal builds (instead of cannoning up) and wall off if possible. at least for me, i try to emulate how i would play an actual player. tha's always a good test. i havent tried this cheating version yet, but will soon.

btw, if u want to be pro sc2, start practicing 15 hours a day or forget it, lol. what a life huh? trust me, being a competent player is different than pro and takes a lot of dedication and time away from real life. i would know personally.

Quote from: sLayed on March 04, 2010, 04:34:24 AM
I've played over 10,000 games of starcraft and I'm a semi-pro SC gamer so this new AI is exciting me quite a lot.  The ones we've got now I beat while sleeping. I can literally micro 20 lings and 5 roaches for 20 minutes vs the ai (ofc at which point i've got like 5 expansions and full tech ).  Why don't you just release the AI and update it after? I'm sure 90% of the people here would rather have this artificial goodness ASAP over the finished version at the high quality you're seeking. Just call it like 5.9 and when it's where you expected it to be, release it as 6.0.   I know this makes the jumps smaller and less awe-inspiring but I don't think that's what you guys are going for .


It's fine with whatever you AI dev's decide, thanks for the great and speedy work.  I guess if I've been waiting 5 years for SC2, I can stand to wait another few days for an improved AI for it .

5 expos full tech? how do u macro effectively? it's hardly manageable w/ 2 fully saturated bases as i've seen in multiplayer reps. also, how do u manage to get 5 expos? were u testing out the tech tree? btw, whats ur apm?

Quote from: ElvishSolution on March 03, 2010, 09:52:41 PM
Still very beatable, at least for me, of course to expect an AI to ever give an experienced SC (and RTS in general) player like me any real trouble is basically unheard.  For casual players I'm even v5.5 without cheating is a challenge.  Probly should look into difficulty levels at this point.

yeah figured it was beatable 1v1. i guess we are similar. not that i'm an rts player, but im a quick learner. i like how things are coming along. hope ai can give me a run for my money soon w/ the release of v6. turd should definitely take a look at this version to imrpove his, but i think he's already implemented some of the imrpvoements.

Any decent players here tested this yet? If you can win in 15mins or less, it's still marginal in real gains at least for ladder training, but any gain is a good gain.

awesome news!

now having said that, i hope that you've gotten the time to address the micro of the units (ie. the scout/army moving back and forth when they see the t1 units/stronger army).

i hope that the new early/mid/late BO's will be able to mesh well together. again, i believe the early bo's are highly important so it is best to just implement std efficient opens (so we can't overrun with our 1st wave vs AI. coupled with an efficient open and decent micro, they may be able to withstand the 1st rush which would be a DEFINITE milestone) and add some variable techs (ie. immortals, banshees, etc). the bo's should be cohesive enough and implement good unit mixes rather than random. i'd love to have a.i. that could counter our unit mix (probably hard to implement).

another thing to implement is the use of warp gates, phase prisms, proxies, nydus worms, burrowed roaches/infestors, medivac+thor drops, etc. as a.i. improves (some may already be implemented? ive yet to see anything technically astute abt the a.i. as i've never been past 12-15 mins, however, blink micro is awesome!).

i believe the current a.i. don't employ overseers or at least dont micro them well? observers are pretty key too and as a.i. gets better, perhaps that can be one element to aid a.i. in countering our unit mixes (ie. we have too many roaches so comp makes immortals, etc).

the cheating a.i. should be adjustable, b/cuz strats that distrupt mineral lines are vastly undermined.

w/ total saturation, will a.i. be able to keep up macro-wise? i suppose the easiest way to good a.i. are improvements to macro. that way at least they have the unit count and can rely on the chokes to hold ground. if they learned to wall w/ T that would be gosu if that were somehow available (map dependant).

btw, if a.i. can repair structures as stated above, that would be key. also the use of the chrono boost, mule, inject larvae, as well as the non-traditional macro skills that each race possess (ie. the structure heal and creep tumor from zerg, etc). again, may have been implemented, but just want to help a.i. macro-wise.

Quote from: itsarabbit on March 03, 2010, 02:14:15 PM

Quote from: r3ddrag0nx on March 03, 2010, 11:25:13 AM
I would be up for distributing my beta key when I do get it, but preferably for some amount of money. But seriously, if anyone happens to get a key, do try and talk to LazyTown or any of the various teams as they'll need some to experiment with.

You shouldn't tell others what to do just because you don't want to. It's ignorant and selfish. Although I do know that you want to be able to play with the beta key, its better to just shut up about it.

ur such a clown. that wasn't what i was trying to say. i'm saying whoever gets it first should send one so they get more time to work with it. if i got mine early, i'd do it, just rather get some sort of compensation or lien on my money.

The best way to review the bot is watch the game you just played.

Anyway, some things I noticed about the Terran AI:

   
• they dont saturate mineral fields (builds 15 scvs and stops for a good while - saturation is at 24)
• the unsaturated minerals results in an imbalance w/ overabundance of gas b/cuz u programmed terran to dual gas early
• instead of building the 3rd rax, perhaps it can tech up? i think the problem in the a.i. lies in optimal build order. if that's the case you should program the standard openings for all 3 races.
• as stated before in my previous post in this thread, the constant back and forth of a.i. army is ridiculous. i've noted that you should program them to retreat and stay put for a little while if the difference in army strength is larger by x amount. meanwhile, they can build their forces. if difference is minimal they should just stay and fight/micro and re-evaluate the strength difference in intervals or something cuz they seem to be constantly evaluating army sizes compared to their own and fog of war fucks with their perception.

I can only provide feedback on early game as any mid game and late is redundant if early is not corrected. little mistakes equate to bigger problems. plus, i'm not too familiar w/ late game anyhow.

btw, turd if u need any testing, build order, etc. help, please let me know. i think i can be of some assistance in that aspect in the same manner that david kim is helping blizzard test in-house.

oh another idea is when the a.i. encounters an abundance of unit x, they should pump unit y (which is the counter to x). they should probably start off with the first possible counter they have available and tech to the best counter if there is an overly large amount of unit x. higher tier units should be weighted since they are naturally stronger than lower tier units. for example, 1 immortal equals x zealots to put it plainly.

UPDATE: upon playin another vs T game, I think a.i. shouldn't pull back his units to defend his base (the ones of his in my base) if his army strength is bigger relative to mine meaning that he can rip apart my base faster than i can his. i had a game where he retreated, rightfully so, but then it occured to me that perhaps if my army was small, he would have retreated too which is bad.

perhaps we can make the T a.i. have the ability to wall off? somehow detect narrow choke points in his main in the early game? perhaps then we would have map dependant a.i.?

 cool