DarkBlizz

Quote from: Jarpoir on March 11, 2010, 02:05:49 PM

as wondering if another cheat AI version could be added, in which the AI doesn't see all of the map, but still yields a much higher mineral bonus?

using this modifiction by darkzero, you can customize the cheating easily.

http://darkblizz.org/Forum2/ai-scripts/ai-standar-(proposal)/

one of the launchers almsot does this.  requires an AI folder with subfolders that are names of AIs with a base.sc2data in it.  Then, presumably copies the one you selected into the right sc folder b4 launching game.

It should be fairly straightforward to select randomly from those folders. I'm not sure I'd use the feature though.

Quote from: turdburgler on March 10, 2010, 12:14:21 PM
http://i43.tinypic.com/55ghds.jpg

I hope you also consider reacting to expansions and sneakiness, and a proactive flow loop where you resort to some canned bags of tricks/sneakiness (drops, dt, nydus, prism, reaver harass) that aren't always part of build, but trigger semi randomly.

The reactive system seems like a strong alternative to the AI not sending wave after wave when it is getting raped on each wave, though I think casual players like facing waves. 6.1, I think is pretty good at understanding that it is losing, and not sending another wave if it is.

6.1 does a good job with core army makeup and progression, but it could significantly increase challenge by forcing the player to multitask more with sneaky play.  Even if the P AI knows it can win by mass carriers after it is ahead, it would be more fun for the player to try a special/sneak wave on them, at least some of the time.

Improvements that I would prioritize ahead of a reactive system, and ahead of some of my suggestions above are:

*use more varied units (sentry, infestor, warp gate, ghost, roach burrow and reaver cliff harrass) - though I understand that some of these units micro abilities (probably reavers hardest) are very hard to implement, some like guardian shield are easy, and emp or infestor abilities are not any harder than storm.
*more variations on canned opening strategy and progression.  There are other valid openings than 3 gate for protoss, even if that is the one the AI would choose more often.  I know people complain if the AI doesn't make the strongest possible choice, but variety is long term strength, and more fun, and more useful practice.
* Active income analysis:  The AI currently puts about 30 workers on 9 mineral patches, when the gains from having more than 18 are minimal.  If it actively monitored the income it was getting from additional workers, it would dynamicly react to any cheating handicap it receives.  More importantly it can calculate how many production buildings it needs to keep queues full, and build more production buildings if it has the income, or consider expansion if doesn't or dynamicly build farms based on income/production levels.
* little things like queens at the right expansions, spamming chrono boost (even without worrying about being smart about it), creep tumors and so on would also boost strength more than I imagine a reactive flow system would.

(though reactive flow sounds like it has great potential for dealing with wall ins, or the AI deciding to turtle in itself, most of that could be 8.0 ideas )


btw, even a small 10-15% cheating mineral boost makes the AI very rush resistant.  PvT on metalopolis is pretty hard for standard early P play.

default replay speed is always 2nd fastest, but battle.net speed is fastest.

sorry about the bad report regarding breaking replays.  I had unrelated extra file issues.

Quote from: turdburgler on March 09, 2010, 04:55:25 PM
I'll back use of something like this. The exact specification has to be debated.

DarkZeros format is pretty good.  Maybe a mini-primer in heading.  Using pure galaxy code makes it easy on AI designers for providing extra customizability

That is excellent.

I love how it shows how much it is cheating as well.

Its actually not too hard to edit or screw up
the format is
property = setting ;

You should only edit the "setting" part.
always leave the ";" afterwards
"//" means comment rest of line
any setting that says "false" has the only valid alternative "true"
if a setting is a number, only change it to other numbers


It looks like this mod breaks replays though. 

v35,

setting 3 players to human still loads 3 AIs
deleting the text in player slots 3 and 4 still loads 3 AIs
someone said they played a 1v2, but I have no idea how they might have accomplished that.
Implementing a disabled option would be clearest.

Its confusing, because there are 2 requests.

the first, live APM tracking is best provided by blizzard, or at least after they release the editor.

The 2nd and key request is being able to analyze replay files.  If you can get APM, you can also get resources, harvesters, army size, score and custum stats like score per minute.  The value of such a tool is that you don't have to load and play through a lengthy replay for just the summary or chart info .

on rev 34,

twilight fortress in your attached map pack is setup as 2 player only???
only loads 1 AI players.
WORST PROBLEM: Only spawns (7 in a row anyway) both players at bottom of map

trying replays (of 6.1.1 cheating ai) causes crash in sc2 in win7(no error number).  no problem loading the replay in original lazy launcher.

To adjust difficulty setting on cheating AI, the quick version is (tested):

in MeleeAI.galaxy, about 20 lines from the bottom, there is 3 extra lines in the cheating AI over the regular one.

    AIHarvestBonus(player, 2.0);  //sets mineral boost. 1.0 is no cheating. 0.8 is 20% easier
    AISetDifficulty(player, c_diffNormalVision, false);  //sets maphack
    AISetDifficulty(player, c_diffLimitAPM, false);  //unsets any apm limit on AI

I've set mine to:

    AIHarvestBonus(player, 1.2); //20% harder
//    AISetDifficulty(player, c_diffNormalVision, false); //no maphack
    AISetDifficulty(player, c_diffLimitAPM, false);

There are other difficulty settings just below, but I wouldn't tamper with them.

If AI designers would standardise on the player variable,  and the c_diffNormalVision and c_diffLimitAPM contants (they prob do, as that was in blizzard original I assume), and standardise on placing it in the MeleeAI.galaxy file, then launchers could adjust the cheat amounts and flags on the fly while loading game.



this tool will let you open up and extract or replace individual base.sc2data files:  http://www.zezula.net/en/mpq/download.html

Quote from: Lenor on March 08, 2010, 01:52:52 PM

Yeah tudrburgler,will you make a 1.5x harvesting bonus AI?It would allow us to choose the AI more adapted to our levels

Thanks for 6.1.1 BTW

sc2launcher 2.2.2 comes with a triglibs version (unpacked) of the 6.1 AI.  But I cannot find any code that sets what cheat modes are enabled/disabled.  If we find, then should be easy to tweak exact handicap to AI

btw, the AI is stronger and more varied on 4 player maps. like bw.  (probably 6 player ones too)

The twilight fortress 3x map has FFA settings.  Its layout has 2 very close bases.  So games on this map always starts with 2 1v1s then the 2 winners facing off after.

It turns out to be easier than 1v1 for some reason.  On cheating AI, it appears impossible unless you go for immediate military buildings, and its still not beatable by me.

Do the other FFA maps with 3 AIs also tend to have 2 close AIs attack each other, while the other AI focuses on player?  My vote is that this pattern makes the game more interesting if FFA AIs would pair off like this.

This is harder than 5.5 for sure.  Very challenging and fun.  Especially if you don't just mass up in your base and wait for it to come to you. 

I liked the tech builds and unit mix in this one.  6.0 forces me to be aggressive/turtle up a lot of units, so this is a nice variety of playstyle.