AMAI 2.2

@Zalamander

elseif race_use_fountain and not IsUnitType(u, UNIT_TYPE_PEON) and not IsUnitType(u, UNIT_TYPE_MECHANICAL) and not (GetUnitTypeId(u) == old_id[racial_ghoul]) and not (GetUnitTypeId(u) == old_id[racial_militia]) and neutral_available[NEUTRAL_HEALING_FOUNTAIN] and (not neutral_guarded[NEUTRAL_HEALING_FOUNTAIN] or (neutral_night_buy[NEUTRAL_HEALING_FOUNTAIN] and daytime >= 18)) then ...

should add conditions:
not IsUnitType(u, UNIT_TYPE_SUMMONED)
and
not IsUnitType(u, UNIT_TYPE_ILLUSIONED) [if this type exist]

I think it is meaningless to send a time-limit summoned unit to fountains. Also blizz announced that phoenix would not be healed by fountains in v1.17.

Hope this help!

@peon


I know, i was meaning to regain AMAI its natural ability of choosing targets and attack. Not possible yet.



I just tested it in 2.31, and this limitation still exist. I just saw profile Menace building a burrow when he has about 12 food remaining, greatly affecting their economy in the beginning. This bug becomes worse in teching strategy, i often have seen AMAI wasted his lumber building unneeded farms and then waiting for lumber needed to tech, doing nothing meanwhile.

While some things like how often u attack / make an expansion are opinion-wise, the time when u should build a farm isn't (think at the first 10 minutes of game). I saw lots of AMAI games and realized that a RIGHT TIME exists!
I strongly suggest to remove the "farm build time" customization and set it a default value for everyone, which could be that one of profile Trollus (good building farm timing)


@Zalamander


Tested it furthermore. I saw an AMAI player having a Tier 2 main and a Tier 1 expansion. He wants to go Tier 3. He preferred teching his expansion to Tier 2 then Tier 3 instead of just upgrading his already Tier 2 main.
Never seen this behaviour with 2.1, so i think is a bug.

Never seen Blizzard AI teching at expansion too


Bye bye and thanks to all AMAI "crew" and forum people

Yeah that teching both bases is odd I just saw it as well after i fixed the hero skill bug, never seen it before in my life, we don't control what hall they should upgrade much I think they only get order to get a hall2 if they don't have one or a hall3 if they don't have one. We will look things up and have a great bugfix time when andy comes back I guess. Most of these bugs you ll mentioned here I know and will be fixed, btw I think they go back to normal behavior if you just order the same restriction again in the commander, it turns off the same way as it turns on for what I know when I have been using it.

Bugs I have seen that can be fixed is at least 2 heroes TP at same time and probably the multi hall upgrading.

Never thought about that, i'll try it and tell you if it works.

Thanks for the answer--


>=-Bye-=<

This AI is very nice, keep up the good work!

Here are the problems i noticed while testing with a normal amai v2.32.
1. After creeping the AI often leaves items around.
2. One NE AI moonwelled itself into its own town
3. AI sometimes TPs after opponet has already TPd
4. Two TPs by one ai in the same area
5. ai can't decide what to do: attack opponet...creep site A...creep site B?(wastes time walking back and forth)
6. starts a job(creeping, attacking) then leaves...similar to 4
7. chose to expand right next to opponet instead of a convenient location right by base(wasn't creeped yet but easily could have been)
8. attacker often TPs to early...can do a lot of damage but instead leaves and lets opponet build back up
9. one team is on attack...is losing battle, but its ally is about to arrive, yet it still TPs right before the ally arrives which would even things up
10. often not enough base defense...not even one or two towers, i may be wrong about this one but i like a few
11. NE AI often seems to take a break and not spend its gold and it becomes very passive building only a few military units
12. the ai often fights on hopelessly in base(or expansion) and instead should just send all of its units to its expansion(base), for it is only wasting units and time.
13. related to 12...ai TP'd to expansion that was overwhelmed with opponets and had no chance of saving.
14. in a team game....one ai was completely dead except for a hero....this hero hardly did anything...i.e. it didn't help its allied ai.
15. my bro sometimes jokes around with this funky strat: as humans he builds a ton of towers and a bloodmage at the beginning and sits in base and techs until he masses Gryphon Riders, then he procedes to destory the ai fairly easily. maybe better air/ranged countering system needed. also the ai shoulda built a bit more siege for the amount of towers to units ratio.

Also i was wondering about the insane AI: can anybody beat it, if so are there recorded games. Does the insane AI cheat in other ways besides double gold and wood? And does it have any improvements over the normal ai besides its cheating?

I wrote to newsmaker of www.ag.ru. And they write in news about AMAI. Here is my own simple translation (site has poor autotranslation only).
Article names: "New brains for Warcraft III"

Quote:

Many people like to criticize strategic intelligence in modern RTS, blaming it in cheating & unnatural behavior for human opponent. Exactly this problem decided to eliminate two enthusiasts, released free modification Advanced Melee AI (in abbreviated form AMAI) for Warcraft 3: Reign of Chaos и Warcraft 3: The Frozen Throne.

Author is newsmaker himself (besides that string about P4 2.6 - it he takes from my letter, and other my comments he put in his news too, but unfortunately he ignored some of my important arguments for AMAI).
In their forum appears a thread about AMAI. As I expected, it is full of lammers – I'll put them in their place.
That site www.ag.ru on average has 14 000 unique visits per day.

1. Nothing we can fix, AI is blind we can't detect items and order them to pick them up. :/

2. Building placement is also totaly rtandom making NE computers build them self in very often since they don't use up the trees and make additional room that way, I made NE always buy a shredder to havest away the trees and make more room for buildings as a small fix to the problem.

3 - 4. We have some issue with it but might be hardcoded, hard to tell since there is a hardcoded TP control as well that we can't shut down and replace, we only have our own TP that reacts earlier and to more different situations than blizzards hardcoded.

5. This haven't been much of a problem lately unless you watched a teamgame with computers in same team, the hardcoded AI always override our orders and send them to help each other all the time. :/

6. A hardcoded problem appearing in 2 cases, If it's early in the game and the computer has only a hero and maybe one unit ot no unit it maybe kill one creep then goes home, and also a hardcoded bug when it succesfully stuns all the creeps it seems to think they are gone so it goes home again.

7. Happens sometimes, the fotmula create chanses that it picks different expansion sites, on some maps it might be 10% chanse that it pick a expansion by the enemy while it's 90% chanse that it expands near own base, so it can happen, Im however going to make it expand closer to home more "all the time" by changeing the formula a bit.

8. Yes a problem when the hero gets to hurt it flees with TP, we are going to make them send away the hero by foot and continue the fight with the units and other heroes more often.

9. Also connected to hero fleeing when it gets hurt.

10. Since AI has stupid hardcoded building placement the towers usualy ends up at stupid locations where they do no good, better use the resources on units instead since they are better used than towers by computers.

11. Maybe it switched to some teching strategy so it waited for a upgrade to be completed so it could spend the money on the things the new strategy was going to use.
Or you noticed the upkeep resistance, all computers save up to 500 gold before entering low upkeep now and up to 2000 gold before entering high upkeep.

12 - 13. Hardcoded problem again, they always try to save whatever is under attack no matter what it takes, they just send every unit until the enemy is gone or what they are trying to defend is lost.

14. Not sure how a AI works in these stuations, they so rarely occure.

15. It does build more siege if enemy build towers and build ant air if enemy build air, unfortunatly it's not enough, the AI can't see ahead of time so when the gryphons start appearing the AI probally just have got his anti tower army ready. Also they are very bad at defending towns from air units. It's very easy to beat even a insane using a strategy like this.

Insane usualy don't have any benefits beside from that dubbe income no.
In AMAIs case the insanes are allowed to use any strategy as starting strategy and they are also allowed to change strategy faster, just becuse this would cause economical problems for normal computers it's limited there but no reason to limit it for insane. AMAI at insane skill is much much harder to beat than normal computers at insane skill becuse of AMAIs dynamic building making it easy for them to make use of all the extra resources. I made a test game and 1 AMAI Insane actualy beats 3 Normal AI Insane in a 1vs3 game with ease.


Nice to see that AMAI at least get some attension. I noticed that it was growing very popular on russian sites shortly after it had reached popularity on the chinese sites, maybe we should do a russian translation for it soon.

IC, well thanks for the reply.

I figured many if not most of these problems would be hardcoded. I was wondering though as blizzard updates the game are there often changes to the code that allows some of these things to be fixable?...i often here of new functions from patch to patch.

And i too noticed the teching to tier 3 at the main base and expansions....this really needs a fix and could give the ai a solid overall improvement without all this extra resources being spent.

Also, could you just give a little run-down of what we might expect to be changed in the future versions of amai.

The 1.17 beta 405 just came out...will you be releasing a new slightly updated 2.32 to optimize for this?

Thanks!

There is plenty of bugs in the AI tat have been there since the release of TFT. Blizzard just don't care at all for the AI. I have already tried to ask them to fix bugs like the NE don't send any wisps to mines if they expand on a mine that was already entangled or hunted before.
They even have very simple bugs in theire own AI script they don't fix, for example the Orc TFT AI tries to build "Berserkers" after they have upgraded the "Berserk" upgrade (The one giving additional HP to grunts), they check for the wrong upgrade. And they check in the wrong order as well they try to build berserkers if the upgrade is "not" done and head hunters if it's done.

This error has been there since the release of TFT and still are.

I think the teching have something to do with something that changed in the AI. it must be it, becuse I haven't seen this bug earlier. So it will be fixed.

What to expect in next version of AMAI:
Release date: Shortly after the official 1.17 patch is released by Blizzard.

Bug fixes:
Hall teching bug fixed.
Hardcoded Team help problem maybe solved(I got some ideas on this one)
Dubble TP usage bug hopefully fixed.
Better expansion choosing.
Most "take a break" bugs hopefully fixed since they are most likely connected to the currently bad heal army function that is in need of improvements.

New features:
New improved army/hero healing.
Zeppelin drop harassment.(hopefully)
Faster dual creeping.(maybe)

Some new supriseing game modes for AMAI might be in order.

Don't really know what kind of new features there will be yet, it's based on how long time it'll be until Blizzard release 1.17.

All features that 1. I think will improve gameplay for the computers and 2. AIAndy thinks is easy enough to code and won't use up too much CPU-power in the AI, will be added if we get enough time to implement and test them.

Now to the current situation on 1.17 beta, I don't think there is any need for a new 1.17 beta version of AMAI since there was no changes at all that AMAI don't automaticaly adapts to already.

I stumbled across this place by accident, I'm fairly new to WC3 and haven't finished the campaign yet, let alone played the AI in a skirmish. However, this all sounds very exciting and the second the Blizzard AI falls with ease, I'll be taking on AMAI.

Having dealt with AI in Total Annihilation, I know what a pain hardcoded elements can be, often there simply to deal with inadequecies in the scripts. It's sloppy, but hey, that's developers for you :)

I hope you have some luck convincing Blizzard to at least let you disable hardcoded elements. I have no idea what their community support is like, but from the above it doesn't sound like your chances are good, but here's hoping anyway.

Thanks for your hard work guys, it's much appreciated by those of us who love AI gaming :)

Zalamander, I think it was you who advised me not to uninstall AMAI from a map and then install a new AMAI version, but rather install AMAI to a fresh copy of a map each time AMAI is updated. Well, I messed up and failed to copy a map that I altered. It would take some work to re-do the changes I did to it (although I can do it, I'm just lazy). Could you give me more details about the downside of continually uninstalling and reinstalling AMAI on a map? You had said that AMAI isn't fully uninstalled or something? Is there any way I can manually go through a map and remove all traces of AMAI so I can keep that map as a fresh master copy again? Sorry to bother you with such a dumb question.

The uninstall is prettyy useless, the only good with it is if you want to "disable" AMAI from the map so it will be blizzard orgiginal AI next time you play with computers on it.
The only thing the uninstall can do is hide the AMAI files in the map so that war3 can't see thwem and use them any more, they will still make the filesize of the map bigger.

The best and only way to really clear them out of the map is to dubble click it so that it opens in World Editor then save the amp over it self again when it has opend in WE, now it's very small again in file size and completely clean.

I my self just install new versions of AMAI on the maps over and over again until the maps is to big in file size then I jsut delete them and make fresh copies from the original map folders.

I think we will make a wardraft-version of AMAI soon as well making it possible for you to start war3 with a alternative .exe file and then when it's started with this file the computers in the game will be AMAI no matter what map you play on.
However there is problems with this connected to multiplayer, since AMAI is updated so often all users need to start war3 with same version of this AMAI executeable. another problem is that it can't be spread since the AI isn't in the map with this methiod, other players joining a game in multiplayer you created with AMAI usualy will get the same AI by downloading the map you created on, this won't work with a wardraft version I think.

You think, ur right. Players can only join the player if they are running the same executable. Without it causes issues as players running different ai files, so its likely it would crash.

Are there similar efforts for WC2 or SC, or is WC3 unique in the level to which you can customise the AI?

Does AMAI effect campaigns or only custom games?

Very much more customiseable in war3. AMAI would be impossible to make for WC2 or SC.
Oh and I forgot to thank you for your last post as well. :)

I can say that it's almost possible to do anything in the War3 AI, the only trubble is that most of the hardcoded AI can't be replaced, unless you want to disable it and replace it totaly but I guess you would need a 50GHz CPU to run a AI made in JASS script that replaces the hardcoded functions of the current AI. :)

The AI is coded in JASS. it's a simple but very usefull script/code language.
Get some info about it at http://jass.sourceforge.net if you want to know more.
What else there is to know about the AI is that it works in up to 8 threads I think, per computer player in the game of course.