Feedback and Test Results: AMAI 2.50 RC 2

Use this thread to consolidate all feedback, test results, and accompanying replay files in this thread.
The version scope for this thread is only for AMAI 2.50 Release Candidate 2

awesome!! ima glad to see continueing improvements on this project. being a dial up user as i am -___-;;; i <3 AMAI keeps playing AI just that much more fun. keep up the good work. i jsut downloaded and once i get some sleep i will leave some feedback of RC2.

Does insane AMAI cheat like the insane Blizzard AI does? Because I couldn't do a thing against it, there were just too many of them. Actualy, after loosing the first battle to mass undead casters I decided to try with some more demolishers and it actualy seemed like I was going to win the battle once we finaly fought, but then 4 statues turned into destroyers and it was gg. :( Also, the AI somehow knew I had an expansion going up?

Well, it could also be that I plain suck. Haven't played melee in ages. I'll try some more games vs normal AI later.

Yeah insane AMAI has double resources same as insane blizzard ai and so this cheating combined with amai becomes really hard to beat.

Well, I just completely defeated a normal AMAI a couple of times, the attached replay being one of the more embarrassing losses for it since I didn't loose a single unit.

The problem is that the AI couldn't deal with some early hero harrasment, lost an archmage to a DH when it had a tower nearby it could use for cover, later let itself be creepjacked and continiued to try to fast expand when it was already obvious that it wasn't being particularly fast about it. Also, no rifles vs hunts, why not? Instead, it tried to tech to gryphs which was stupid without a secure expansion. The AI simply doesn't seem to be able to pick another strategy when the one it's currently using doesn't work. Fast expand definitely needs more contingency plans.

Edit: Another game of humans vs orcs, the AI tried to tower rush but failed miserably, it canceled building construction the moment my units touched the constructing towers, but kept trying to build new ones. It tried to keep with the strategy long after it had already failed, if it can't get a tower up by the time the enemy discovers it and attacks then it should abort the strategy.

The problems are the AI is just blind and can't react or micro in any way with use from the scripts we do.
It's impossible to make a AI able to handle a hero harass or to see what's actualy going on and change strategy at once. Sure the strategy changing can be more senitive but the result is often horrible it changes strategies all the time than and end up getting nothing(Have tried it).

You see the real AI that lies in the bottom the one that does all the actual unit movements and attacking and skill using is "hard coded" as we say and can't be modified at all. The scripts can only control it to some extent but not in realtime it sends new status updates and orders every 2sec or something depending on the priority there is on some orders.
Shoulden't be hard for you to understand it can't handle a hero rush if it can only get a snapshot of the situation and make a decision every 2sec. :)

One other irritating thing that is completely hard coded is the building placement when base is contructed, only for some special things like direct orders is it able to set locations for thoes towers. And AI scripts cant see wood at all so there is no way to detect the forest edge and make it build the lumber collection building near there.

@Anitarf: As Zal said this is also a problem why tower rushes, ancient expansions, millitia creeping are really difficult for the amai to use effeciently. So many different odd situations can occur with such humanly tactics. More so tower rushes as its difficult to detect when it should give up. In some situations if it had not given up it would have won but gave up. Its a very difficult thing to control, but i'll look into the code and see if such a check will work for more favourable conditions.

@Zal: I have actually managed to make lumber collection points build as close to lumber as possible using the autoharvest and is in current RC. A hidden unit autoharvests at start location so after a while we return position of the hidden unit thats harvesting wood. This returns a location to build lumber building as close to wood as possible. Such abuse to the systems :P but it works on all standard maps :D

hehe nice one. I have test reports from my friend WargH mentioned on site btw.

Havent managed to get hold on you to tell you about them but he got error messages AI saying it failed to use neutral hero tavern at some maps.
He also found a kind of abuse, he plays night elf and hides a stealthed unit in the computers base during night this will make the AI stuck at home since it can detect enemy units in base even tho it cant see or attack them.
Anthing possible to do about this perhaps, either ignore the stealth units or use any means available to detect them and kill them.

@Zal: Thats odd i'm sure stealth units supposed to be ignored already. I'll check it out.

Hello, Strategy Master. I ran a RC2 AMAI tournament, and the result showed:

Using 3 official maps: Turtle Rock, LostTemple & TwistedMeadows. The rule is something like WCG World Final. Any player who win 2 games will win his opponent. But each score are recorded. All AMAIs are set to normal mode (not insane).

Well done!!! There are no obvious bugs now, and it seems that Night Elf has been improved. And I saw Orc's tower rush although it failed in the end. I'm interested in future release.

Hi Strategy Master:

I've been testing AMAI RC2 in the spare time, and I haven't found problems, the strategies are great, and the AI behaves smoothly in general.


but...

I think there is a serious bug with the Zeppelin usage, AMAI is just stupid using Zeppelins, even worst, it loads the units but it get stuck, the units are not unloaded.

Here's my replay.

Hardcoded AI usualy interfere with zepplin control in some wierd random way some times. It just pickup units without reason or anything and we have yet to figure out and logic behind hardcoded AI zepplin control for what I know.

Some times it do manage to get units to a target on a island location but then it's usualy very buggy and one way trp for the units since the AI in war3 works with all units "the army" as one big unit things goes bad when a few units gets separated from the heard.

The map Desert Strife in question is not suited for AMAI play due to the way the pathing system works. The zeppelin usage by AMAI is just to save near dead units but not for navigating to un-reachable locations by foot and is most likely that behaviour was hard-coded blizzard zepplin usage.

I have interesting pathing projects in development on another mod which if works can be moved over to AMAI to hopefully enable play on these types of maps where player starts enclosed somewhere.

Actually I'm working in something similar, it's a for a project focused in naval fights, and it requires that the AI use boat to transport units to the enemy island (something still impossible :P). I checked AMAI on this map by purpose, because even the harcoded blizzard AI management of Zeppelin transportion is in general buggy and unreliable.

If you add this as a AMAI job, it would make AMAI the best of the best. I'm doing this work in my few spare time, so if I get something, I will inform you because it can be quite useful for this project.

@Moyack: Sounds great. More people trying to solve this complication will help us work out a solution faster