Matrix Games Forums

Armageddon invasion starts in 2014Command Rommels Panzers in Battle Academy!Servers UpdateThe Deal goes Fourth!Command Ops gets a Massive Update!Lost Battles AAR: Smolensk 41War in the East: Lost Battles is here!Conflict of Heroes: Awakening the Bear! Let's Play videoLost Battles Scenario Manual excerpts available!A sneak peek from Allied Corps!		 Forums  Register  Login  Photo Gallery  Member List  Search  Calendars  FAQ 

My Profile  Inbox  Address Book  My Subscription  My Forums  Log Out

AI decisions at Pearl

 
View related threads: (in this forum | in all forums)

Logged in as: Guest
Users viewing this topic: none
  Printable Version
All Forums >> [New Releases from Matrix Games] >> War in the Pacific: Admiral's Edition >> AI decisions at Pearl Page: [1]
Login
Message << Older Topic   Newer Topic >>
AI decisions at Pearl - 3/9/2010 12:37:58 AM   
Jim D Burns


Posts: 2730
Joined: 2/25/2002
From: Salida, CA.
Status: offline 		With the new patch, I just restarted a grand campaign playing as allies vs. the AI, and I find myself frustrated over the AI decision to linger near Pearl. I tried not to take advantage of the AI, so I left all my bombers stood down and only sent a few subs out to try for a lucky shot.

Finally by the 10th, I sent some PT boats north in frustration to try and shake the AI into action to leave the area, which appears to have gotten them to move off a bit to the northwest this turn.

The net result is this by the 11th:




As you can see KB’s air groups are now chewed up pretty thoroughly, for a small cost to me in air to air losses of Pearls P-40s.

The AI needs to be able to somehow tell when its air groups are too fatigued to go into harm’s way. With all its attacks after Dec. 7th, the net result is it sank one extra BB and another sub and damaged some more ships, but at the cost of a huge chunk of its elite pilots.

I know this is more of a rant than a bug report, but is there any hope of ever seeing the AI be able to judge the cost effectiveness of its attacks? If not perhaps the AI script that forces it to loiter off Pearl for weeks should be changed to just an extra day.

I did nothing exceptional here except place my surviving fighters on 30% CAP at 15k after the historical surprise turn. Even though my fighter groups had fatigue in the 30s or more and morale far below 50, they still managed to shred KBs strikes on the subsequent days, probably due to excessive fatigue for KBs air groups.

Had I tried to be more aggressive I’m sure I could have bagged a CV or two by using my bombers and sending surface ships out on night attacks, but I held back hoping in vain KB would see the light and leave.

So Andy, any chance you’ll nix the weeks long death grind for the AI?

Jim




Attachment (1)

_____________________________
Post #: 1
RE: AI decisions at Pearl - 3/9/2010 1:25:14 AM   
Misconduct


Posts: 1777
Joined: 2/18/2009
From: Cape Canaveral, Florida
Status: offline 		I had the same exact situation Jim D, the KB was sitting around Pearl for so long I sent a few subs out and in one of my previous threads showed the results: sank 2 Cv's of the KB with 1 sub.

(in reply to Jim D Burns)
Post #: 2
RE: AI decisions at Pearl - 3/9/2010 1:34:34 AM   
Andy Mac

 

Posts: 13621
Joined: 5/12/2004
From: Alexandria, Scotland
Status: offline 		Thats not a script issue so I cannot change it

Scen 6 removes even the chance of it happening

(in reply to Misconduct)
Post #: 3
RE: AI decisions at Pearl - 3/11/2010 4:31:53 PM   
janh

 

Posts: 1140
Joined: 6/12/2007
Status: offline 		Andy, could you describe the game mechanics behind AI decisions a bit more? How does AI determine where to send ships? What determins whether they linger at Pearl, and would say an emergency situation or a great opportunity to inflict heavy damage cause AI to move the KB elsewhere? Or does such a reaction have to be scripted?

I recall you guys did implement a new AI in AE that does both forward and reverse propagation to find the best course. So how does that translate into the AI mechanics now?

(in reply to Andy Mac)
Post #: 4
RE: AI decisions at Pearl - 3/11/2010 4:55:37 PM   
Sardaukar


Posts: 5493
Joined: 11/28/2001
From: Finland/now in Malta
Status: offline
quote:

ORIGINAL: Jim D Burns

With the new patch, I just restarted a grand campaign playing as allies vs. the AI, and I find myself frustrated over the AI decision to linger near Pearl. I tried not to take advantage of the AI, so I left all my bombers stood down and only sent a few subs out to try for a lucky shot.



Are you playing Hard or Very Hard difficulty level? On those levels, AI TFs are not affected by fuel & ammo concerns etc. (IIRC) and that's why they can linger longer. On historical difficulty they usually "return to base" quite quickly, because then AI is aware of lack of fuel/ammunition. They should RTB on any level when lacking planes, though..but it can be that they get this for free on higher difficulty too.


_____________________________

"To meaningless French Idealism, Liberty, Fraternity and Equality...we answer with German Realism, Infantry, Cavalry and Artillery" -Prince von Bülov, 1870-


(in reply to Jim D Burns)
Post #: 5
RE: AI decisions at Pearl - 3/11/2010 8:22:01 PM   
Jim D Burns


Posts: 2730
Joined: 2/25/2002
From: Salida, CA.
Status: offline
quote:

ORIGINAL: Sardaukar
Are you playing Hard or Very Hard difficulty level? On those levels, AI TFs are not affected by fuel & ammo concerns etc. (IIRC) and that's why they can linger longer. On historical difficulty they usually "return to base" quite quickly, because then AI is aware of lack of fuel/ammunition. They should RTB on any level when lacking planes, though..but it can be that they get this for free on higher difficulty too.


Nope historical difficulty. I had read about some of the crazy AI abilities for the higher difficulties, that's why I went with historic.

Jim

_____________________________



(in reply to Sardaukar)
Post #: 6
RE: AI decisions at Pearl - 3/11/2010 10:27:08 PM   
Andy Mac

 

Posts: 13621
Joined: 5/12/2004
From: Alexandria, Scotland
Status: offline 		Understand Jim but IMO taking out some of those advantages on historic because of the whining has neutered histroic difficulty would never even consider playing on that level persoanlly now.

The AI needs a certain level of cheating to be trully effective (all AI's do)

Janh

AI at Stratetic level forms TF's in accordance to pre determined scripts as dictated by the AI script designer
TF's are also formed in response to needs and to move resources/fuel and units etc
All reactions are scripted and are determined by trigger events - I have about 200 trigger scripts int eh current AI scripts where the AI will respond to specirfic actions - but if something happens that I aint thought of the AI will react in weird and wonderfull ways or not at all.

"I recall you guys did implement a new AI in AE that does both forward and reverse propagation to find the best course. So how does that translate into the AI mechanics now?"

Did we dont understand that one I wrote a series of AI scripts that the coders used as the basis for the AI but it doesnt 'think' 
"

(in reply to Jim D Burns)
Post #: 7
RE: AI decisions at Pearl - 3/12/2010 1:02:41 AM   
Jim D Burns


Posts: 2730
Joined: 2/25/2002
From: Salida, CA.
Status: offline
quote:

ORIGINAL: Andy Mac
The AI needs a certain level of cheating to be trully effective (all AI's do)


No problem Andy, this game is really just a warm up for PBEM play anyway. I doubt I’ll get very far into it, I just need to brush up my rusty game skills a bit before I delve into the realm of true competition again.

Jim


_____________________________



(in reply to Andy Mac)
Post #: 8
RE: AI decisions at Pearl - 3/12/2010 1:59:05 AM   
ADB123

 

Posts: 1557
Joined: 8/18/2009
Status: offline 		I've done Historic campaign starts against the AI in all versions of AE, including Patch 3, and I've never seen it hang around Pearl (Probably 8 starts or more). Most of the time it goes West, once in a while it goes north.

(in reply to Jim D Burns)
Post #: 9
RE: AI decisions at Pearl - 3/12/2010 6:10:18 PM   
marc420

 

Posts: 217
Joined: 9/23/2002
From: Terrapin Station
Status: offline 		All AI's don't have problems. Its just a question of how much effort gets put into them. Remember, AIs have been written that beat grand masters at chess.

This is the sort of stuff that sends me away from buying a game.  I don't play PBEM.  My life is busy, so I often can't play computer games for a weeks at a time.  Then, when I get a chance to play, I want to crank through a bunch of turns. In my life, that's usually between about 10pm and 4am.  If there's a PBEM opponent out there who can deal with not playing a turn for three weeks, and who can then send me back my turns on a 15 - 20 min turnaround at 3am when suddenly I've got time to crank out turns, then maybe I'll play PBEM.  Until then, I judge games on their ability to play against the AI because that's the only way these games fit into my life.

And frankly, playing a game one turn a day is just slow torture.  And I don't like the feeling of obligation of having to do a turn every night for the next three months when I do start a game.  I've tried PBEM, and I simply don't like it.

When playing against an AI, what I want is a smart AI that doesn't cheat.  I'm playing the game because I'm interested in the historical situation.  So, an AI that's competitive through 'cheats' like unlimited ammo and very non-historical production and by search routines that nerf the whole concept of spotting and naval search isn't very interesting to me.  By doing that, the whole historical situation that I wanted to immerse myself in is destroyed.  When I set an AI setting to "Very Hard", what I want is a smarter AI, not a cheating AI. 

I've had chess programs where making the AI 'harder' simply gave it longer to think, let it try more options, find the best course etc.  I suppose no chess player would put up with a computer game where the opponent suddenly got a new rook every turn to 'balance' the AI.  I wouldn't mind at all an AI 'very hard' setting that cranked up a 'AI thinking' time between turns to 10 to 15 minutes  on an AI game if it resulted in an AI that didn't destroy its main fleet by keeping it parked off of Pearl as described in this thread.  When I have to start making bad decisions on my own in order to try to nurse an AI through a game, then my interest level in that game goes to zero very quickly.

Good AIs can be written. Its just a question of how much effort gets put into it.  When I'm reading about AE and all the new changes, what you see conspicuously missing is an "AI team" that put as much effort into improving the AI as was put into a "Map team" improving the map.  The results appear to show in the game.  The screen shots show a very pretty map.  Game companies seem to want to spread this myth that good AIs are impossible.  The problem is just simply that they don't put nearly enough effort into it.  Good AIs do take a lot of effort, and many game companies just don't put that effort into it.  A good AI would get a LOT more attention than the map.

Every thread in this forum that mentions the AI just screams at me not to buy this game.  I know I'm not everyone, but to me this results in a $90 game that is simply unplayable for me with the way my life works.  Too bad since this is the sort of heavily detailed game I'd like to play. 


_____________________________

Guard against the impostures of pretended patriotism. ~George Washington

(in reply to ADB123)
Post #: 10
RE: AI decisions at Pearl - 3/12/2010 7:17:55 PM   
John Lansford

 

Posts: 2638
Joined: 4/29/2002
Status: offline 		If you can't play a computer game for weeks at a time, then AE isn't for you in the first place.

Andy,

Is there any way to get the AI to realize that forming CV TF's with carriers that don't have any planes on board is a bad idea?  I had that happen in mid-42 around Guadalcanal; I landed a division there, took the base, then spotted a 2 CV TF headed south towards me.  The airstrikes were very weak, though, only a dozen planes or so although with a sizable escort.  Even their strikes against my carriers (presumed to be a max-attack) was less than 20 attack planes, from carriers ID'ed as Soryu and Akagi.

Basically, it appears that the AI is wearing out its carrier squadrons and not rebuilding them before sending the ships back out again.  Is there any way to put in a test to keep that from happening?  I've sunk at least 2 CV's that had little to no offensive ability for that reason.

(in reply to marc420)
Post #: 11
RE: AI decisions at Pearl - 3/12/2010 8:01:01 PM   
jb123


Posts: 270
Joined: 8/6/2009
Status: offline 		I'm having a similar problem with the AI CVs in an Ironman game set to hard. I'm going through a very stunted, strange version of the battle of midway right now.

AI sent a 2CV and CVL TF to escort a SCTF and amphib right up to Pearl Harbor.

They "raided" PH where I had about two hundred fighters up and a good thousand ack ack guns. Since the game began I've been cheating for the AI. I've restarted a few turns after I accidentally sunk a CV that was doing something really stupid.

At PH I had no choice, I had to protect a bunch of tankers and other ships at port.

Surprisingly the AI zeroes did amazingly well. They shot down tons of p-40b/es and buffaloes. My fighters were all experienced. But I shredded his bombers.

Instead of pulling out the cvs to replace the bombers, he sent them two weeks later to support the Midway invasion.

He still had no bombers, and about half strength fighters. I destroyed him with LBA and one CV flying f4-3s, devastators and sbd-2s.

Ahh well, I still enjoy the game, I just cheat for the AI a lot by keeping fighters and divisions to the rear when I would deploy them forward against a human. It usually works, until the AI decides to raid PH in mid june 42.

(in reply to John Lansford)
Post #: 12
RE: AI decisions at Pearl - 3/12/2010 8:41:37 PM   
Andy Mac

 

Posts: 13621
Joined: 5/12/2004
From: Alexandria, Scotland
Status: offline 		jb123 can I check which version that Ironman game was started under - I thought we had caught all the AI sending TF's to pearl stuff in the patch 2 AI script update ??

Re Carrier fighters and Bombers we have done everything we can in every test we run the AI breaks off to replenish its aircraft when they run out so any saves of instances where the AI isnot breaking off after getting its air groups shredded or is operating with no aircraft in TF's would be good more saves of this we can see the better as we thought we had it nailed

(in reply to jb123)
Post #: 13
RE: AI decisions at Pearl - 3/12/2010 8:42:59 PM   
Andy Mac

 

Posts: 13621
Joined: 5/12/2004
From: Alexandria, Scotland
Status: offline 		jb if you are peeking anywat the AI names TF's by the script they are operating under so for the PH attack you should be able to see it in the named TF which script is driving it there

(in reply to Andy Mac)
Post #: 14
RE: AI decisions at Pearl - 3/12/2010 9:51:21 PM   
wpurdom

 

Posts: 410
Joined: 10/27/2000
From: Decatur, GA, USA
Status: offline 		Chess is not representative of AIs.

It's a well-understood game with limited options into which enormous efforts have been put. There are few enough options in chess as to the number of moves, that as computers have raised their capacities, a certain amount of brute force apporach has been possible - i.e. try all the alternatives for many turns. That's not possible in a game like AE.

Even Go has been much more difficult to program than chess.

Have you read any of Dr. Herwin's posts - most AI's "have the intelligence of a colony of ants."

(in reply to Andy Mac)
Post #: 15
RE: AI decisions at Pearl - 3/12/2010 10:59:11 PM   
janh

 

Posts: 1140
Joined: 6/12/2007
Status: offline
quote:

ORIGINAL: wpurdom

Chess is not representative of AIs.

It's a well-understood game with limited options into which enormous efforts have been put. There are few enough options in chess as to the number of moves, that as computers have raised their capacities, a certain amount of brute force apporach has been possible - i.e. try all the alternatives for many turns. That's not possible in a game like AE.


Unfortunately this totally correct. Chess is a "multilinear problem" with limited space of options. That's why you can have a "blue gene" think it through a little longer, and winning. Of course there are some very smart algorithms still that qualify good and bad moves and have it think through the better ones, but still that is a finite number. With something of the scope of WITP or AE this comes simply down to too many options.

I do, however, agree with you that any improvement in core-AI would be awesome. I have been following this off and on for a while and still wait for a reason to buy it, but I can't invest enough time or reliability to ask someone for PBEM. I have been following Andys explanations and tutorials on scripting, which seems promising. But from the incomplete picture I have it seems that the available scripting framework may be insufficient from programming a really dynamic response. ARMA 2 is an example for a very powerful scripting framework (http://community.bistudio.com/wiki/Category:Scripting_Commands_ArmA2), but it allows enormous possibilities (of course it is not turn-based strategy, so here it could be much simpler).

What I would guess would be abstracted functions (not specifying at which base, sea area, which specific ships etc), say for example one mimicking the AI dynamically reacting to a human feinting with a CV-TF.

Bluuurrp (just for the idea, forget the syntax otherwise):
if ( {"CV" || "CVL" || "CVE"} count _playerTF>2 && {"BB" || "BC" || "CA"} count _playerTF>0 && AI knowsabout _playerTF>0 && fighterready(_playerTF)>20 && TBFready(_playerTF)>20 && _playerTF distanceToNextBase(position_playerTF, radius=20, basesize>=3)<10 && availableShips(position_playerTF, radius=20, "CV")>=3) then
{pause other scripts using these units; form TF if these and these support ships ready; assign free commander; send out on some pattern; bring them back and continue previous scripts if ships return intact};

Andy, can you do something dynamic like this with the scripting language?


(in reply to wpurdom)
Post #: 16
RE: AI decisions at Pearl - 3/13/2010 3:12:45 AM   
bklooste

 

Posts: 1103
Joined: 4/10/2006
Status: offline 		As Mr Burns stated the issue isnt lingering at Pearl but the AI attacking when pilots are too fatigued.

_____________________________

Underdog Fanboy

(in reply to janh)
Post #: 17
RE: AI decisions at Pearl - 3/13/2010 4:48:49 AM   
Rainer

 

Posts: 1161
Joined: 11/21/2000
From: Neuching, Bavaria, Germany
Status: offline 		If you have your mind set why bother to annoy others? Ego problems?

_____________________________

WitP/AE Devs Fan Boy
1.7.11.23k beta
Data base changes by Andy Mac October 16, 2012
Scen #1 Allied vs AI Level Hard Daily Turns
Art Mods by TomLabel and Reg

(in reply to marc420)
Post #: 18
RE: AI decisions at Pearl - 3/13/2010 10:33:22 AM   
bklooste

 

Posts: 1103
Joined: 4/10/2006
Status: offline
quote:

ORIGINAL: marc420
Good AIs can be written. Its just a question of how much effort gets put into it. 


Sorry i dont think so....teh effort is too much for even minimal effects , unless users are willing to pay 1M per copy , the returns of scale of strategy games is too low these days with everyone playing WOW and Xbox .. Chess is not really AI , I have written it and its brute force , these days since there are only 64 squares you can fit every square in a single 64 bit quad word and these masks contain every possible move for a piece so they can be evaluated incredibly quickly. Also though they can see 8-10 moves in future the strategy is awfull so they still relly on opening book DBs which are prepared by human GMs for the first 20 years , they also use end game tables ( So yes chess computers cheat !) . Chess games also have the advantage of a huge established base consisting of thousands of man years of ideas and trials , eg early strategy versions of chess were useless and were abandoned , deep blue cost 10s of millions and thats for a problem space of 6 different types of pieces with simple interaction over 64 possible location .

Free move games , or games with lots of hexes that require strategy are a completely different and much harder .

Lets do an excercise say AE sells 10,000 copies ( its probably more like 2K) and you put $10 of the budget on AI thats 100K which buys you about 1 man year which is penuts for an IT project .


_____________________________

Underdog Fanboy

(in reply to marc420)
Post #: 19
RE: AI decisions at Pearl - 3/13/2010 10:39:18 PM   
jb123


Posts: 270
Joined: 8/6/2009
Status: offline
quote:

ORIGINAL: Andy Mac

jb123 can I check which version that Ironman game was started under - I thought we had caught all the AI sending TF's to pearl stuff in the patch 2 AI script update ??

Re Carrier fighters and Bombers we have done everything we can in every test we run the AI breaks off to replenish its aircraft when they run out so any saves of instances where the AI isnot breaking off after getting its air groups shredded or is operating with no aircraft in TF's would be good more saves of this we can see the better as we thought we had it nailed


i started under one of the hotfixes for patch two in December. I thought that Ironman version wasn't fixed. I guess I was wrong.

I usually turn the other way when they do stuff like that. I may post some saves in tech next week when I have access to the game files.

Thanks for the post, I'm not complaining, just observing.

(in reply to Andy Mac)
Post #: 20
Page:   [1]
All Forums >> [New Releases from Matrix Games] >> War in the Pacific: Admiral's Edition >> AI decisions at Pearl Page: [1]
Jump to:





New Messages No New Messages
Hot Topic w/ New Messages Hot Topic w/o New Messages
Locked w/ New Messages Locked w/o New Messages
 Post New Thread
 Reply to Message
 Post New Poll
 Submit Vote
 Delete My Own Post
 Delete My Own Thread
 Rate Posts


Forum Software © ASPPlayground.NET Advanced Edition 2.4.5 ANSI

0.172