RE: AI craziness

Matrix Games Forums

Forums Register Login Photo Gallery Member List Search Calendars FAQ

My Profile Inbox Address Book My Subscription My Forums Log Out

RE: AI craziness

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

Logged in as: Guest

Users viewing this topic: none

Printable Version

All Forums >> [Current Games From Matrix.] >> [World War II] >> War In The Pacific - Struggle Against Japan 1941 - 1945 >> RE: AI craziness

Page: << < prev 1 2 3 [4]

Message

<< Older Topic Newer Topic >>

RE: AI craziness - 7/14/2004 8:47:30 AM

JohnK

Posts: 285
Joined: 2/8/2001
Status: offline

quote:

ORIGINAL: ZOOMIE1980

Yes, it is giving me one hell of a tough fight. But it is very "linear", show virtually NO inclination to adapt to my situation. I've played the first three-four weeks six times now, with various levels of reenforcement, but it basically does almost the exact same thing each time, regardless of the force it encounters. The variablity is in the natural randomness one gets for random number generation and the different precise timeline the AI gets put on based on the human moves. I have yet to see it attempt a single flanking manuever around a powerful defensive position, for instance. It just bludgeons the position until it beats you down.....which it obviously knows it can at this stage of the game.

BTW, playing the "hard" setting right above "historical".

I'm actually fighting for Rabaul, sent Sparrow Force there and the Australian/NZ cruisers...Dec. 20th or so..(playing on "Hard".)

My Cruisers are sitting there in a TF, the Bettys at Truk simply keep bombing the base to little effect, never have tried to attack the cruisers....assorted Japanese TFs of APs, AKs, and Cruisers constantly pass nearby, but none ever try to land or enter the hex.

Other than the plodding Singapore and Phillipine offensives, the computer has landed forces at Menando and one of the north Borneo ports that are too small to take either.

Not real impressed with the AI.

< Message edited by JohnK -- 7/14/2004 6:48:11 AM >

(in reply to ZOOMIE1980)

Post #: 91

RE: AI craziness - 7/14/2004 9:13:42 AM

ZOOMIE1980

Posts: 1284
Joined: 4/9/2004
Status: offline

quote:

ORIGINAL: JohnK

quote:

ORIGINAL: ZOOMIE1980

Yes, it is giving me one hell of a tough fight. But it is very "linear", show virtually NO inclination to adapt to my situation. I've played the first three-four weeks six times now, with various levels of reenforcement, but it basically does almost the exact same thing each time, regardless of the force it encounters. The variablity is in the natural randomness one gets for random number generation and the different precise timeline the AI gets put on based on the human moves. I have yet to see it attempt a single flanking manuever around a powerful defensive position, for instance. It just bludgeons the position until it beats you down.....which it obviously knows it can at this stage of the game.

BTW, playing the "hard" setting right above "historical".

I'm actually fighting for Rabaul, sent Sparrow Force there and the Australian/NZ cruisers...Dec. 20th or so..(playing on "Hard".)

My Cruisers are sitting there in a TF, the Bettys at Truk simply keep bombing the base to little effect, never have tried to attack the cruisers....assorted Japanese TFs of APs, AKs, and Cruisers constantly pass nearby, but none ever try to land or enter the hex.

Other than the plodding Singapore and Phillipine offensives, the computer has landed forces at Menando and one of the north Borneo ports that are too small to take either.

Not real impressed with the AI.

Mogami and some others have warned us that if you get much at all outside historical norms, at least in the initial offensive stage, the AI will probably "break". This may be one of those.

When I was working on AI's for the Pentagon's simulations, we used a rather simple "brute force" intelligence technique. Had a lot of the success sanity check points I've alluded to that would branch to "Plan B" and "Plan C" sub routines. MASSIVE amounts of code and even more MASSIVE amount of data storage, as it used disk based database servers to service that design. Obviously something well beyond what current game programmers do.

The thing is, though, the time has arrived for these types of solutions to make it onto home computers. There is nothing "cosmic" about these AI techniques. They are essentially massive brute force approaches...nothing like some the research you may have read about in these areas. But through well designed toolkits these types of efforts should be something that is managable with current technology.

No help for WitP. But hope for the future. As for WitP, this AI is still pretty good, so long as you don't do too much outside the box. Your's may be outside the performance envelope, though. Give it some time, and see if it eventually reacts in a meaningful way. I've discovered that sometimes it takes the AI a while to finally deal with your moves.

(in reply to JohnK)

Post #: 92

RE: AI craziness - 7/14/2004 6:11:21 PM

JohnK

Posts: 285
Joined: 2/8/2001
Status: offline

Probably should have clarified, this is AFTER the Japanese made their initial landing on Rabaul; so with Sparrow force added I've now matched them in troops there.....

I've got a squadron of Wirraways and 4 Hudsons there; I've been spotting a Japanese CA force near Rabaul for about a week; it never actually goes there...and they never reinforce their troops there.

Will be interesting to see what happens (no sign of the KB in the game since PH, either.)

(in reply to ZOOMIE1980)

Post #: 93

RE: AI craziness - 7/14/2004 6:38:47 PM

Grotius

Posts: 5798
Joined: 10/18/2002
From: The Imperial Palace.
Status: offline

Yes, I've seen the AI land with too few troops too, but both sides made that mistake repeatedly during the war. That's happened to me in my current game at Menado, just like with you. That doesn't bother me, especially since the AI invariably reinforces its initial invasion force, though this can take time. Case in point: the Japanese finally reinforced its force at Menado, and it just fell.

In my game, on "historical," I wouldn't call the AI offensive in the Phillipines, DEI, Rabaul and Malaya "plodding." If anything, it's quicker than historical. But I guess I do need to play on the harder difficulty levels. I plan a game as the Japanese on a harder setting, after this first learning game.

I didn't try reinforcing Rabaul because I found the Japanese LBA quite willing to attack my surface ships around Singapore and the Phillipines. I assumed it would do the same out of Truk, but that is a pretty long-range attack. (I don't remember every trying to attack anything out of Truk in UV.) Sounds like it's working for you. Actually, I did think about moving some Aussies in, but I was stymied by the no-load rule. I didn't think of using the Sparrow force (where are they, Timor?). I guess I also kinda figured I'd play this first game quasi-historically. Though that doesn't mean I send Prince of Wales and Repulse north to test LBA. :)

Incidentally, KB has repeatedly appeared in my game, first to attack ships in port around the DEI, then around the Phillipines, most recently to cover IJA shipping rounding Singapore. And CVL Ryujo has hung around the DEI for the whole game, it seems, launching occasional attacks on ships in port and sinking a couple transports here and there.

Anyway, I think the AI is pretty good, as PC wargame AI's go. If the devs can fix the silliest AI behavior, I'll be happy with it, since I already feel like it's giving me a good game.

(in reply to JohnK)

Post #: 94

RE: AI craziness - 7/14/2004 6:41:39 PM

The Gnome

Posts: 1233
Joined: 5/17/2002
From: Philadelphia, PA
Status: offline

quote:

Incidentally, KB has repeatedly appeared in my game, first to attack ships in port around the DEI, then around the Phillipines, most recently to cover IJA shipping rounding Singapore. And CVL Ryujo has hung around the DEI for the whole game, it seems, launching occasional attacks on ships in port and sinking a couple transports here and there.

Ryujo was a thorn in my side around PI throughout December until the Skipjack put two torpedoes in her side. ;) Haven't seen her in a while but I doubt that's bad AI.

I'm getting to the point in the game where I'm expecting KB to show up in DEI, but even if they're a no show for another month I don't think that's completely unreasonable.

(in reply to Grotius)

Post #: 95

RE: AI craziness - 7/14/2004 6:56:22 PM

JohnK

Posts: 285
Joined: 2/8/2001
Status: offline

Oh, I sent Repulse and POW north to Khota Baru......

The BBs had left, Japanese had 4 cruisers guarding the transports.

Not a single salvo from the Repulse and POW hit ANYTHING. I got one Japanese DD, but my CLs and DDs were badly shot up. Repulse and POW came out without a scratch. I didn't see or hit any transports.

Went back to Singapore, they've been there in a TF ever since. Other than one fragmented Betty Attack (about 12 planes) they've never been attacked from the air.

Sparrow force begins in Darwin.

I don't see sending them to Rabaul after the Japanese invade Rabaul as horrendously ahistorical; it SHOULD not work from the Allied point of view, even against the AI.

(in reply to Grotius)

Post #: 96

RE: AI craziness - 7/14/2004 7:03:06 PM

ZOOMIE1980

Posts: 1284
Joined: 4/9/2004
Status: offline

KB has finally showed itself near Sumatra for me on 1/24/42. I did spot it well west of Bataan about a week earlier when it launched a massive airfield attack on Juno of all places? Then, a week later, it showed up. Not sure if it going to stay there or head for the Malacca Strait. It has yet to attack anything.

Also no sign of IJA moving forces north to Burma from Malaysia yet. Singapore fell five days ago. Bombardement of Rangoon is now in its 13 straight day with no change at all in force structure or results (a piddling 10-15 casualties per day).

AI took Clark two days ago but has shown no sign of moving any forces to Bataan or Manila. The Jap force at Manila has been bombarding now for 18 straight days with no change in force size and no change in results (about 40-50 casualties per turn). Wonder how much longer that will go on before they move more troops in?

AI is very plodding, systematic and relentess so far. It always seems to land too few troops on contested landings and about 5-7 days later reinforce, slowly. If that's not enough, they reinforce again about 5-7 days after that.

(in reply to The Gnome)

Post #: 97

RE: AI craziness - 7/14/2004 8:26:47 PM

The Gnome

Posts: 1233
Joined: 5/17/2002
From: Philadelphia, PA
Status: offline

quote:

ORIGINAL: JohnK

Oh, I sent Repulse and POW north to Khota Baru......

The BBs had left, Japanese had 4 cruisers guarding the transports.

Not a single salvo from the Repulse and POW hit ANYTHING. I got one Japanese DD, but my CLs and DDs were badly shot up. Repulse and POW came out without a scratch. I didn't see or hit any transports.

I did the same thing, ended up getting the CA to retreat after seriously damaging Mikuma. Went back the next turn to hit the transports and found 15 sitting there unescorted, the result? 30 shells into 1 small AK and 1-2 shells into two others... but I already have thread for that in support. ;)

< Message edited by The Gnome -- 7/14/2004 1:27:46 PM >

(in reply to JohnK)

Post #: 98

RE: AI craziness - 7/14/2004 8:50:41 PM

ZOOMIE1980

Posts: 1284
Joined: 4/9/2004
Status: offline

quote:

ORIGINAL: The Gnome

quote:

ORIGINAL: JohnK

Oh, I sent Repulse and POW north to Khota Baru......

The BBs had left, Japanese had 4 cruisers guarding the transports.

Not a single salvo from the Repulse and POW hit ANYTHING. I got one Japanese DD, but my CLs and DDs were badly shot up. Repulse and POW came out without a scratch. I didn't see or hit any transports.

I did the same thing, ended up getting the CA to retreat after seriously damaging Mikuma. Went back the next turn to hit the transports and found 15 sitting there unescorted, the result? 30 shells into 1 small AK and 1-2 shells into two others... but I already have thread for that in support. ;)

Hmm, wonder if their target selection algorithm is any different that ones in Grigsby's early games like GuadalCanal and North Atlantic '86? I know in those games all the capital ships were at the top of list followed by cruisers, destroyers and then the frigates. The target selection code started at the bottom of the list and worked up, testing each ship for selection. It would repeat the list walk until a ship got selected. While the frigates had very low selection values, there were a lot of them so they got picked the most often as the odds were, even if low, individually, out of 20 or so of them one still got picked before it got to the DDG, CA, CG, CGN, and CV's.... this abstracted the notion that the CV's were always at the center of the TF, immediately surrounded by the Aegis cruisers who in turn were surrounded by the DDGs and ASW DD's with the FFG's forming the outer ring of picket ships.

Problem was, with that simplistic algorithm, if all the ships had roughly equal selection values the first ship always got hammered and the others only got scratched. We have the same problem here in our call center management telephony applications. How to keep the first agent on the list from getting all the calls while the agents down the list get to read a book all day out of boredom. As a result we have developed a lot of work-load distribution algorithms for this kind of problem. Seems to be the same sort of problem here.

(in reply to The Gnome)

Post #: 99

RE: AI craziness - 7/14/2004 9:23:35 PM

The Gnome

Posts: 1233
Joined: 5/17/2002
From: Philadelphia, PA
Status: offline

quote:

ORIGINAL: ZOOMIE1980

quote:

ORIGINAL: The Gnome

quote:

ORIGINAL: JohnK

Oh, I sent Repulse and POW north to Khota Baru......

The BBs had left, Japanese had 4 cruisers guarding the transports.

Not a single salvo from the Repulse and POW hit ANYTHING. I got one Japanese DD, but my CLs and DDs were badly shot up. Repulse and POW came out without a scratch. I didn't see or hit any transports.

I did the same thing, ended up getting the CA to retreat after seriously damaging Mikuma. Went back the next turn to hit the transports and found 15 sitting there unescorted, the result? 30 shells into 1 small AK and 1-2 shells into two others... but I already have thread for that in support. ;)

Hmm, wonder if their target selection algorithm is any different that ones in Grigsby's early games like GuadalCanal and North Atlantic '86? I know in those games all the capital ships were at the top of list followed by cruisers, destroyers and then the frigates. The target selection code started at the bottom of the list and worked up, testing each ship for selection. It would repeat the list walk until a ship got selected. While the frigates had very low selection values, there were a lot of them so they got picked the most often as the odds were, even if low, individually, out of 20 or so of them one still got picked before it got to the DDG, CA, CG, CGN, and CV's.... this abstracted the notion that the CV's were always at the center of the TF, immediately surrounded by the Aegis cruisers who in turn were surrounded by the DDGs and ASW DD's with the FFG's forming the outer ring of picket ships.

Problem was, with that simplistic algorithm, if all the ships had roughly equal selection values the first ship always got hammered and the others only got scratched. We have the same problem here in our call center management telephony applications. How to keep the first agent on the list from getting all the calls while the agents down the list get to read a book all day out of boredom. As a result we have developed a lot of work-load distribution algorithms for this kind of problem. Seems to be the same sort of problem here.

It might be, I don't know. I know people were saying it's due to a convoy scattering rule. Of course then the three ships I fired on should have roughly equal hits assuming the others got away. Plus when is enough, enough? I think that first ship is going to go down... no nee to keep firing until the vertical stern goes completely under.

< Message edited by The Gnome -- 7/14/2004 2:24:06 PM >

(in reply to ZOOMIE1980)

Post #: 100

RE: AI craziness - 7/15/2004 2:18:03 AM

JohnK

Posts: 285
Joined: 2/8/2001
Status: offline

quote:

ORIGINAL: ZOOMIE1980

AI is very plodding, systematic and relentess so far. It always seems to land too few troops on contested landings and about 5-7 days later reinforce, slowly. If that's not enough, they reinforce again about 5-7 days after that.

Yeah, this makes me think there isn't scripting, but actually the AI might be better if there WAS early Japanese scripting.

They're landing lone SNLFs at defended bases and not reinforcing them at all in my game.....though based on what you've said they may reinforce them in a day or two.....

(in reply to ZOOMIE1980)

Post #: 101

RE: AI craziness - 7/15/2004 4:37:12 PM

ZOOMIE1980

Posts: 1284
Joined: 4/9/2004
Status: offline

More than a day or two. I've seen them wait up to two weeks to finally land additional forces. Kendari was one example in my game. They landed, did bombardment attacks for about 4 days, a shock attack which was bad for them, then roughly 10 days of bombardment attacks and finally landed more troops and took it.

And I have a historical oddity developing in the PI. It appears they Japanese are going to take Bataan BEFORE they get Manila. They are now deliberately attacking in Bataan but are on their fourth week of bombardment only attacks in Manila and are now have almost ZERO effect, even though Manila is completely out of supplies.

(in reply to JohnK)

Post #: 102

RE: AI craziness - 7/15/2004 8:19:13 PM

Grotius

Posts: 5798
Joined: 10/18/2002
From: The Imperial Palace.
Status: offline

Bombardment attacks increase the disruption and fatigue of the targets, no?

(in reply to ZOOMIE1980)

Post #: 103

RE: AI craziness - 7/15/2004 8:26:10 PM

Mr.Frag

Posts: 13410
Joined: 12/18/2002
From: Purgatory
Status: offline

quote:

Bombardment attacks increase the disruption and fatigue of the targets, no?

Not quite but in a round about way ...

Bombardment is going to hit exposed support types which will spa strength of the unit.

Aircraft on ground attack specifically target the biggest baddest unit in the hex and make it extrememly tired so it is basically useless for the ground combat phase.

(in reply to Grotius)

Post #: 104