|
That's fascinating to me. I've been trying to a bit more study on AI programming and really I dunno why they didn't go with some fairly brute force approach with it. The weakness of things like alphabeta pruning in this kind of game is just how long it would take even if you managed to limit the amount of choices to more broad decisions like "send AttackForceFoo against TargetBar" or more generalist strategy decisions. I've seen a few blurbs on how the Total War series was using monte carlo alphabeta with that kind of approach and they have to really struggle with keeping turn times down. But this game is groggy as poo poo, with player turns taking a considerable amount of time to carry out anyways... I'd figure you could really let loose with some crude-but-powerful approaches that take like, five minutes a turn or some poo poo and eat up half your system memory because the consumer base for this sort of game would more a lot more tolerant if it meant having some nasty AI. That and the AI would be able to handle all the fiddly stupid bullshit effectively, at least, in theory, so it should already have a pretty keen edge on a human. The other fascinating thing is that it has that problem of inability to commit enough towards a goal. Thats a problem in a lot of strategy games, and its really one of the most fatal weaknesses that result in theoretically superior AI forces folding like a leaf against a halfway competent player. At least this AI is persistant, but it really, REALLY should be able to at least go "Well that operation I attempted with X resources met with miserable failure, but I REALLY REALLY want this to succeed, so lets use that as a basis to calibrate what it would take to win, and then send 133% of that just to be safe." (Like I said tho, I'm just studying this, haven't done much application of it, so if anyone who knows more wants to correct / address what I wrote, please go ahead)
|
#
?
Nov 13, 2017 17:14
|
|
|
|
| # ? May 17, 2024 01:53 |
|
|
I wonder what happens if we let two instances of AlphaGo loose on this game...
|
|
#
?
Nov 13, 2017 18:04
|
|
|
   Owch, that was a bad run!   Our Bettys get eyes on target, but take heavy losses to the CAP without getting in a hit.  With fighter support our carrier strike gets in sight, but flak is heavy.  We finally get a hit in.  Their own strike is fended off nicely. Only one bomber even sees a carrier.  That should put paid to her.  The flak is still heavy though.  God dammit! That puts a dampner on a good day! Lets hope damage control are up to their normal high stand......   We took out a carrier, but drat was that a bad hit at the last second.  That's pretty much kill confirmed.  The Kaga's going to be okay.
|
|
#
?
Nov 13, 2017 19:00
|
|
|
What does 23 flood damage mean? Is that a lot? Will it keep on taking water while moving to port? This game makes it really tough to see what any of the numbers actually mean. Also, can anyone recommend a decent browser-based strategy game I can play at work?
|
|
#
?
Nov 13, 2017 19:57
|
|
|
It's not trivial but should be okay with no further damage. Once you get above 50 it starts to get dangerous. However, the ship will need to transit more slowly to keep from making the damage worse, which makes it more vulnerable to subs and aircraft.
|
|
#
?
Nov 13, 2017 20:31
|
|
|
 Bguy32Not confirmed via points yet
|
|
#
?
Nov 13, 2017 20:31
|
|
|
Can it still perform air operations?
|
|
#
?
Nov 13, 2017 20:47
|
|
|
Ron Jeremy posted:Can it still perform air operations? I believe so, as it's squadrons appear to have landed safely on it.
|
|
#
?
Nov 13, 2017 20:51
|
|
|
13 November 1943 A much less interesting day in OTL, as all I've got is another RN escort destroyer (HMS Dulverton) getting hit by a glide bomb in the Dodecanese. This time, the damage is enough to warrant scuttling the ship.
|
|
#
?
Nov 13, 2017 21:40
|
|
|
pthighs posted:It's not trivial but should be okay with no further damage. Once you get above 50 it starts to get dangerous. Still cruising at 11 knots in the pic, and almost no systems damage should reduce flooding further even at sea. FLOODING INCREASES ON CV KAGA DAMAGE CONTROL PARTIES TRAPPED IN FLOODED COMPARTMENT ON CV KAGA DAMAGE CONTROL PARTIES TRAPPED IN FLOODED COMPARTMENT ON CV KAGA Or not.  (Of course this game models random DC mishaps, duh)
|
|
#
?
Nov 13, 2017 22:40
|
|
|
11 knots is stupid slow in warship terms. Dangerously close to sub bait.
|
|
#
?
Nov 13, 2017 22:53
|
|
|
goatface posted:11 knots is stupid slow in warship terms. Dangerously close to sub bait. 11 knot cruise speed, 21 knots full. Kagas undamaged speeds are 16 and 28 respectively. She got off really light.
|
|
#
?
Nov 13, 2017 22:59
|
|
|
Today we celebrate our independence day.
|
|
#
?
Nov 13, 2017 23:44
|
|
|
Grey Hunter posted:
My lucky ship! Careful with that, it was super explodey when damaged...
|
|
#
?
Nov 14, 2017 03:52
|
|
|
goatface posted:11 knots is stupid slow in warship terms. Dangerously close to sub bait. Fully expecting the next day to be all the subs waking up
|
|
#
?
Nov 14, 2017 04:50
|
|
|
Dead CVL and some messy air losses and one damaged CV. Not bad.
|
|
#
?
Nov 14, 2017 07:16
|
|
|
Kodos666 posted:I wonder what happens if we let two instances of AlphaGo loose on this game... quote:. I'd figure you could really let loose with some crude-but-powerful approaches that take like, five minutes a turn or some poo poo and eat up half your system memory because the consumer base for this sort of game would more a lot more tolerant if it meant having some nasty AI. That and the AI would be able to handle all the fiddly stupid bullshit effectively, at least, in theory, so it should already have a pretty keen edge on a human. Unlike GO, here you don't know where the enemy fleet is. You now have a belief state you need to deal with (for each boat, probabilistically what is the probability that is in this area?) which makes brute force even less effective. You'd need to start dealing with hand-rules (kind of what you mentioned of "I tried with X resources, didn't work, let's try with more") or some smart statistical object to narrow the problem down to something manageable. I'd love a game at the intersection between groggy wargame and programming game. But I don't think there is any. I think the best you can mod in WITP is some basic AI parameters.
|
|
#
?
Nov 14, 2017 11:55
|
|
|
Given There's a single Score value for each side in this game, AlphaGo or a similar algorithm might actually be able to have a go at learning to optimise this game. Although, I dispair to think how long it would take for it to figure out where all the menus are and learn how to use them.
|
|
#
?
Nov 14, 2017 12:47
|
|
|
nothing to seehere posted:Given There's a single Score value for each side in this game, AlphaGo or a similar algorithm might actually be able to have a go at learning to optimise this game. Although, I dispair to think how long it would take for it to figure out where all the menus are and learn how to use them. The algorithm bould become conscious of itself and kill all humans, for the crime of exposing it to such a UI.
|
|
#
?
Nov 14, 2017 13:01
|
|
|
As long as we get at least a year or two to play WITP with an awesome UI, i'd say its worth the trivial price of total annihilation.
|
|
#
?
Nov 14, 2017 14:30
|
|
|
nothing to seehere posted:Given There's a single Score value for each side in this game, AlphaGo or a similar algorithm might actually be able to have a go at learning to optimise this game. The problem is not so much about the unknown score function, it's really about the size of the state, action space and rule weirdness that makes this impossible to brute force. In chess maximizing greedily the score is a good strategy, here it might not be the case. For example right now Grey could push more aggressively with the Kido Butai and pulverize waves of patrol boats in the middle of nowhere. He would burn fuel (which I don't think is present in the score) in exchange for a few small points but we can all agree it would be strategically silly.
|
|
#
?
Nov 14, 2017 16:32
|
|
|
It's kinda weird that grog devs aren't more gung ho for AI. In a way you're conceptualizing and creating THE PERFECT GENERAL, you'd think that they'd be all over that.
|
|
#
?
Nov 14, 2017 16:59
|
|
|
CarrKnight posted:The problem is not so much about the unknown score function, it's really about the size of the state, action space and rule weirdness that makes this impossible to brute force. Aren't those the same arguments that were used against AlphaGo getting good enough to beat humans? I'm not an expert, but from what I understood of articles about the algorithm behind AlphaGo those are things it learnt to overcome by playing against it self all the time: the biggest issue here is that rules, objectives, and resources aren't symmetrical, which has been true for any solved game so far I think.
|
|
#
?
Nov 14, 2017 17:16
|
|
|
nothing to seehere posted:Aren't those the same arguments that were used against AlphaGo getting good enough to beat humans? The increase in complexity when we deal with hidden information is huge.
|
|
#
?
Nov 14, 2017 17:31
|
|
|
aphid_licker posted:It's kinda weird that grog devs aren't more gung ho for AI. In a way you're conceptualizing and creating THE PERFECT GENERAL, you'd think that they'd be all over that. AI is a hard CS concept and grog developers are notoriously low-skilled.
|
|
#
?
Nov 14, 2017 18:09
|
|
|
Would a CVL be something akin to a CVE?
|
|
#
?
Nov 15, 2017 05:00
|
|
|
Kibayasu posted:Would a CVL be something akin to a CVE? They're similar in some capacities, such as their plane counts, but serve separate roles. CVLs, or light carriers, still retained relatively high speeds in order to operate alongside fleet carriers, or otherwise be used in combat against enemy ships. CVEs, or escort carriers, on the other hand tended to be quite slow, good for convoy escort or similar duties but incredibly unsuitable for fleet actions. In both cases the reasons for building them was simply because they were a lot cheaper than a true fleet carrier. Neither were incredibly offensive weapons, as both tended to carry a disproportionate number of fighters compared to the rest of their air wing, but escort carriers tended to carry almost entirely fighters, with maybe a small number of bombers to prosecute submarines, while light carriers tended to carry a squadron or so of torpedo bombers. Lord Koth fucked around with this message at 05:25 on Nov 15, 2017 |
|
#
?
Nov 15, 2017 05:20
|
|
|
Kibayasu posted:Would a CVL be something akin to a CVE? CVL is a light carrier, goes the speed of a fleet carrier, carries about 1/3 the number of planes a fleet carrier does. CVE is an escort carrier, goes the speed of a cargo ship, carries about 1/4 the number of planes as a fleet carrier.
|
|
#
?
Nov 15, 2017 05:21
|
|
|
Isn't it common practice for the AI in strategy games to cheat in some manner to make up for difficulty of simulating everything that a human player has to do? I know I've seen RTS games where the computer player gets a huge resource bonus and greatly decreased build times to make up for the fact that its tactics are basically "build x units and send them to the enemy base" and would be laughably easy to beat if it had to play by the same rules as a human player.
|
|
#
?
Nov 15, 2017 05:23
|
|
|
Kibayasu posted:Would a CVL be something akin to a CVE? Kind of? CVL sorta covers a broad category of ships, from early fleet carriers that would be considered CVLs due to their limited capabilities, to cruiser-conversions like the Independance here, to custom built mini-fleet carriers like the Majestic and Colossus, with aircraft compliments ranging from 24-40+. CVEs generally where very cheap and vulnerable, almost disposable, and mainly used to ferry aircraft or protect convoys from lone raiders and hunt subs. IIRC CVEs where often merchant-conversions, whereas CVLs are purpose built for combat. They where something only the UK and US really invested in, and a bunch of them ended up sold off to middle power countries (Canada, Australia, Spain, Brazil, etc) as flagships that would in some cases operate all the way through the 20th century and beyond. In contemporary naval power, every carrier currently in operation that isn't an American CVN could broadly be considered a CVL, including Marine Assault Ships, Russia's Admiral Kuznetsov, the Chinese Liaoning and yet-unnamed Type 001A, France's Charles de Gaulle, etc... That said it's entirely possible that the Aircraft carrier is now in the position that Battleships occupied in the 1940s, rendered an expensive convenient target for submarines and ballistic missiles.
|
|
#
?
Nov 15, 2017 05:55
|
|
|
Flavius Belisarius posted:What does 23 flood damage mean? Is that a lot? Will it keep on taking water while moving to port? This game makes it really tough to see what any of the numbers actually mean. As mentioned above, the flooding and fire damage values are generally manageable/survivable below 50. Engine damage impacts on speed, obviously. The one to watch out for is systems damage, which represents how well the ship can continue to carry out its duties. This can mean air ops but also includes damage control teams, fire fighting systems etc. In this case, the Kaga is functioning quite well, and so will get the flooding under control and will even reduce the damage a little over the next few days. When you see heavy fires/flooding AND high systems damage, you're hosed.
|
|
#
?
Nov 15, 2017 06:01
|
|
|
Deep Blue gets turned into a WITP AI (shut up I know this isn't possible lalalala) It calculates every possible move of every possible ship and plane and man on the map and prints the following result after years of processing. "The Tokyo Bay Fortress is in Tokyo Bay."
|
|
#
?
Nov 15, 2017 06:34
|
|
|
ponzicar posted:Isn't it common practice for the AI in strategy games to cheat in some manner to make up for difficulty of simulating everything that a human player has to do? I know I've seen RTS games where the computer player gets a huge resource bonus and greatly decreased build times to make up for the fact that its tactics are basically "build x units and send them to the enemy base" and would be laughably easy to beat if it had to play by the same rules as a human player. I'm pretty sure I read on the matrix forums that the WitP-AI gets to cheat in that it can teleport ships to some degree. When a script wants a certain TF composition departing a certain base it can teleport idle ships there from other bases to get the TF going. I guess modelling all the transits and allocating escorts and so on would be too much complexity. I'm not sure, but I guess there's a range restriction though. I don't think the AI gets to teleport a fleet carrier from San Fransisco to Sydney.
|
|
#
?
Nov 15, 2017 11:20
|
|
|
algebra testes posted:Deep Blue gets turned into a WITP AI (shut up I know this isn't possible lalalala)
|
|
#
?
Nov 15, 2017 15:38
|
|
|
Caconym posted:I'm pretty sure I read on the matrix forums that the WitP-AI gets to cheat in that it can teleport ships to some degree. When a script wants a certain TF composition departing a certain base it can teleport idle ships there from other bases to get the TF going. In the WitP timeline, the Philadelphia experiment is real and successful.
|
|
#
?
Nov 15, 2017 16:59
|
|
|
   Another day, another pointless invasion of Luganville. They re-embark in the next phase.  This is more worrying, this may tip them over the edge into winning territory!  I have literally no idea what this task force is doing here. Not a clue! The more worrying thing is that this means there is a carrier supporting their invasion of Jaluit!   We drive off a sub.  The Independence is still alive!  It's just as well I don't go around trusting kill reports....   We cross a river and the Chinese troops flee before us.  I'm a little annoyed there were no afternoon attacks on the Independence, but I need to sweep the carriers up the Jaluit now to deal with whatever is there.  This must have been an supply run or something.
|
|
#
?
Nov 15, 2017 18:49
|
|
|
  The Guns of Juilit do their work again.  They take offence to the existence of the Custer.  Goodbye mate.  Normally I'd hunt you down. But not today.  Another surface raid brings home the bacon.  The Pompon gets past the escorts today.  I hope that carrier hangs around!  We come out ahead here.
|
|
#
?
Nov 15, 2017 18:49
|
|
|
Grey Hunter posted:
Destroyed on field: 11  Also after sighting the Independence? Near 200-point score increase. Jobbo_Fett fucked around with this message at 19:07 on Nov 15, 2017 |
|
#
?
Nov 15, 2017 19:05
|
|
|
Am I correct that the AI just used two seaplane tenders in the Jaluit 'invasion'?
|
|
#
?
Nov 15, 2017 19:05
|
|
|
|
| # ? May 17, 2024 01:53 |
|
|
Pershing posted:Am I correct that the AI just used two seaplane tenders in the Jaluit 'invasion'? Destroyer seaplane tenders
|
|
#
?
Nov 15, 2017 19:08
|
|