a white rabbit
Posts: 2366
Joined: 4/27/2002 From: ..under deconstruction..6N124E.. Status: offline
|
quote:
ORIGINAL: vonPryz quote:
ORIGINAL: a white rabbit i'm suggesting more storage space, so Elmer has a decent memory for comparison. As you wrote above 1 big save is running at 800KB so 10 moves, his own and the player's is 16MB ish, then he needs somewhere to store his own provisional planning unless you want to wait for him to recalculate evry time you load, so that's some 20MB, then he needs space to try out the planning in theory during play, no idea there.. So I gather you are purposing something like the libraries chess programs use? Those contain most if not all variations about common opening strategies and countermeasures. It might work to some degree against humans: if one is often using certain tactic, the AI might recognize it and act accordingly. quote:
ORIGINAL: a white rabbit the best iGo program was some 3.5 GB -- the "thinking" bit, the .exe if you really wanna get picky was some 150KB Most of that data is made of well-known games and patterns. Still, go has way smaller decision space than TOAW or other wargames. For any kind of AI, the binary size hardly matters. Algorithms usually fit into smallish space, but supporting data is the one taking lots of space. Then again, disk access is tremendously slow in retrospect to RAM, so any deep analysis requires quite a lot of memory. quote:
ORIGINAL: a white rabbit It's still a y/n system, no maybe.. The obvious difference between wargame and most of the boardgames is that moves do not have fixed outcome. That is, the system is at least in some parts definitely "maybe" instead of "yes/no". In chess you can always capture opponent's piece - unless the move is illegal. It doesn't matter at all whether the capturer is a pawn, a bishop or even a queen - legal capturing move always leads to capture. Of course, sometimes capturing isn't too smart thing to do, but that is another a matter. Wargame move outcomes are often more versatile. Can my 2nd armored regiment break the opposing mech infantry batallion in order to establish a bridgehead? It might succeed or it might not - there is no definite answer. Taking the strengths and supporting units into account, one might estimate that the chance for success is, say, 70%. In the chess case it is relatively easy to calculate all the possible moves. The outcome is certain for each legal move, so multiple outcomes don't need to be worried about. Now what happens to the outcome calculation's branch that is trying to find out my next move after 2nd armor attacked? It has to create several branches for both success and failure. How much damage did the unit take in attack? In which direction did the enemy unit withdraw? Did we discover more units? How strong are those? Do they pose immediate risk to my unit or future threat to my units? All this makes the decision tree to grow a LOT, so min-max isn't the answer. -P ..but there are only ever a limited number of "legal" moves on a map, a btn without bridging equipment can't cross a major river, but go up a level and the div, as a single unit, that has a bridging unit as part of the zoom-out OOB can, no bridging unit and it can't either. A layered approach simplifies much.. ..ok, combat, at best these can only be worked out on a best-case basis, as it is now, Elmer doesn't attack at bad odds unless forced by the formation orders, so the "legal" definitions already exist.. ..new units appearing following an attack, how long do you want to give the AI to recalculate and compare with previous stored details, that's all. Are we playing speed-toaw, are you being penalised for taking more than 10 minutes to make a move ? No, then why should Elmer suffer ? ..now known patterns...yup, i agree, the iGo uses a lot of space for previous games, but iGo consists of a limited number of patterns? / moves ? / placings ?, as does a wargame, really. You can take your armoured div thru the Pripet marshes, but the chances are good you'l feed it on to open tank ground. The basic battle formations were codified in the 1700s, even tho they existed before then, refuse left/refuse tight/refuse center, advance right etc etc and in all truth, they're still used today (i can hear the screams now), form a square to repell cavalry, or box-defence is i think the modern term, and repel the NVA, same formation ; guns at the corners and center, kneeling and standing ranks of bayonet armed infantry, hordes of screaming Zulus crashing into the mine-fields and triggering the automatic computer-controlled weapons, ballista teams feverishly winding up the propulsion, legionaries lobbing pilums then womp as the Greek-fire pours from the A10s (whatever) onto the attacking Russian hordes.. ..maybe we'll need some patterns, but there's some really good gamer's around who can draw those, or the designer can do it, and there's always the historical version, we can give Elmer patterns, scen by scen, we got the space, in spades, so please, let's use it.... ..i'm 52, i've wargamed since i was 9-10 (ahhh, match-stick firing cannon, the easy days), i've dreamed of a machine that could play a decent, me-hammering game with no cheats since 1985, a major future for gaming is in a decent AI, we got the machinery, now let's get the thinking right, cos we got the machinery to do it...
_____________________________
..toodA, irmAb moAs'lyB 'exper'mentin'..,..beàn'tus all..?,
|