Given DK has that kill screen, at some point there is just simply what is the absolute maximum possible score given a perfect game with perfect luck... I wonder what that score is... and how close the record is to it now... how much wiggle room remains at this point?
> Wow! > > Given DK has that kill screen, at some point there is just simply what is the > absolute maximum possible score given a perfect game with perfect luck... I wonder > what that score is... and how close the record is to it now... how much wiggle room > remains at this point?
I've been wondering the same thing.
These questions sound like they might be right-up Don Hodge's alley.
If you have a problem, if no one else can help, and if you can find them, maybe you can hire... The D-Team.
I think there isn't such a thing as a 'perfect game' in DK, because you can hang out and wait for more barrels to be produced (and firefoxes, and perhaps somewhat firballs, once you've hammered them). It seems that's where the majority of extra points are, which has to balanced with the remaining bonus. Guestimate?
Consider it high comedy....sincere tragedy....whatever...don't take it personally.
> Wow! > > Given DK has that kill screen, at some point there is just simply what is the > absolute maximum possible score given a perfect game with perfect luck... I wonder > what that score is... and how close the record is to it now... how much wiggle room > remains at this point?
The search space of possible Donkey Kong games, while finite, is almost unfathomably large. I'm skeptical that the "best possible" game is a solvable problem. However, I could imagine a DK-playing AI that would (in an emulator) look ahead at the outcomes of various moves (essentially playing and then rewinding to a previous save state in case of a bad outcome), allowing it to take incredible risks through the entire game for huge rewards. I imagine this would rack up scores far higher than any human realistically could--perhaps 2,000,000 or more. Of course, making such a program would be a huge effort and not particularly practical.
Quote: The search space of possible Donkey Kong games, while finite, is almost unfathomably large. I'm skeptical that the "best possible" game is a solvable problem. However, I could imagine a DK-playing AI that would (in an emulator) look ahead at the outcomes of various moves (essentially playing and then rewinding to a previous save state in case of a bad outcome), allowing it to take incredible risks through the entire game for huge rewards. I imagine this would rack up scores far higher than any human realistically could--perhaps 2,000,000 or more. Of course, making such a program would be a huge effort and not particularly practical.
I thought the routines that handle the attract modes already did this sort of thing.
Consider it high comedy....sincere tragedy....whatever...don't take it personally.
>The search space of possible Donkey Kong games, while finite, is almost unfathomably large...I imagine this would rack up scores far higher than any human realistically could--perhaps 2,000,000 or more.
Do you really think so? That they are still as far as only about 50% of a potential maximum? Personally I would IMAGINE (I can only imagine since I've never watched a record setting DK and it's techniques, or observed their loss-of-opportunity moments despite setting a record) that the top players are already maximizing every potential opportunity to control and group barrels, maximizing hammer-use and taking advantage of known exploits to achieve all they can... taking any risk necessary to achieve a top score. It's not like they are avoiding risk since the consequence is simply a restart. Unlike pacman, which has a defined maximum score... dk has SOME randomness to it due to barrel behavior and grouping... but still... at some point there is a maximum number of barrels in the time allotted... a maximum number of opportunities to group the barrels... and a maximum amount of time and levels. To come up with a theoretical maximum would be a matter of counting how many barrels are possible in the time on each level... assume the maximum score is produced by every barrel, and derive a theoritical maximum from that. Someone here must care enough about DK to crunch that basic formula out.
You may be right... but seems to me, they are already approaching that asymtotic maximum potential score.
> > Wow! > > > > Given DK has that kill screen, at some point there is just simply what is the > > absolute maximum possible score given a perfect game with perfect luck... I wonder > > what that score is... and how close the record is to it now... how much wiggle room > > remains at this point? > > The search space of possible Donkey Kong games, while finite, is almost unfathomably > large. I'm skeptical that the "best possible" game is a solvable problem. However, I > could imagine a DK-playing AI that would (in an emulator) look ahead at the outcomes > of various moves (essentially playing and then rewinding to a previous save state in > case of a bad outcome), allowing it to take incredible risks through the entire game > for huge rewards. I imagine this would rack up scores far higher than any human > realistically could--perhaps 2,000,000 or more. Of course, making such a program > would be a huge effort and not particularly practical.
People have programmed AI bots for games like Guitar Hero.
You let a geek get bored, and you end up with the weirdest, most useless stuff.
> I thought the routines that handle the attract modes already did this sort of thing.
Unlikely. Attract mode AI is always either playback of canned input or very, very simplistic. Especially on DK ROM space was super expensive (and it's not like they're going to run a neural net on a Z80).
> People have programmed AI bots for games like Guitar Hero.
There is no AI involved in that, you just have to detect the gems coming down the note highway and trigger a solenoid at the right time. If the bots also did things like strategic use of overdrive and whatnot, that'd be AI (but none of the ones I've seen actually do that).
> Unlikely. Attract mode AI is always either playback of canned input or very, very > simplistic. Especially on DK ROM space was super expensive (and it's not like they're > going to run a neural net on a Z80).
Beyond Dark Castle on Mac seemed to have a half-decent AI for the demo.
> > People have programmed AI bots for games like Guitar Hero. > > There is no AI involved in that, you just have to detect the gems coming down the > note highway and trigger a solenoid at the right time. If the bots also did things > like strategic use of overdrive and whatnot, that'd be AI (but none of the ones I've > seen actually do that).
Wouldn't a similar process be used if someone tried this with a video game? Just let it look at the screen and see where things are, and act appropriately. AI bot was the first thing that came to mind, sorry about that.
> > The search space of possible Donkey Kong games, while finite, is almost unfathomably > large...I imagine this would rack up scores far higher than any human realistically > could--perhaps 2,000,000 or more. > > Do you really think so? That they are still as far as only about 50% of a potential > maximum? Personally I would IMAGINE (I can only imagine since I've never watched a > record setting DK and it's techniques, or observed their loss-of-opportunity moments > despite setting a record) that the top players are already maximizing every potential > opportunity to control and group barrels, maximizing hammer-use and taking advantage > of known exploits to achieve all they can... taking any risk necessary to achieve a > top score. It's not like they are avoiding risk since the consequence is simply a > restart. Unlike pacman, which has a defined maximum score... dk has SOME randomness > to it due to barrel behavior and grouping... but still... at some point there is a > maximum number of barrels in the time allotted... a maximum number of opportunities > to group the barrels... and a maximum amount of time and levels. To come up with a > theoretical maximum would be a matter of counting how many barrels are possible in > the time on each level... assume the maximum score is produced by every barrel, and > derive a theoritical maximum from that. Someone here must care enough about DK to > crunch that basic formula out. > > You may be right... but seems to me, they are already approaching that asymtotic > maximum potential score.
Well, 2,000,000 was definitely a guess, but doesn't to my mind seem unreasonable for a theoretical maximum (probably only actually reachable by a save-state-wielding AI like I described). According to this article, Dean Saglio scored 1,136,400 in MAME, so there's a lower bound. One could also, if they had access to a lot of high-score videos, track the record score for every individual level and add them together. I think you're under-emphasizing the consequence of losing a like during a Donkey Kong high-score attempt. Sure, you can start over, but a full game takes a couple of hours. If a player played as aggressively as possible, they'd never reach the kill screen. Things like barrels and fireballs are too unpredictable (unless, say, you're a shamelessly cheating AI that checks the program RAM to figure out the next move). In truly uninhibited play, you could do things like repeatedly jumping back and forth over fireballs, not worrying about them turning the wrong way. You could also strategically lose lives, finding the level you can score highest in and playing it three times until time ran out. It seems like most high-score players have struggled just to reach the kill screen.
You might get somewhere with tracking all of the possible ways to score points, and trying to place a limit on how many times they could occur in each stage. This could at least get you an upper bound, though that would probably be more than could even theoretically be achieved.
That's pretty awesome.. Saglio's MAME DK score is verified by Twin Galaxies, so maybe your guess of 2 million isn't so far off after all! Saglio could, I would imagine (I got a real active imagination ) at some point dump all the contenders should he ever repeat his score on a real DK machine. Any body know is he's ever shown up at a live event?
I'm used to getting at least eighty grand on an off day, though have peaked at over 200k even on MAME after the video re-write. I played DK on 139u3.1 tonight. At first, it kicked my fucking ass. I had trouble initially breaking thirty thousand. It plays different from earlier versions.
So. If Saglio is playing current MAME, I'd really like to see him get on an original machine.
> I'm used to getting at least eighty grand on an off day, though have peaked at over > 200k even on MAME after the video re-write. I played DK on 139u3.1 tonight. At first, > it kicked my fucking ass. I had trouble initially breaking thirty thousand. It plays > different from earlier versions. > > So. If Saglio is playing current MAME, I'd really like to see him get on an original > machine.
What plays different to you?
A LONG time ago the input method was changed to support something called steadykey iirc. When this was introduced something about the way the keys are read is different. If you play using a real joystick with a restrictor plate, you wouldn't notice a difference. If you used a keyboard, it makes it harder than hell to get up those ladders. You are probably used to hitting the UP button just before the ladder while still running. Since you were already holding the LEFT or RIGHT button down, Jumpman doesn't climb the ladder.
If you compile your own MAME, you can change make a change hack to drivers\dkong.c. Look for "PORT_4WAY" and change it to "PORT_8WAY". I know it is a hack, but it makes the game playable on a keyboard.
EDIT: This may have been longer than I thought. I found this in whatsnew_037b14.txt...
Quote: What plays different to you? A LONG time ago the input method was changed to support something called steadykey iirc.
No no. I mean the barrels and firefoxes appear and behave differently. I've noticed this since some time after the video re-write, but there seems to be some very recent serious difference.
Index 
Threaded 
Wiebe's got the record
Reply
[Re: 
) at some point dump all the contenders should he ever repeat his score on a real DK machine. Any body know is he's ever shown up at a live event?
