Thread in 'Discussion' started by tepples, 29 Dec 2007.

  1. tepples

    tepples Lockjaw developer

    Tetris The Absolute: The Grand Master 2 is emulated in some hard-to-find versions of MAME. But has it been subjected to a tool-assisted speed run of Master or Death? This would give a lower bound on how well one could possibly play.
  2. As far as I know, there isn't one.

    It's extremely hard to find a version of Mame that runs TAP, and doesn't have input/rendering lag, and can be used to tool-assist.
  3. Muf


    Didn't someone have a source code diff that can be applied to vanilla MAME source code to get it to run TAP? That way you could compile your own MAME and tool assist.
  4. I used to have a video (avi) of one.
  5. I wouldn't stop anyone from trying, but I would advise them to label the video with a note saying the video is tool-assisted.
  6. I have a video of death completed in 4:32 that is pretty much impossible to tell apart from a TAS for large parts of it, so I doubt anything below 4 minutes is possible.

    We'll probably never know for sure though since unlike most TAS attempts, it won't matter how much time and patience you have - if you're not already sufficiently competent at the game, a perfect (or even efficient) playthrough is nigh impossible.

    The few people truly capable of the doing a TAS justice are unlikely to be motivated to do so.
  7. "We'll probably never know for sure though since unlike most TAS attempts, it won't matter how much time and patience you have - if you're not already sufficiently competent at the game, a perfect (or even efficient) playthrough is nigh impossible."

    Good TASers are always better at making a TAS for a game than the best player of that game could (make a TAS for the game). Don't ask me why. It's just from personal experience. I bet it could be done in less than four minutes with proper re-recording / frame stepping software. It's just a matter of a good TASer learning how the game works. He'd probably wind up teaching us a thing or two after he's done, too. You'd be surprised.
  8. tepples

    tepples Lockjaw developer

    Assume it takes an average of 5 frames to shift and lock each tetromino. Then for each section, take ((ARE + 5) * 75) + (lineClearDelay * 7.5). Add it all up and you have an estimate of the theoretical best time. Let's plug in the Death numbers from the wiki: average ARE = 9.1; average line clear delay = 3.9.
    Then (((9.1 + 5) * 75) + (3.9 * 7.5)) * 10 = 10867.5 frames, or just over three minutes.

    In the extreme-slow-motion recording typical of a TAS, it's a bit like playing Lockjaw set on Arika + no floor kick + Death + 2.0 second lock delay. In fact, it's even easier, as you can get 100% efficiency at synchro.
  9. I'm not saying a TAS couldn't beat the current unassisted best, but to be done properly it would need to be exclusively tetrises to minimise line clear delay (except for 100-300 of course).

    It would take considerable effort (and by that I mean months of practice) for someone who hasn't played 20G with Arika rotation to reach the competence required for all tetrises, even with the ability to pause. Hell, most people unfamiliar with TAP would probably struggle to even survive even with tools.
  10. Yes, but if you don't suck at the game, then you have a sizable advantage. And remember, the current holder all 3 TASes is a fairly strong tetris player. Defintiely Death 300 ability, maybe even higher...
  11. tepples

    tepples Lockjaw developer

    The NES Tetris TAS goes back-to-back-to-back, though it does run at 0.5G and manipulate the randomizer.

    Even with the foresight of knowing all 720 or so next pieces?
  12. Foresight is correct, as luck cannot be manipulated very much. Once the game has started, the list of pieces is set in stone.
  13. Completely irrelevant due to the different ruleset and absence of "luck manipulation" in TAP.

    Yes. Please don't make me repeat myself. I don't wish to sound overly harsh but you seem to be underestimating the challenge of 20G with the movement limitations of TAP. When the next piece becomes unplayable, an inexperienced player won't know how far to backtrack or even which earlier moves were mistakes.
  14. i don't think you're getting the difference between playing live and TASing. with the TAS, you simply try every piece over, frame by frame, until you get it perfect. there's zero time pressure. all the guy would have to do would be to look at some videos, and he'll get a grasp on how he needs to manipulate blocks in certain situations. if he comes to a place where he would break the b2b tetris, then he backs up and corrects it. if that doesn't work, he tries again, and again, and again.
  15. tepples

    tepples Lockjaw developer

    Better yet, the player doesn't backtrack; instead, he writes a program to do all the backtracking for him. Plenty of AI students have written Lisp programs that search several moves ahead in the game state of Chess. Likewise, TAS jockeys have written programs to search game trees for advantageous states. For instance, large parts of one of the speedruns of were played by a robot called BisqBot. (Read the interview.) Like Chess, TAP with a predefined tetromino sequence is a game of perfect information. But unlike in Chess, the opponent is predictable. The movement limitations of TGM rotation at 20G appear to make the search tree complexity even less than that of Chess, which would appear to fall within the reach of A*.
  16. I can see I'm not getting my point through effectively, so I'll just reiterate that the lack of time pressure is irrelevant. I know how TASing works, and I'm saying that trying over and over again will be useless. Perfect stacking in TAP (or any tetris game that doesn't use the 7-bag for that matter) has not been solved, even with complete knowledge of the pieces to come.

    Efficient and clean stacking under these conditions requires a level of competence likely beyond anyone who would be inclined to attempt a TAS. It's not like megaman where if the player misses a jump by a frame he can just rewind. If the player is fundamentally incapable of identifying poor moves, how can he correct them? He will literally keep hitting dead-ends indefinately. It really doesn't matter how much determination or patience the person has, it just won't work.

    And as for writing an AI to look for optimal moves - I challenge anyone to write an AI that can even beat Death mode.
  17. DTET 20G AI has already been done so I doubt writing an AI that can just complete death mode would be much trouble at all.

    TAP with tools is very straightforward, since you have a sequence of pieces that you KNOW you will be using (assuming you start at the same time every time) so like it or not, luck manipulation won't be a factor (the time taken to pause/delay the next piece would also be a factor.)

    You have a clear list of things you can and can't do, everything is well-documented. Even for someone who's never played a TGM before I doubt it would take more than a few days (much less several months) to get the hang of things .

    How about planning everything out in advance before even touching the game?
  18. This doesn't make any sense to me - it's impossible for the reason you just quoted.

    The argument I've been trying to make is that there probably aren't any TGM players in the TAS community. If a decent TGM player decided to commit the time and effort into it, it might be possible.

    Otherwise, I just don't see someone who hasn't played the TGM series before being able to do an efficient TAS, even with a complete piece list and all the time in the world at their disposable. A complete piece list is not much help if you simply can't stack cleanly, and a novice just won't recognise the mistakes to be amended. Becoming competent with the rotation system in just a few days seems ludicrous to me, although I'd love nothing more than to be proved wrong.

    I'm not familiar with DTET, although I know it has more wall kicks and double rotation. Also, I assume it's a versus AI - will it survive for 5+ minutes on it's own? Either way, it must have been written by a player. Someone not already familiar with the game would struggle to write a similar AI for TAP.

    Getting back on topic however since I don't like being so negative, if writing an AI that can attain death GM would be as straightforward as you think and of course there is no reason for that AI player to waste any frames (other than the extra frames for line clears below tetrises) - that same AI should theoretically set an excellent target time for a TAS to beat. Know anyone that could do it?
  19. Actually...
  20. This thread is sort of confusing. What's wrong with taking the world record as the lower limit to try for?

    And in terms of small technical details that would be exposed by a TAS... That's a rubbish argument too. TAS'ing the game is entirely unnecessary, all you have to do is probe the game deeply. In fact, DAS (oh god acronyms) works much differently than most players realise, and it interacts with ARE in an interesting way that does in fact affect high level play. If this isn't on the wiki yet, expect it to be soon.

Share This Page