What is Yomi?

Discussion in 'Game Design' started by Kdansky, Mar 24, 2014.

  1. Kdansky

    Kdansky Well-Known Member

    I just had a radical idea. After Keith got banned at FS when he said (in different words) that Yomi isn't a skill. I think there is some truth to it, because really, "I think that you think that I think that you think [stop at the Yomi point]" is quite out there. Humans can't read minds.

    But when we get down to it, that skill is only required in actual RPS, and we usually don't play RPS, though I could go on a tangent here about Japanese culture, but I digress. What games like SF4 or Yomi (The Game) actually are, is weighted RPS. Rock (combo starter) gives you 10 points, Paper (shoryuken) gives you 3 points, and Scissors (Throw) gives you 5 points. As was pointed out, the ideal strategy is not All-Rock, because that strategy loses to all-paper quite badly. The ideal strategy is a mix of all three, where you play them more often if they are better. In our example 10 / 18 Rock, 3 / 18 Paper, 5 / 18 Scissors. This will give you the maximum against any opponent who cannot read your mind.

    But we're not playing weighted RPS. We're actually playing weighted RPS where the numbers are hidden. You never know if your opponent has a big combo in hand in Yomi (the game), and you don't know exactly how much damage a combo in SF4 is going to do, because it might get dropped due to execution error, and you don't know how much meter your opponent is going to spend, and whether or not he knows that one character specific combo, or whether or not he can get the 1-frame links to connect. I'm sure you get my point: We know it's weighted RPS, but the numbers are heavily obfuscated.

    And here is where the skill of Not-Yomi comes in: It's not so much about reading minds, but about estimating the numbers correctly, and playing along them. Just like in Poker, you need to know the statistics, and unlike Poker, you need to do the math in the span of milliseconds. And then on top of that you need to be able to RNG in your head, which is something humans are incredibly bad at [this is a scientific fact, see google]. That's my theory: Yomi is misnamed, it's not about reading the opponent. It's about pretending to be a computer, and knowing the numbers.
    Jon Perry and Nachtfischer like this.
  2. pkt-zer0

    pkt-zer0 Active Member

    Well, you might want to put a different name for the skill you described, but that doesn't make "yomi" a misnomer. Conscious/unconscious pattern recognition being described as "reading" seems fitting, I would say.

    Suppose you're playing plain old RPS against a guy who goes R, P, S, R, P, S, R, P, S... as his attempt at the least exploitable strategy (all three options with equal probabilities, in this case). If you are able to recognize this pattern, and modify your play accordingly, you can do better than if you don't, and just keep playing randomly, as the optimal strategy dictates.

    In practice, usually the patterns are a lot more subtle, especially if there are no/lax time constraints, since they are more easily obfuscated then. But still, it is a thing. Why would it not be? I mean, the counter-claim would be that if you're playing against a guy who plays rock 100% of the time, you still can't win more than a third of the time, because pattern recognition does not give you any sort of edge whatsoever.
    Dasick likes this.
  3. Leartes

    Leartes Well-Known Member

    Ok, I just deleted a longer post because some things became clear to me while writing.

    pkt-zer0, I think you consider the wrong strategy as optimal. I mean, "optimal" in the classic sense means that there is no move in the whole strategy space that exploits you. In practice your opponent does not play the whole space. In poker for example, he is likely to fold 37. Now your optimal move changes (though perhabs 37 has 0 probability in the equilibrium, thus changing nothing). The classic equilibrium strategy is not optimal if you restrict the space in such a way that your opponent may never have 37 in a certain gamestate.
    In your example, if I play someone that does R, P, S, R, P, ? then I would give him more than 1/3 to play S next and thus alter my response.

    I think this transfers to yomi, but not to fighting games. Fighters are much more psychological since people don't have time to think, calculate and ask random.org.
  4. Kdansky

    Kdansky Well-Known Member

    I was thinking specifically about Yomi The Game as an example, because it lacks execution and time constraints. Fighting games are chock full of other stuff (execution, timing, reflexes, recognising moves, remembering setups) which trump all the Yomi you could ever do.
  5. pkt-zer0

    pkt-zer0 Active Member

    Leartes: I'm not sure if this is what you were getting at, but the repeating R, P, S sequence was not intended as an example of optimal play (hence "attempt at the least exploitable strategy").
  6. evizaer

    evizaer Well-Known Member

    There's enough wiggle-room in answering "what should I play in this MU + board situation?" combined with cards available in hand: you can actually gain an advantage from understanding the biases of your opponent and playing to counter them. This lets you do better than perfectly weighted random play--but it has to be customized to the meta and your opponent's biases. I think that's a pretty strong indicator that yomi is a thing. "Reading the opponent's mind" is a way of expressing this process.
    deluks917 likes this.
  7. Nachtfischer

    Nachtfischer Well-Known Member

    It seems like this only really works if we assume that only one of the players is doing it. If they both are, then we indeed run into the kind of arbitrary "A thinks that B thinks that A thinks that B thinks..." stuff.

    An example to illustrate: In Hearthstone there's often the decision of attacking the opponent's hero directly or attacking the minions he has on the board. Some players are really biased to "go for the face" (attack the hero). Whether they're just bad or they have a deck designed around rush-down doesn't matter. Point is, you observe that they almost always attack you directly instead of clearing the board. So you could try to start "reading the mind" of your opponent. With the thesis in mind of "he will probably not remove my minion" the best decision might indeed change. You might just want to play a minion, that your opponent could easily remove (and you think that would be bad for you, so you wouldn't play it without the reading) but you just assume that he won't. So you play it nevertheless.

    Now, what if your opponent starts doing the same thing? He would go: "Huh, I think that he thinks that I think going for the face is the best idea. So, this time I won't." Or would he? He might also go: "Huh, I think that he thinks that I think that he thinks that I think going for the face is the best idea. So, he might suspect me to drop that behavior. So, I will do it all the more!" At that point it's just pure guessing when your opponent stops following that.

    I know that Hearthstone isn't the best example game for mind reading, but that example just came to mind, because I actually saw it happen in some video. Obviously you have a lot of more important information in that game. It's mostly about action/reaction and knowing what the opponent could have (instead of what he "generally likes to play").

    In a game like Yomi however, there's much less of that. It's simultaneous action and therefore theoretically much more about that inifnite "I think he thinks..." chain. Not sure how much that applies in practice, though, as the whole "infinite chain" approach is kind of impracticable in itself.

    Interestingly the self-proclaimed (I assume he's not lying) number 1 on the Yomi online leaderboards recently wrote this:
  8. Leartes

    Leartes Well-Known Member

    You missunderstand what I wanted to say. It is reasonable to define the optimal play as the (mixed) strategy with the highest expected outcome. Now the question is over what space is this expectation? In classic game theory (e.g. nash equilibrium) you assume your opponent also plays optimal and you end up in an equilibrium where the expected outcome is the value of your game. In games that occur in practice this is not neccessarily the case. If your opponent shows you non-equilibrium play then your optimal strategy vs this opponent changes since e.g. exploitive play leads to a higher outcome vs this opponent.
    If you name this "yomi" then yes, yomi exists in all sorts of games. I think I'm missing a proper mathematical definition of yomi :D
  9. pkt-zer0

    pkt-zer0 Active Member

    Proper mathematical definition, you say. The edge you gain from yomi is the information value of the gamestate history in a match. So, the difference between your answer to "what is your move?" and "your opponent played R, P, S, R, P; what is your move?". Maybe that's too broad a definition, though. Hmm.
    Leartes likes this.
  10. evizaer

    evizaer Well-Known Member

    I think it's the difference between"what is your move?" and "your opponent played R, P, S, R, P; what is your move?" MINUS how the hidden information that was revealed by that sequence affects your valuation. (Like what your opponent has in their hand that you know about, what's in their discard pile, the interactions they could exploit based on abilities on their cards, etc.)
    Leartes likes this.
  11. Leartes

    Leartes Well-Known Member

    Yes this seems to cover what I was trying to say. First I hung up on the "you do this, I do that, but you know that I know ..." thing too much. Seeing it the way evizaer describes it, it is pretty much what I did all the time when I played poker. I mine a lot of data from a site (or ask a friend for their database) and adjust play to the tendencies observed in the opponents.

    Do you think that there is much yomi in a single game of yomi vs a complete stranger (of equal skill level)?
  12. evizaer

    evizaer Well-Known Member

    I am not particularly qualified to comment.

    I would guess that you need non-trivial play data--the kind of data that comes from knowledge of meta--to read someone. You need to know that they are in the meta before using the info about meta. So there is going to be a feeling out phase with a stranger where you assess their conformance with the meta while monitoring for other patterns. Hopefully you are playing a BoX series, so you can actually start picking up on potential exploits.

    The meta acts a pre-packaged set of assumptions to feed your reading engine so that you have something as a base, and don't have to just rely on actual play observations 100%. I'm guessing that this will noticeably speed up pattern detection--it may not, though, because both players are aware of it and need to pick up on player-specific patterns within the gamut of the meta, anyway. Also the meta involves all MUs, and in a game that doesn't receive a REALLY HIGH amount of play, a new MU to you can throw your meta knowledge in the trash.
  13. Disquisitor Sam

    Disquisitor Sam Well-Known Member

    This actually only works out to be statistically true. Ideally, you want to throw 10/3/5 - R/P/S as you say, and you want to throw them in a random arrangement so you cannot be predicted. This is more relevant to Rock, Paper, Scissors than you think - it is commonplace for competitors in such tournaments to randomly determine their play order using a computer prior to even showing up. Then it's just a question of memorizing your order. Then more people start doing it because they pretty much have to, and the whole game falls apart.

    But that only happens when there are no limitations on the move that you choose and a fixed value on each. In reality there's a lot more going on than just choosing a move as randomly as you can. Certain moves are not as good in terms of doing damage but have spatial advantages instead. If a situation arises when a specific move would get you free damage, you'd definitely want to use it, even if it's not a lot of damage. Now change the situation to where you won't get free damage, instead you might get some free damage (the opponent can still counter). Really it's not free damage in this case because the cost is the risk that the opponent has set up the situation to bait you. If you only consider damage in your valuation of given moves, you might get away with the preordaining strategy. But you couldn't possibly have calculated which spatial moves were going to be valuable when. This forces you back into playing the game for real - you have to make a human decision when the unforeseeable big moment for a big (or little!) move presents itself. And then when that happens, you've messed up your preordained move set. Do you start over? Make a substitution?

    I've thought about playing Yomi online via the website, but I eventually decided not to. It's actually not very much about the occasional player who's bound to be using a RNG and making for an unsatisfying match, because I think that most Yomi players aren't okay with winning games randomly - they want to distinguish themselves. I think a much bigger reason is that I think the game loses a lot when not played face to face. Body tells, acting, and the like might be enough to shake up the match significantly. Even if a profound acting performance fails to convince the opponent to stop playing preordained, the opponent can make an act of his own to convince the first guy that his acting worked.
  14. keithburgun

    keithburgun Administrator, Lead Designer Staff Member

    Yomi is misnamed, because the card game Yomi isn't actually about reading. Further, "reading", while not completely non-existent, is not enough of a thing to base a game upon.

    Yomi has a lot more to it than reading, and some have said that it should actually be called "Hand Management" or "Valuation". Those aspects of the game, I actually like.

    The fact that high level players play with an RNG sometimes and it either helps them, or doesn't seem to hurt them, means that they should just do that, and probably everyone should just do that. The game should probably have this as a baked-in rule and formally get rid of the mostly fake reading myth somehow (would be a design challenge).

    I've talked before about "game placebo" and how people tend to see patterns where there aren't any (Evizaer has a little video up on that in the FS thread). I think in the case of Yomi, the game kind of depends on players' willingness to have some part of themselves that "Believes" in the Reading part. It's somewhat slightly taken on faith. And the way that Sirlin & co respond to any of what's being said here is also pretty telling - they basically do this elementary school behavior thing of "YOU'RE NOT PLAYING RIGHT!", the complete opposite of the Play to Win message. Or they try to say that using an RNG is "cheating", even though whether or not you use one is like 100% not enforceable.

    The pattern recognition thing is also basically horse-shit, because we're in a situation where another human, reasonably close to yourself in his basic intelligence, is trying to have an unreadable pattern (Basically, trying to play somewhat randomly). Yeah if someone were to COMMIT TO "RPSRPSRPSRPS" you could exploit that. No one does that, or anything like that. People play patterns that they themselves don't even understand, so this idea that a different player is going to understand it is ridiculous to me.

    If you think you have a "read", or even that you "guessed their pattern", what makes you think this? Is it the fact that you indeed played a Throw when they Blocked? Because that also will happen if you're just playing random cards.

    In short, players play cards, without really knowing why. The cards match up, and a combat winner is pronounced. It's not random, but it may as well be random.

    On hard data: I've been interested to know for awhile the combat win % of certain players. I'd like to see if that varies much at all for any given player. My feeling is no one gets it above like 60%, and those who have something like 60% just haven't played enough games to bring it back down to 55%ish.
    HairMachine and deluks917 like this.
  15. Disquisitor Sam

    Disquisitor Sam Well-Known Member

    Let me offer an unrelated example and see where people's thoughts lie. In World of Warcraft PvP, caster classes cast spells out of unique schools. A mage has the fire, frost, and arcane schools, which is distinct from ability trees (as if I needed to mention that WoW is hardly elegant, efficient, or even sensible). A shaman has abilities in the "Elemental" tree, focused around offensive spells, but different spells in Elemental fall into Nature, Fire, or Frost schools.


    Certain spells have cast times before they take effect, where the caster must not move and clearly telegraphs casting a spell. He may stop the cast time at any point for no mana or cooldown cost, just the time he spent casting. This is significant because certain classes have "interrupt" abilities, abilities that when used will not only stop the cast but then further prevent the casting of any spell in the entire school for some time after that. The catch is that you have to actually interrupt the cast. If you interrupt while your target is doing nothing, you have no effect, and you waste the cooldown on your interrupt, leaving your opponent free to cast spells unmolested.

    These dynamics give rise to a kind of meta strategy called "juke casting." The first order strategy is to just cast the spell and hope it goes off. Depending on the proximity of the interrupter class, this is easily counterable since the cast is clearly telegraphed. Second order strategy is to try to bait the interrupt. You start casting the spell, but then stop casting some amount of time into the cast. Due to reaction times and server lag, it's quite common for the interrupter to throw the interrupt late and not catch the cast at all, after which point you can just cast with impunity. You have "juked" him. But if you discover your opponent is good enough to know about juking, he's not so easy to bait. He knows not to interrupt the moment he sees a cast.

    So you'd think that third order strategy is to just cast without juking and take advantage of his hesitancy to interrupt. However it's not quite that simple. In the likeliest example of a 1.5 second cast time, this is still a pretty long span of time for good players, as you might imagine. The difference between second and third order strategy isn't really binary, but rather a battle of willpower along a spectrum. It's kind of a test of nerves: every millisecond that you leave your cast time running increases the likelihood of being interrupted, but also that much further tempts your opponent to throw the interrupt. If you're trying to juke your opponent, but juke too early and he doesn't throw the interrupt, it's that much more time that you go without casting the spell you want, time in which you are probably being damaged or some other time sensitive thing is happening without you. If you juke too late, you get interrupted and you don't get the spell at ALL for several more seconds, which is even worse as it might be a game ending moment. The opposite is true for your opponent: too early and you get to free cast for several seconds, too late and you get your spell off after all (though the opponent still has his interrupt ready for the next time). Throw into this the idea of potentially three different schools of spells and a whole other kind of "juke" where you lead off with a school you don't actually care about and try to bait him into interrupting that, and you can see it gets pretty messy.

    I feel like this idea is a lot less subject to the preordained strategy that was discussed earlier. It carries with it a spectrum of decisions rather than just an RPS counter. But it still seems linked to the idea of reading and predicting your opponent. What are people's thoughts on this?
    Last edited: Mar 24, 2014
  16. Disquisitor Sam

    Disquisitor Sam Well-Known Member

    By the way, Keith, thanks for clarifying your position on this topic. You cleared up a misconception that I had about your thoughts on this.

    That said, I'm not sure that I'm ready to accept that position just yet. I think the things you say are valid criticisms and are well grounded. I'm definitely on board with the part about game placebo and that humans tend to see non-existent patterns because of how we work. However, that doesn't make your explanation completely conclusive. Yes, humans DO that. The question I'm trying to answer to myself is "Are they doing it in this case?"

    I'm not saying you're wrong, or even illogical. I'm saying that I personally do not have a complete enough knowledge of psychology to make a call on this. I think I'll need to do some reading on related topics to see if false patterns are the only thing at play here.
    keithburgun likes this.
  17. Leartes

    Leartes Well-Known Member

    I find this interesting mainly, because I know this to be a concept in many games (e.g. guildwars, dota). Especially in the dota laning stage this is especially important as you even cancel your normal attacks to make your opponent miss hits. (e.g. you attack a unit that dies in 2 hits, your opponent will do the same so that he can get the lasthit before the creeps kill it, then you cancel your attack and if he goes through he is in attack cooldown afterwards and will 100% miss the lasthit). It also occurs often in fights with some spells that have a long cast animation. In this game we rarely have hard interrupts, but we have tons of options for counterplay. It is very important with mobility skills where you want to bait them out to make your target jump into a bad spot when you don't have the means to get them without a mistake on their part.
  18. CrystalChaos

    CrystalChaos New Member

    Yomi isn't mind reading exactly. It's the ability to determine what your opponent's going to do next turn or next action through conscious or unconscious pattern recognition. This skill is applicable to RPS, weighted RPS, and "weighted RPS where the numbers are hidden", though to different degrees and in different ways.

    While it's not possible to know with absolute certainty whether your opponent has a big combo in hand or not, knowing with a high degree of certainty is very possible. I don't dispute, however, that Yomi the game has unclear weighted RPS, as that's pretty much the point of the game.

    Yomi has very little to do with numbers (at least in the sense that you mean it here). It is informed by the relative damage and combat strengths of both characters, but is fundamentally not a probability-based ability. Instead, it is the ability to know intuitively with high confidence what the opponent is going to play. Computation, statistics, and "RNG in your head" are valuation skills and are pretty close to the opposite of yomi.

    On a personal note, I've managed to achieve fairly strong results in both tournaments and quick match using no concept of valuation, RNG, or "pretending to be a computer" whatsoever (outside of basics like calculating the odds of my opponent having a throw in their hand or counting to see whether they have any dodges left).
    Why do you believe that it's impossible or "impracticable" to determine where your opponent is in the infinite chain?

    Can't figure out how to quote quotes, so I'll just say that it's kind of interesting that ntillerman would say that, as I have completely the opposite opinion. I find that yomi is only useful against players who have strong knowledge of the game, as otherwise it's not really very possible to figure what your opponent is trying to do (as they don't really what they're doing either). For this reason I've always tended to do much worse against new players/players with relatively worse results than I have against players with better results.
    My opinion is that there is still a significant amount of yomi in a single round of Yomi, though it's much more important in longer sets or against players you have a lot of experience against. Similar character matchups against the same person also make yomi more reliable.

    May respond to other things later if I feel like it.
    Last edited: Mar 24, 2014
    Leartes likes this.
  19. deluks917

    deluks917 Well-Known Member

    Given the prevalence of "tied" combats in yomi (block vs block, dodge vs dodgem, etc) its seems unlikely anyone straight wins 50% nevermind 60% of combats.

    I am actually not sure if anyone has a .6 or better ratio of won combats to lost combats. I think this is actually plausible for at least some character/player combinations. For example it seems possible to me that someone could win 60% of their combats as Rook/Midori or especially Troq. The reason here is not that they has such good yomi. Its that vs a grappler you afford to lose ALOT of attack vs block combats. Especially late game the plan A is to spam fast attacks (contingent on reasonable assumptions on how many AA they can get, number of doges/special blocks left etc).

    I agree with the opinion no one's yomi is really that strong. But I just wanted to point it out combat win% does not seem like a super simple thing to consider, especially not with grapplers involved.

    (one semi-reasonable first approximation to a character's strength in yomi is how low a percentage of combats you can win while still winning the match. This works ok except for Perse I think).


    I do not think most people on this site have played competitive pokemon (its the gameboy game spalyed on a sim with a number of added rules, no grinding, thousands of people playing online any time of day). But the game also involves double blind decisions. Imo the game really taught me alot more aboutal game "yomi" than the actual game yomi. In that community people refer to yomi as "prediction." As one might expect from FS.com discussions there is a split among people who think prediction is basically guessing and people who think it is a skill.
  20. Bucky

    Bucky Well-Known Member

    One thing I had to learn the hard way about Yomi was that playing to maximize my combat win% isn't very effective at winning the game, compared to building a hand and making combos.

Share This Page