Life In 19x19

I know we've discussed David Sirlin as a person and an author here before. And whatever you may think of the theories he presents in 'Playing To Win', I have to say... The man can make some very compelling games.

Not to go into salesmanship or gushing or anything, but I just discovered Yomi, a card game simulating a traditional Street Fighter style fighting game. It's very deep, very compelling, and relies almost entirely on reading your opponent's possible mindset. There's luck involved, I've lost games because I had some bad beats where I didn't draw a good variety of cards, but even so, being able to effectively out think your opponent is most of the game.

Anyway, it struck me as the sort of deep strategy game which would appeal to people here, especially the poker/video gamer crowd. I'm in love, because it's finally a fighting game where my aging and already poor reflexes don't weigh me down. In short, it's a fighting game where you remove reaction timing, and your judgement of your opponent is all that matters.

Anyway, for those who want to know more, you can find physical yomi, and a better description of the game here., and play it free (kind of. 2 characters a week, rotating like League of Legends.) here.

Anyway. I'm Daemondym over on Fantasy Strike. If anyone else is into it, drop me a wave.

i went through the tutorial and boy am i bad at that game. i went from 53 points of health with the cpu having 9 to 1 point of health and the cpu having -7 (it took me 12 turns to finish off the cpu.)

I played some, it's fun, although it has a lot of luck, a bit more than I'd prefer I think.

The CPU plays completely at random, FYI. The Puzzle Strike bot has brains, but the Yomibot just chooses cards at random and does everything it possibly can even if it's completely retarded. You can do stuff like pick Lum and constantly attack and basically not lose ever.

Zombie wrote:The CPU plays completely at random, FYI. The Puzzle Strike bot has brains, but the Yomibot just chooses cards at random and does everything it possibly can even if it's completely retarded. You can do stuff like pick Lum and constantly attack and basically not lose ever.

I'm pretty sure this isn't true. I can pretty reliably bait the yomibot into certain behaviours. It has a (poor) AI, that tries to value its cards based on your character's probabilities, and your previous few turns. Granted, it's easier to bait the bot than even the most new of new players.

The reason you may be able to play attacks and never lose is that this is terrible play. The yomibot is programmed to assume you'll not do suicidal and crazy things, the way most opponents would assume, when reading, that you wouldn't do crazy things.

Like, for instance, if I attack 3 times, and my opponent blocks 3 times, MOST of the time, my opponent will not block the fourth time. They may dodge, trying to catch me in trying to attack for the fourth time, because I think they'll pull a fast attack and I've been saving my fastest attack. They may play that fast attack. Or they may throw because I may dodge because of that fast attack.. but generally, after a string of a single action, you can say with increasing probability that the next card thrown will not be that sort of action. After 3 or 4, the yomibot probably has no idea what to do with you, zombie. AI's tend to break when given inputs that make no sense, after all.

emeraldemon wrote:I played some, it's fun, although it has a lot of luck, a bit more than I'd prefer I think.

I think it has less luck than say, poker.

While luck may limit your options some, there's a broad range of strategies for each given character, based on cards. Much of the skill is figuring out the thought process in your opponent's head. Playing the person is a big part of Yomi, more than say, go, which makes it feel more unreliable, but it really is a skill.

Granted, individual games can be won or lost on the luck of a topdeck draw, getting just the right card to make the hand perfectly function. But by and large, the mechanics of the game are secondary to being able to get into someone's head, and pull their strings. In this way, it's largely like a computerized fighting game.

When playing Mortal Kombat 2, which I did every night for about 3 years, while working in a gas station, I could set the playerchoice on random. It didn't matter which character I was playing, or who my opponent was playing. The game was entirely me knowing my friend's reactions to certain behaviours well enough to work out what they'd do. Knowing every character inside and out took them entirely out of the picture. Of course, Mortal Kombat 2 was a pretty poor excuse for a fighting game. The characters were wildly unbalanced, many of them were pallet swaps with similar or identical move sets, and the sweep timing was broken, not to mention that the input system was crap, no queue or anything to punish button mashing..

But the point remains. Yomi is about head games, as much as it's about valuation of a hand, or even what you have in your hand.

CSamurai wrote:

Zombie wrote:The CPU plays completely at random, FYI. The Puzzle Strike bot has brains, but the Yomibot just chooses cards at random and does everything it possibly can even if it's completely retarded. You can do stuff like pick Lum and constantly attack and basically not lose ever.

I'm pretty sure this isn't true.

I thnk I'd rather believe Thelo (aka the guy who wrote the website and digital version of the game), who has multiple times stated that Yomibot is brainless. We're creatures that seek patterns and reasons in everything, oftentimes attributing patterns to things that are, in reality, random. It is worth noting that Yomibot does have a pattern because of the deck distributions which mostly skew towards attack.

CSamurai wrote:

emeraldemon wrote:I played some, it's fun, although it has a lot of luck, a bit more than I'd prefer I think.

I think it has less luck than say, poker.

While luck may limit your options some, there's a broad range of strategies for each given character, based on cards. Much of the skill is figuring out the thought process in your opponent's head. Playing the person is a big part of Yomi, more than say, go, which makes it feel more unreliable, but it really is a skill.

Granted, individual games can be won or lost on the luck of a topdeck draw, getting just the right card to make the hand perfectly function. But by and large, the mechanics of the game are secondary to being able to get into someone's head, and pull their strings. In this way, it's largely like a computerized fighting game.

When playing Mortal Kombat 2, which I did every night for about 3 years, while working in a gas station, I could set the playerchoice on random. It didn't matter which character I was playing, or who my opponent was playing. The game was entirely me knowing my friend's reactions to certain behaviours well enough to work out what they'd do. Knowing every character inside and out took them entirely out of the picture. Of course, Mortal Kombat 2 was a pretty poor excuse for a fighting game. The characters were wildly unbalanced, many of them were pallet swaps with similar or identical move sets, and the sweep timing was broken, not to mention that the input system was crap, no queue or anything to punish button mashing..

But the point remains. Yomi is about head games, as much as it's about valuation of a hand, or even what you have in your hand.

This is pretty much accurate. There is no absolute correct move most of the time in Yomi - there are simply moves that work and those which don't. Also, nevermind poker (which has a ton of luck, you need tons and tons and tons of repeat plays to ascertain who is actually good or not), this has less luck than stuff like Magic. No manascrews or anything. The one bad thing about the game is that it fries my brain much faster than the other games I play.

For the beginners: Learn valuation first and start worshipping card advantage in the early game. Seriously. To read properly, you first have to have a decent grasp on what is even sane to do, because no one can be right all the time (and sometimes winning combat is actually bad idea). When you have a solid base of valuation, you can use that to beat people who don't know what they are doing (something pure reading dies against), and then get the advantage by good reading. (the opponent, that is)

I am skeptical about your (or anyone's) ability to read my mind over the course of a game. If we played a large number of Yomi matches (say 10 or so), maybe. But I doubt a single game (~30 card choices, from the games I've played) gives you enough information about me.

The website lists my rank as "Student 2 (Grasshopper)", which has no meaning to me, but I'd guess I've probably played about 10 games total. I do think there's some interesting strategy involved, and it's possible I'll learn more as I play. But let's compare to go: I'm sure we could find people on this forum who could beat me 10 times out of 10 in go. Is there enough skill in Yomi that someone could beat me 10 times out of 10? 9/10? Reading would become more important for a longer set like that, But I'd still give myself decent odds to take 1 game out of 10 off of anyone.

This isn't to say the game is bad, if I weren't having fun I wouldn't have made it to grasshopper

I played a few vs the computer, it's kinda interesting.

The block/attack/throw dynamic is basically rock/paper/scissors. Choosing between the three randomly* is the best you can do, unless you can take advantage of some predictable (i.e., non-random) play from your opponent.

[*] Humans are really bad random number generators. Specifically, we generate runs of the same thing much less often than we should.

P.S. 2000 posts!

Random is unlikely to be the best because the payoff for each option is not the same (e.g. for many characters block is pretty great because a card advantage is quite significant and pure block at the start would likely beat random choice). In addition, the probability for each option is not equal (more problematic when hand size is small) and there is a hierarchy within the same option (due to speeds for throw and attacks). There are also added complexities (e.g. special abilities on some cards).

Interesting game, though I blame the OP for providing me with another excuse to procrastinate :p

Well, the distribution perhaps ought to be weighted to one or the other. If you make the grid mapping choices to outcomes (the grid they use when explaining the prisoner's dilemma, I forget what it's called), there will be multiple reasonable choices, and you should randomly choose each with a certain probability to maximize your expected value.

Yes, though the weighting would be very difficult and perhaps unfeasible to find for human players (especially because the opponent's hand size, what is in their discard pile, and their abilities would all affect the weighting). In addition, the opponent likely also has some valuation approach (at least in part), which would affect their matrix. I'm nowhere near proficient in game theory, but I suspect that it's neither feasible nor optimal (unless you restrict opponent's strategy).

The more I think about it, the more I think it could be a fun AI project...

illluck wrote:Yes, though the weighting would be very difficult and perhaps unfeasible to find for human players (especially because the opponent's hand size, what is in their discard pile, and their abilities would all affect the weighting). In addition, the opponent likely also has some valuation approach (at least in part), which would affect their matrix. I'm nowhere near proficient in game theory, but I suspect that it's neither feasible nor optimal (unless you restrict opponent's strategy).

Just to restate what I was saying with more official words, there's probably no "pure strategy Nash equilibrium" for most plays (http://en.wikipedia.org/wiki/Strategy_(game_theory)#Pure_and_mixed_strategies). It's nearly always infeasible for a human to do the math during a game, of course.

emeraldemon wrote:I am skeptical about your (or anyone's) ability to read my mind over the course of a game. If we played a large number of Yomi matches (say 10 or so), maybe. But I doubt a single game (~30 card choices, from the games I've played) gives you enough information about me.

The now-retired best player had a winrate in the 80's. He does consider Bo1 Yomi to be unplayable (about which many on the forums disagree, though all agree that a longer set is a better test of skill).

To the probability folks, good luck on quantifying stuff like safety and card advantage. It's RPS, yeah, but the payoffs are not only uneven, but also unclear. Plus you have to figure out what your opponent is likely to do and have (which is very possible).

Zombie wrote:

emeraldemon wrote:I am skeptical about your (or anyone's) ability to read my mind over the course of a game. If we played a large number of Yomi matches (say 10 or so), maybe. But I doubt a single game (~30 card choices, from the games I've played) gives you enough information about me.

The now-retired best player had a winrate in the 80's. He does consider Bo1 Yomi to be unplayable (about which many on the forums disagree, though all agree that a longer set is a better test of skill).

To the probability folks, good luck on quantifying stuff like safety and card advantage. It's RPS, yeah, but the payoffs are not only uneven, but also unclear. Plus you have to figure out what your opponent is likely to do and have (which is very possible).

Even in Bo1, there are certain indicators. Even ignoring things like card distribution, and probabilities, let's talk about competitive rock paper scissors.
No, really, that crap exists. It lasts 3 throws. So, if you're going into rock paper scissors competitions, you've got way less time to read your opponent. The best RPS players play their opponents though. You can make certain guesses about consecutive throws, based on the previous throws. A lot of it is headgames, with a dash of game theory and the unlikelyhood of three consecutive anything leads to all sorts of situations where that's the right move in competitive play.

By comparison, reading your opponent's patterns is way easier in Yomi, even in Bo1. Does your opponent attack a lot? Dodge? Block? How do they use their throws? What do they open with? What do they play after you play a couple of the same move? As in fighting games, Signalling, that is, intentionally playing a certain pattern to draw out certain reactions, is a powerful weapon in Yomi. If you attack 4 times running, you're signalling that you're on a rampage. Your opponent has to figure out whether you're going to change up into something else, or stick with an attack for the 5th move. Knowing the character they're playing will help you know what their options are, but only getting in their head will really answer that question.

Some players go so far as to track previous moves in spread sheet form, so that they can consult it to see what your patterns and tells are. I find this somewhat cheap, but with my memory, I could do the same in my head, so what do I care? It's like a joseki dictionary. It'll only tell you what is written down, not what's in someone's head.

Do the players chat during the match? Choice of words, exclamations of frustration, all of these are important factors into getting into their head, and guessing what they're up to.

Best of 1 is hard, but by no means the hardest mind reading trick around.