League of Legends XXXII: Cupcakes and Kittens and Squirrels. Oh my!

[Dada]

I really wish your response did more than just contradict mine, without providing any points of why you actually think I am wrong.

[Zen Master]

I really wish your response did more than just contradict mine, without providing any points of why you actually think I am wrong.

... where did I contradict you and say you're wrong?

[Dada]

Oh, the data is definitely there,

Right there

[Zen Master]

Right there

But it's in the match history. And if that doesn't satisfy, then lets say - for the sake of argument - that what happens is this:

Your picks, bans, runes, masteries etc. are all recorded locally, translated to a simple numerical code, and uploaded to the central server every so often, like every 10, 100 or whatever number games.

It's not hard to do, and it's not an insurmountable task, neither locally nor for the server. The server then spits out updated list daily or whatever, telling the easily exitable public what's what.

It's seriously not something that cannot be done. It's just a lot of effort for something that would have little to no true value.

[toasty]

It's just a lot of effort for something that would have little to no true value.

I think the argument has been that, given the amount of work required to properly find and organize the data, it might as well be impossible. The combinations possible in LoL, after accounting for Runes, Masteries, Picks and Bans, ignoring player skill, are... effectively infinite. Trying to properly deal with that level of data would require supercomputers and guys with PhD's running around creating equations and running numbers. Its effectively impossible.

[Dada]

But it's in the match history. And if that doesn't satisfy, then lets say - for the sake of argument - that what happens is this:

Your picks, bans, runes, masteries etc. are all recorded locally, translated to a simple numerical code, and uploaded to the central server every so often, like every 10, 100 or whatever number games.

It's not hard to do, and it's not an insurmountable task, neither locally nor for the server. The server then spits out updated list daily or whatever, telling the easily exitable public what's what.

It's seriously not something that cannot be done. It's just a lot of effort for something that would have little to no true value.

Gah. What I'm saying is not that we can't collect the data - as you describe above, that is fairly easy. Neither am I attacking the math and computation part. As Toasty mentions, that would require an enormous effort, but thats not my point.

My point is that our data (picks&bans) depends on a total of 16 different variables. Even discounting runes, masteries and player skill, there exist over 10^31 different combinations of picks and bans, and we need statistically significant amounts of data for each of these. If we were able to collect information from all the games ever played on LoL, that would not be a nearly large enough dataset to generate any statistically significant results.

[Talesin]

I think he's more saying that the actual data isn't there, not the ability to store it, well actually to store that quantity of data you'd need a mainframe that probably amounts to a large city even up to a planet.

Math:

Spoiler

Show

If you had 100 million players all playing at once you'd have 10 million games going on. If every game is a 20 minute surrender you get 30 million games done per hour. Let's assume they all play 5 hours a day you've got 150 million games a day. If they played for a whole year they'd get 54,750,000,000 games done.

Given the total number of possible games is 10,000,000,000,000,000,000,000,000,000,000 the 54.75 billion games played by those 100million players is a drop in the ocean given the number of games overall. It would take 1.83*10^20 years to do all the games at the rate listed above.

I think i've got that all correct but feel free to correct me

Additionally all the games in LoL weren't played with 95 champions available or with 3 bans per team. This means you'd need to play the total number of games between each patch or they data wouldn't be precise because an additional factor is present in the new games (the new champion) that wasn't present in the old games.

[Zen Master]

I think he's more saying that the actual data isn't there, not the ability to store it, well actually to store that quantity of data you'd need a mainframe that probably amounts to a large city even up to a planet.

Math:

Spoiler

Show

If you had 100 million players all playing at once you'd have 10 million games going on. If every game is a 20 minute surrender you get 30 million games done per hour. Let's assume they all play 5 hours a day you've got 150 million games a day. If they played for a whole year they'd get 54,750,000,000 games done.

Given the total number of possible games is 10,000,000,000,000,000,000,000,000,000,000 the 54.75 billion games played by those 100million players is a drop in the ocean given the number of games overall. It would take 1.83*10^20 years to do all the games at the rate listed above.

I think i've got that all correct but feel free to correct me

Additionally all the games in LoL weren't played with 95 champions available or with 3 bans per team. This means you'd need to play the total number of games between each patch or they data wouldn't be precise because an additional factor is present in the new games (the new champion) that wasn't present in the old games.

Well ...... you're aware that statistics for individual champions exist, right? It does not, however, take into account the exact build. But it's easy enough to find what champion has the greatest win propability.

So ... while not 100% accurate in every aspect (far from it, in fact), it's done. Not do-able - done. People are doing it, now.

Wanting to account for every single possible variation is something you guys came up with, not me.

Somewhere between the extremes of the fairly sloppy statistics available now, and the total attention to every conceivable detail you guys speak of, there is a compromise that is easily detailed enough, and easy to do in practical terms. If one were so inclined.

Very similar things exist for WoW.

[Dada]

Well, I interpreted the original discussion to be on the "optimal" model, which was what my arguments were based around. Also, thank you Talesin for illustrating my point.

Anyway, yes, of course you can make a simpler model (e.g. champion win-rate), but then the usefulness of it drops as well. I mean, would you ever look up the win ratios of champions before deciding your pick?

Champion win rates show which champions people are good at winning with, which is different from which champions are actually strongest. There is some overlap, of course, but it is really not a usefull metric for making decisions concerning a single game.

[Talesin]

Well ...... you're aware that statistics for individual champions exist, right? It does not, however, take into account the exact build. But it's easy enough to find what champion has the greatest win propability.

Yeah but does that probability take in account the other champions picked or are they a "If Trynda is picked and Shen is picked, Trynda's team wins 55% of the time and Shen's does 45%"?

If so, then a database with that level of simplicity could be available but I would question its usefulness on the fact that all its telling you is that independent of any other variables Trynda beats Shen. This is something I would assume powers lolcounter or anyone saying that "X counters Y"

So ... while not 100% accurate in every aspect (far from it, in fact), it's done. Not do-able - done. People are doing it, now.

Wanting to account for every single possible variation is something you guys came up with, not me.

I believe people came up with that so we could answer your question of would it provide a winning team. You'd need to look at a huge number of factors to get a team that is statistically more likely to win the game rather than just beat a team containing shen.

Somewhere between the extremes of the fairly sloppy statistics available now, and the total attention to every conceivable detail you guys speak of, there is a compromise that is easily detailed enough, and easy to do in practical terms. If one were so inclined.

Very similar things exist for WoW.

I would say I disagree on the practicality sheerly because I think you're underestimating the amount of information available but I don't know a vast amount about the data available so I could be wrong.

However as a thought experiment, given infinite time and power, yes a database could be made that gave you a team with a much better chance of winning based on their first ban. It could even be possible to create one that gave you a guaranteed win but I would be hugely surprised if either scenario was remotely feasible given current computing power.

[Math_Mage]

Well ...... you're aware that statistics for individual champions exist, right? It does not, however, take into account the exact build. But it's easy enough to find what champion has the greatest win propability.

That can already be done. Our statistics aren't as COMPLETE as what's available to Riot, but they're fairly extensive. Sites like lolking, lolbase, lolstatistics, riotstats, and so on have been been doing that forever.

But it's so utterly useless for picking team comps that everyone sort of assumed you meant the more meaningful model, which is picking based on win % as adjusted to the picks and bans so far (e.g. "Accounting for the fact that Shen, Kassadin, Ryze, Ahri, Shaco, and Lee Sin were banned, and the other team first-picked Janna, what champion pick at this point has the best chance of leading to victory?"). Which, of course, requires you to have a good guess about what each champion's win % is when picked after each combination of picks and bans.

And this sort of naive optimize-at-each-step model is still very simplistic, because it fails to account for picks that look suboptimal at time of choice, but are part of a team comp that, as a whole, outperforms the stepwise optimized comp. (The good thing is, at least we probably won't need to acquire even MORE data to feed into the algorithm--we just need better algorithms.)

The unfeasibility comes in when one considers that the number of possible matches is so astronomically high that we don't even have ONE sample of every possible match, or even MOST possible matches. This is the data needed to rigorously support a meaningful model.

That said, one can take shortcuts. For example, one could force the model to handle only AD-Support bot lanes and to include a jungler, which drastically cuts down on the number of combinations. You lose the ability to make a comprehensive assessment that includes unconventional setups...but then, it's not like unconventional setups are all over the place in the current system.

[Zen Master]

This is getting to be very pointless.

No offense to anyone, but I never claimed the database would be a good idea - quite the opposite, in fact. I've claimed it's possible, and clearly it is, because similar - if less specific - things exist already.

Furthermore, I never intended to discuss statistics, which are surprisingly dreary and dull, even when one expects them to be so.

So I bow and resign from the discussion. I shall utter not a word further on statistics. Or on databases. For the most part, we don't even disagree on anything except whether or not it's something that can be done or not - which was never my point to begin with.

[TheAmishPirate]

Could've sworn I posted in here a few months back...oh well!

Hello Lol Thread! My friends got me hooked on original Dota in high school, and I hopped over to LoL in college and have loved it ever since. I mostly play AP carries, but I try to keep a good roster of champions for other roles too.

Username is TheAmishPirate, and I'm on the NA servers. Hope to see some of ya'll in game!

[Neoseanster]

In regards to the statistics discussion: uh, it's been going back and forth a bit, but I'm not sure you guys are really getting anywhere fast. Yes, it might be theoretically possible to make a pretty good model of what's good against what given what circumstances.

No, this isn't really feasible on a meaningful level because it requires you to gather and analyze a truly ridiculous amount of data. There are hundreds of variables that would need consideration to make it truly accurate: I'd bet we're talking not just "takes a few afternoons to design and lags your computer to try and analyze" ridiculous, but "would requires years to design and more processing power than humanity as a whole has available to it right now" ridiculous.

I mean, think about it. There are ninety five-ish champions in the game right now, and ten slots in a game. If I'm not totally borking up the math (95*94*93*92*91*90*89*88*87*86), that means there, uh, 36,668,784,925,297,324,800 different combinations of champions on each team possible in the game right now. And that's before you account for bans, player skill and preferences, different builds, rune pages, masteries, etc etc etc.

Of course, you'll be able to prune it down a fair bit if you cut out "unviable" options, and there are other ways to cut down on complexity at the cost of a greater margin of error, but it remains a prohibitively difficult task, and even if successful you'd probably only find things like "a team with champions X, Y, and Z will have a 4.5% greater chance of victory if they're against a team with champions A, B, and C, but not L."

Do we really need to go back and forth on the matter much further? I think that the majority of the posts for the past two or three pages have been about this.

Sigh, lost over 70 elo over last 2 days. I should really stop playing in the mornings - while I'm ecstatic over getting to play with guys like aAa Linak and Gosu Pepper, I'm reeeaaallly not on their level.

I got to support imaqtpie a few weeks ago, that was fun. <3 We wrecked our lane, but sadly our team was too heavy.

[ex cathedra]

I mean, think about it. There are ninety five-ish champions in the game right now, and ten slots in a game. If I'm not totally borking up the math (95*94*93*92*91*90*89*88*87*86), that means there, uh, 36,668,784,925,297,324,800 different combinations of champions on each team possible in the game right now. And that's before you account for bans, player skill and preferences, different builds, rune pages, masteries, etc etc etc.

Yup. Taking bans into account, there are 11,546,801,012,126,245,723,024,527,360,000 different permutations. 11.5*10^30. That's the amount of unique game compositions possible in league of legends, but it doesn't take roles in account. One comp could be Kennen, Alistar, Galio, Gangplank, Kog'Maw. Where are they going? What are those lanes? No one knows!

[Qaera]

11.5*10^30.

Are you kidding, qt? 1.15e31~

~ ♅

[Baxter190]

Yup. Taking bans into account, there are 11,546,801,012,126,245,723,024,527,360,000 different permutations. 11.5*10^30. That's the amount of unique game compositions possible in league of legends, but it doesn't take roles in account. One comp could be Kennen, Alistar, Galio, Gangplank, Kog'Maw. Where are they going? What are those lanes? No one knows!

Kennen top, Galio mid, Pirate jungle, and Kog'Maw alistar bot.

[toasty]

Kennen top, Galio mid, Pirate jungle, and Kog'Maw alistar bot.

Kennen bottom with GP, Alistair Jungle, and Galio Top. Kog'maw middle ofc.

[Lix Lorn]

*sob* Finally, a forum with a LoL-thread.. I can die in peace now.

...Lol Forums? :P

Yup. Taking bans into account, there are 11,546,801,012,126,245,723,024,527,360,000 different permutations. 11.5*10^30. That's the amount of unique game compositions possible in league of legends, but it doesn't take roles in account. One comp could be Kennen, Alistar, Galio, Gangplank, Kog'Maw. Where are they going? What are those lanes? No one knows!

Ooh, I'd send-

Kennen top, Galio mid, Pirate jungle, and Kog'Maw alistar bot.

...aww.
I'd have done that though, although I had Galio and Kennen switched.

[ex cathedra]

Are you kidding, qt? 1.15e31~

~ ♅

Well excuuuuuuse me, Princess.

I don't remember how to do basic algebraic maths, okay.

[fred dref]

So it turns out that when you click on mid, you should pay attention as to whether what was once a safe destination will still be one by the time you get there. I was watching a standoff at the enemy Blue when I found myself accidentally walking into the entire enemy team and dying, which let them catch two other guys out of position, kill them, take Baron and then the game. So yeah, not watching where I was going kinda just lost my team the game. Sorry guys!

[ex cathedra]

Kennen bottom with GP, Alistair Jungle, and Galio Top. Kog'maw middle ofc.

Nah bro, it's all about the support Galio comps.

[PEACH]

Obviously it's Kog jungling, Kennen supporting Alistar bot, Galio solo top, and GP mid...

Ugh. The problem with making a ****ty comp involving GP is that GP works everywhere.

[Darth Mario]

Obviously it's Kog jungling, Kennen supporting Alistar bot, Galio solo top, and GP mid...

Ugh. The problem with making a ****ty comp involving GP is that GP works everywhere.

Simple. Either roam GP or GP just goes and plays Dominion instead.

[Tesla_pasta]

New galio skin... uh sweetness?
also Teemo OP

[Duos]

Ok, me and some other people need to have a chat. The involved parties know who they are.

Clearly there is something going on and I want to get it resolved in a peaceful and coherent manner before somebody (most likely me, because I have a smart mouth) says or does something stupid.

Either do it through PM or just PM me a time when it's feasible to talk on mumble, because I want to keep drama out of the thread.

[MachineWraith]

Hello, GitP LoL'ers. After much wheedling from my friends, I'm gonna start playing League of Legends. I've never played it or, really, any similar games before. The closest I've come, I think, is playing a lot of Last Stand mode in Dawn of War II. Based solely on aesthetics, I want to play as Fizz.

Advice? Whether build advice, starting player advice, whatever, it's all appreciated.

[Drascin]

New galio skin... uh sweetness?

Really? Wooooo! I love Galio, let's see what...

...oh god that looks uglier than sin . What does a man have to do for a Goliath Galio?

[Math_Mage]

Hello, GitP LoL'ers. After much wheedling from my friends, I'm gonna start playing League of Legends. I've never played it or, really, any similar games before. The closest I've come, I think, is playing a lot of Last Stand mode in Dawn of War II. Based solely on aesthetics, I want to play as Fizz.

Advice? Whether build advice, starting player advice, whatever, it's all appreciated.

Welcome to our playground of legends!

For build advice, I would point you to Solomid.net, which has a vast array of well-constructed guides by reputable players. For Fizz in particular, I would probably advise a heavy ability power build in mid lane, such as Boots of Speed + 3 HP pots --> 2 Doran's Rings --> Rabadon's Deathcap --> Lich Bane, but as I don't play Fizz, that's just a starting template that should probably be heavily modified as you become more familiar with the champion. There are also attack damage builds; AD Fizz usually jungles. (Jungling is wandering across the map killing neutral creeps rather than enemy minions in the early-mid game; you can also try to kill enemy laners by ganking them, or steal the neutral creeps from the enemy jungle.)

For play advice, for now I would recommend you simply practice getting last hits for gold. Picking up lots of last hits is the most reliable way to get gold, which gets you items, which makes you strong, which allows you to win fights later on. Practice a few different kinds of champions--Fizz, for example, is typically an AP burst champion, whereas Ashe is a ranged AD carry, Janna is (usually) a support, Warwick is a bruiser, and so on. This helps you understand what people are going to do with these roles, and how you should coordinate in your role to help your teammates fulfill their roles.

And get on Mumble, we don't bite. Most of the time.

[lord_khaine]

Really? Wooooo! I love Galio, let's see what...

...oh god that looks uglier than sin . What does a man have to do for a Goliath Galio?

Yeah, so far it looks really ugly.