|
On September 30 2015 16:34 Celial wrote:Sufficiency, can you whip something similar up with, like, 400k games instead of 4k...? I mean I'm pretty dumb, but even I know that 4k games are not really a representative amount of games. Idk if its feasible in terms of time investment, or if you even have the time. http://www.goldper10.com/stat/2420-what-role-should-i-main-to-climb.htmlEdit: EUW would be most appreciated cuz its my server, but you are from Canada so I guess you have better data for NA. Shouldn't be that much difference in soloqueue meta tho I guess.
data is conceptually bad There is no indication of how gold leads are acquired or how hard it is to acquire a gold lead in a particular role.
For instance if i look at that data i might say that the easiest way to win the game is to give my ADC who is not me a gold lead at any point in time. Which is to say i play support and win the lane for them.
Otherwise the answer is simple, because every role when it has a gold advantage tends to win the game. So play any role and you will climb... so long as you get a gold advantage in your games... which is well, kind of obvious and not answering the question that people want.
To be fair though the question is dumb. Its asked by people who don't understand what they're good at or why their teams win. If people knew what they were good at they would instinctively know what roles and champions they should be focusing on because then the real question "how do i get a gold lead at any point in the game" is obvious, you get a gold lead by playing roles which utilize your advantages and negate your weaknesses.
On September 30 2015 18:22 Prog wrote:
4k games are probably way more representative than you think.
Not broken down like they have it. 5 Roles all tied up in each others stats. If you don't randomly sample and so pick about 1/5th of the data for each role you've got independence problems. You additionally have independence problems for each of the next 10 minutes issues.
Edit:
Re AD items:
You can't really get around there being a "high multiplication" in league. Tank stats are multiplicative. AP is multiplicative. You either have your consistent damage dealer have the highest multiplicative stats or you break the early/late game progression of team fight power in ways that don't make sense with regards to the game. So that is pretty much right out.
Crit is bad because randomness is bad. Because the law of large numbers* does not ever apply to league. And this means that any amount of crit besides 0% or 100% means that many fights are entirely determined by luck not just because of the amount of crits but also because of the order of crits.
And in a game like league that is super bad.
Fuck crit. Fuck crit forever.
*LLN's basically say that the probability limit of an average converging to the average is 100%. So for league this means that over a high number of attacks you will crit about the same amount of time as your crit %. But in league we don't care about averages we care about sums. And sums don't converge and that means if you get ahead by 200 HP early, you're likely to stay ahead by 200HP until the end of time. LLN says 200/n attacks = 0 as n->inf. But we don't care about that we care about the absolute amount of HP you're ahead when a champion dies. And in that case, no convergence. Just pure unadulterated random fucking walks. Fuck crit forever.
|
But Crit isn't even a pure dice roll. It has been explicitly said since the Renekton patch that it is unlikely to score multiple crits in a row.
EDIT:
going back to my previous point, critical strike isn't a pure dice roll. It's unlikely to get multiple crits in a row - but how much is the effect no one seems to know.
Therefore, all criticisms of crits seem to be based on anecdotal evidence when the above effect failed to prevent multiple crits in a row and turned the tide of a fight. While I believe such events will happen, it is probably far more unlikely than what some people claims.
|
data is conceptually bad There is no indication of how gold leads are acquired or how hard it is to acquire a gold lead in a particular role.
That's the biggest problem with the article. gold difference doesn't tell you anything. It's just a number, and it's general consensus that when you are winning, you are ahead in gold. So basically the article says 'When you are winning (aka gold lead), these are your chances of winning'. It does not tell you which role is best for carrying in yoloQ.
A much better but not that much more meaningful analysis would be to see the win rate based on which and how many lanes are winning. Do you really think the chance of winning is high when bot is winning but mid and top are losing? That would tell us more about the carry potential of every role, not this rubbish.
But we don't care about that we care about the absolute amount of HP you're ahead when a champion dies. .
Also sample size (#of AA) is so low that convergence would be a miracle. Although the crit system is not strictly RNG, so I can imagine that it does start converging much faster than it naturally would (but if we're talking about 1-5 AA here, then convergence is nigh impossible with most crit chances).
Therefore, all criticisms of crits seem to be based on anecdotal evidence when the above effect failed to prevent multiple crits in a row and turned the tide of a fight. While I believe such events will happen, it is probably far more unlikely than what some people claims.
I wish Riot would release the way they calculate crit chance, there are a bunch of mathematicians who could give input over it. But the point still stands, crit is (relatively) random and not predictable, and it can completely change a small skirmish.
|
Nobody remembers crit is psuedorandom. They'll bring up Wildturtle over and over and pretend RNG confirms their biases.
I just think it creates bad itemization problems/is a boring stat.
|
pseudorandom mitigates the problem, but does not solve it.
|
Well it doesn't really matter how random or pseudorandom it is as long as a single crit can dictate the outcome of a fight.
|
On October 01 2015 04:22 Sufficiency wrote:
But Crit isn't even a pure dice roll. It has been explicitly said since the Renekton patch that it is unlikely to score multiple crits in a row.
EDIT:
going back to my previous point, critical strike isn't a pure dice roll. It's unlikely to get multiple crits in a row - but how much is the effect no one seems to know.
Therefore, all criticisms of crits seem to be based on anecdotal evidence when the above effect failed to prevent multiple crits in a row and turned the tide of a fight. While I believe such events will happen, it is probably far more unlikely than what some people claims.
Doesn't actually matter unless crit is both deterministic over a specific time frame and that determinism is necessitates proper numbers of crits for all participants over the specific time frame which is also the same time frame of every fight in the game. Which is to say that it doesn't actually matter because there is no way for the algorithm to determine whether or not someone is going to die before they get their requisite crits off and then give them those crits or to determine that someone will die and has too many crits and so goes back and takes crits away or to smooth out the fact that someone might die faster due to a crit streak than they should have given the proper number of crits and so give them more HP.
We can see an easy example of how advantage is problematic even with perfect determination.
At 50% crit suppose for a second that the system flips between crit/no crit every attack. But that the first crit is actually random, at least in the sense that its sufficiently difficult to determine whether or not you're going to start your chain on a crit. Over any 2 attacks we always crit 50% of the time. This is as "perfect as it gets"
Alright, then 50% of the time one side has a ~1.5 auto attack advantage which flips to neutral and back at each next auto attack for the entire fight. 25% of the time blue side has the advantage, 25% of the time purple side has the advantage, 25% of the time they both start on crits, 25% of the time neither starts on crits.
Think about that for a second. If the system is most consistent and least random that it can possibly be. The very best we can do is that one side has a 1.5 auto attack advantage 50% of the time.
For smaller numbers of crits the auto attack advantage length extends but the likelihood of that advantage is lower.
But that doesn't matter because lots of pokes and team fights happen. And so the probability that this happens at least once in a game, or once in a tournament, goes really really fast to numbers we don't like.
edit: And its important to note that a 1.5 auto attack advantage can be super huge. The tank falls 1.5 auto attacks faster which means now the enemy team is routed, still attacking our tank while we get to hit their squishies. And/or the AD is killed 1.5 auto attacks sooner and so our AD now gets to clean up for free. So on and so forth.
You cannot ever fix crit except by removing the randomness entirely.
|
You're seeing this in a vacuum way too much.
Some tanks are stronger tanks than other tanks
Some adc's have higher attack speed, higher range, higher base values
Item tresholds matter too.
Only if you had a simulation where both comps are exactly the same, exactly the same items, lvls and start attacking at exactly the same and both teams would mirror eachother perfectly, then you can make a case.
And you saying 1.5 auto attacks faster, are you kidding me? That's like 0.2 seconds you can 'maybe' start auto attacking the squishy. If they don't have another tank, if you're positioned to do so, if you have more hp, if you're stronger, etc etc etc.
|
Yea some tanks are stronger than other tanks, this is an advantage they get from choosing a stronger tank, or farming better, or outplaying the enemy. That is OK
Some AD's are stronger than other AD's and some assassins are stronger than other assassins and this is an advantage they get from pure random chance that has nothing to do with an outplay, a better farming scheme, or choosing higher damage. That is bad.
Fuck crit forever
|
Some AD's are stronger than other AD's and some assassins are stronger than other assassins and this is an advantage they get from pure random chance that has nothing to do with an outplay, a better farming scheme, or choosing higher damage. That is bad.
OK that's going a bit far. Vayne isn't a good duelist bc she crits more, she excels because of her whole kit (ult boosts crit power though, but bonus AD isn't random, she just gets more out of crits). And there are tons of scenarios where random crits won't be the decisive factor in a fight (all the stuff you mentions are usually the case), it's just when it does happen it can be quite frustrating, especially in a situation you are technically on even footing.
|
Which is to say that it doesn't actually matter because there is no way for the algorithm to determine whether or not someone is going to die before they get their requisite crits off and then give them those crits or to determine that someone will die and has too many crits and so goes back and takes crits away or to smooth out the fact that someone might die faster due to a crit streak than they should have given the proper number of crits and so give them more HP.
It might help to use point form instead of one single sentence.
Crits are fine. You are just arguing over some theoretical implications which may or may not happen. The only facts we know are:
1. Crit is not a dice roll; you are unlikely to crit twice in a row. Therefore, even in an ideal situation, it's actually a lot harder for an ADC to win in a duel against a squishy simply by a lucky streak of crits.
2. Randomness is always in the game. While there are no "hard" random components beside critical strike, the players have incomplete information over the game and other players behaviours. Therefore, for all intents and purposes, there will always be randomness. By the way, I am not going to go into a philosophical argument as to why the world is actually entirely deterministic, blah blah blah.
|
Plim(1-(1-P)^n, n->inf) =1
If it "may or may not happen" in any particular instance its more or less guaranteed to happen over time. And its not like we haven't seen crit randomness decide fights! And that was just the ones that were super obvious, which is not typically how team fight advantages go [due to how advantages stack and accrue in league]
WRT the quoted sentence: The quoted sentence means that you cannot design an algorithm that works perfectly under every situation, there must be situations where the crit algorithm fails to produce a non-advantageous result. Indeed, its more likely than not that there are situations which fail to produce a non-advantageous result. I even showed how a "perfectly non-random" solution produces significant advantage 50% of the goddamn time
We are all OK with "player created randomness". Like players randomly shoot out skillshots, or randomly dodge in certain directions or choosing the place you're going to be at any given point randomly; we are fine with that. That is a result of a strategic choice of the participants.
Crit is not. Fuck crit forever
|
On October 01 2015 05:44 Goumindong wrote:
Plim(1-(1-P)^n, n->inf) =1
If it "may or may not happen" in any particular instance its more or less guaranteed to happen over time. And its not like we haven't seen crit randomness decide fights! And that was just the ones that were super obvious, which is not typically how team fight advantages go [due to how advantages stack and accrue in league]
WRT the quoted sentence: The quoted sentence means that you cannot design an algorithm that works perfectly under every situation, there must be situations where the crit algorithm fails to produce a non-advantageous result. Indeed, its more likely than not that there are situations which fail to produce a non-advantageous result. I even showed how a "perfectly non-random" solution produces significant advantage 50% of the goddamn time
We are all OK with "player created randomness". Like players randomly shoot out skillshots, or randomly dodge in certain directions or choosing the place you're going to be at any given point randomly; we are fine with that. That is a result of a strategic choice of the participants.
Crit is not. Fuck crit forever
The problem is that player-created randomness is still randomness. For example, it's always possible that I will get ganked by a hell-bent jungler who wants to destroy me at level 2. It's no more interesting than something that is generated by a machine. Do you want to also remove fog of war?
If you hate randomness so much you should probably go play chess. Oh wait, even in chess tournaments there are some randomness due to bracket selection and white vs black.
|
In chess its both impossible to eliminate first move advantage/disadvantage and they play enough games that its unlikely to make much of a difference. If games weren't independent and you carried through advantages then first move would be a real problem. But they are and they don't. Of course even then first move advantage/disadvantage is one of the most discussed/debated topics in chess for that very reason you outlined .
Player created, I.E. strategic randomization is indeed more interesting than machine generated rng.
|
On October 01 2015 05:52 Sufficiency wrote:Show nested quote +On October 01 2015 05:44 Goumindong wrote:
Plim(1-(1-P)^n, n->inf) =1
If it "may or may not happen" in any particular instance its more or less guaranteed to happen over time. And its not like we haven't seen crit randomness decide fights! And that was just the ones that were super obvious, which is not typically how team fight advantages go [due to how advantages stack and accrue in league]
WRT the quoted sentence: The quoted sentence means that you cannot design an algorithm that works perfectly under every situation, there must be situations where the crit algorithm fails to produce a non-advantageous result. Indeed, its more likely than not that there are situations which fail to produce a non-advantageous result. I even showed how a "perfectly non-random" solution produces significant advantage 50% of the goddamn time
We are all OK with "player created randomness". Like players randomly shoot out skillshots, or randomly dodge in certain directions or choosing the place you're going to be at any given point randomly; we are fine with that. That is a result of a strategic choice of the participants.
Crit is not. Fuck crit forever The problem is that player-created randomness is still randomness. For example, it's always possible that I will get ganked by a hell-bent jungler who wants to destroy me at level 2. It's no more interesting than something that is generated by a machine. Do you want to also remove fog of war? If you hate randomness so much you should probably go play chess. Oh wait, even in chess tournaments there are some randomness due to bracket selection and white vs black.
I'm not sure what you're arguing against at the moment. Goumindong clearly has no problem with randomness so long as it's the direct result of a player choice (e.g. the player chose to dodge in a "random" direction, or ganked at a "random" time etc.). It's randomness that doesn't involve player agency he dislikes (e.g. crits). Fog of war and belligerent junglers are beside the point.
|
On October 01 2015 06:03 Seuss wrote:Show nested quote +On October 01 2015 05:52 Sufficiency wrote:On October 01 2015 05:44 Goumindong wrote:
Plim(1-(1-P)^n, n->inf) =1
If it "may or may not happen" in any particular instance its more or less guaranteed to happen over time. And its not like we haven't seen crit randomness decide fights! And that was just the ones that were super obvious, which is not typically how team fight advantages go [due to how advantages stack and accrue in league]
WRT the quoted sentence: The quoted sentence means that you cannot design an algorithm that works perfectly under every situation, there must be situations where the crit algorithm fails to produce a non-advantageous result. Indeed, its more likely than not that there are situations which fail to produce a non-advantageous result. I even showed how a "perfectly non-random" solution produces significant advantage 50% of the goddamn time
We are all OK with "player created randomness". Like players randomly shoot out skillshots, or randomly dodge in certain directions or choosing the place you're going to be at any given point randomly; we are fine with that. That is a result of a strategic choice of the participants.
Crit is not. Fuck crit forever The problem is that player-created randomness is still randomness. For example, it's always possible that I will get ganked by a hell-bent jungler who wants to destroy me at level 2. It's no more interesting than something that is generated by a machine. Do you want to also remove fog of war? If you hate randomness so much you should probably go play chess. Oh wait, even in chess tournaments there are some randomness due to bracket selection and white vs black. I'm not sure what you're arguing against at the moment. Goumindong clearly has no problem with randomness so long as it's the direct result of a player choice (e.g. the player chose to dodge in a "random" direction, or ganked at a "random" time etc.). It's randomness that doesn't involve player agency he dislikes (e.g. crits). Fog of war and belligerent junglers are beside the point.
Crits give player agency too. You can always estimate how much damage you can do on average, since your chance to crit is known. It's no different from estimating your chance of getting ganked.
|
For some actual game discussion, I played two Yasuo games this afternoon and it feels so much better to play him now that I have 25-28 ping. In fact, it's awesome. I can block most skillshots with W and following up to even short knockups like Vi's Q and Draven E is a breeze. I think I might actually start playing him more seriously from now on.
Also I had a very interesting second game where the gold chart was like this:
+ Show Spoiler +
But my team (blue side) won while being 18 kills behind at the game's finish. Really strange game.
|
On October 01 2015 06:08 Sufficiency wrote:Show nested quote +On October 01 2015 06:03 Seuss wrote:On October 01 2015 05:52 Sufficiency wrote:On October 01 2015 05:44 Goumindong wrote:
Plim(1-(1-P)^n, n->inf) =1
If it "may or may not happen" in any particular instance its more or less guaranteed to happen over time. And its not like we haven't seen crit randomness decide fights! And that was just the ones that were super obvious, which is not typically how team fight advantages go [due to how advantages stack and accrue in league]
WRT the quoted sentence: The quoted sentence means that you cannot design an algorithm that works perfectly under every situation, there must be situations where the crit algorithm fails to produce a non-advantageous result. Indeed, its more likely than not that there are situations which fail to produce a non-advantageous result. I even showed how a "perfectly non-random" solution produces significant advantage 50% of the goddamn time
We are all OK with "player created randomness". Like players randomly shoot out skillshots, or randomly dodge in certain directions or choosing the place you're going to be at any given point randomly; we are fine with that. That is a result of a strategic choice of the participants.
Crit is not. Fuck crit forever The problem is that player-created randomness is still randomness. For example, it's always possible that I will get ganked by a hell-bent jungler who wants to destroy me at level 2. It's no more interesting than something that is generated by a machine. Do you want to also remove fog of war? If you hate randomness so much you should probably go play chess. Oh wait, even in chess tournaments there are some randomness due to bracket selection and white vs black. I'm not sure what you're arguing against at the moment. Goumindong clearly has no problem with randomness so long as it's the direct result of a player choice (e.g. the player chose to dodge in a "random" direction, or ganked at a "random" time etc.). It's randomness that doesn't involve player agency he dislikes (e.g. crits). Fog of war and belligerent junglers are beside the point. Crits give player agency too. You can always estimate how much damage you can do on average, since your chance to crit is known. It's no different from estimating your chance of getting ganked.
The point is the player can mitigate getting ganked from the fog of war, but the player cannot mitigate the sequence of crit-noncrit in a series of 5 autoattacks by the enemy AD carry.
Mitigate isn't really the right word, but you know what I mean.
|
On October 01 2015 06:08 Sufficiency wrote:Show nested quote +On October 01 2015 06:03 Seuss wrote:On October 01 2015 05:52 Sufficiency wrote:On October 01 2015 05:44 Goumindong wrote:
Plim(1-(1-P)^n, n->inf) =1
If it "may or may not happen" in any particular instance its more or less guaranteed to happen over time. And its not like we haven't seen crit randomness decide fights! And that was just the ones that were super obvious, which is not typically how team fight advantages go [due to how advantages stack and accrue in league]
WRT the quoted sentence: The quoted sentence means that you cannot design an algorithm that works perfectly under every situation, there must be situations where the crit algorithm fails to produce a non-advantageous result. Indeed, its more likely than not that there are situations which fail to produce a non-advantageous result. I even showed how a "perfectly non-random" solution produces significant advantage 50% of the goddamn time
We are all OK with "player created randomness". Like players randomly shoot out skillshots, or randomly dodge in certain directions or choosing the place you're going to be at any given point randomly; we are fine with that. That is a result of a strategic choice of the participants.
Crit is not. Fuck crit forever The problem is that player-created randomness is still randomness. For example, it's always possible that I will get ganked by a hell-bent jungler who wants to destroy me at level 2. It's no more interesting than something that is generated by a machine. Do you want to also remove fog of war? If you hate randomness so much you should probably go play chess. Oh wait, even in chess tournaments there are some randomness due to bracket selection and white vs black. I'm not sure what you're arguing against at the moment. Goumindong clearly has no problem with randomness so long as it's the direct result of a player choice (e.g. the player chose to dodge in a "random" direction, or ganked at a "random" time etc.). It's randomness that doesn't involve player agency he dislikes (e.g. crits). Fog of war and belligerent junglers are beside the point. Crits give player agency too. You can always estimate how much damage you can do on average, since your chance to crit is known. It's no different from estimating your chance of getting ganked.
Your chance of getting ganked is a function of the strategic choice of the enemy jungler. Your chance to crit is not a strategic choice of anyone.
|
On October 01 2015 06:23 Fildun wrote:Show nested quote +On October 01 2015 06:08 Sufficiency wrote:On October 01 2015 06:03 Seuss wrote:On October 01 2015 05:52 Sufficiency wrote:On October 01 2015 05:44 Goumindong wrote:
Plim(1-(1-P)^n, n->inf) =1
If it "may or may not happen" in any particular instance its more or less guaranteed to happen over time. And its not like we haven't seen crit randomness decide fights! And that was just the ones that were super obvious, which is not typically how team fight advantages go [due to how advantages stack and accrue in league]
WRT the quoted sentence: The quoted sentence means that you cannot design an algorithm that works perfectly under every situation, there must be situations where the crit algorithm fails to produce a non-advantageous result. Indeed, its more likely than not that there are situations which fail to produce a non-advantageous result. I even showed how a "perfectly non-random" solution produces significant advantage 50% of the goddamn time
We are all OK with "player created randomness". Like players randomly shoot out skillshots, or randomly dodge in certain directions or choosing the place you're going to be at any given point randomly; we are fine with that. That is a result of a strategic choice of the participants.
Crit is not. Fuck crit forever The problem is that player-created randomness is still randomness. For example, it's always possible that I will get ganked by a hell-bent jungler who wants to destroy me at level 2. It's no more interesting than something that is generated by a machine. Do you want to also remove fog of war? If you hate randomness so much you should probably go play chess. Oh wait, even in chess tournaments there are some randomness due to bracket selection and white vs black. I'm not sure what you're arguing against at the moment. Goumindong clearly has no problem with randomness so long as it's the direct result of a player choice (e.g. the player chose to dodge in a "random" direction, or ganked at a "random" time etc.). It's randomness that doesn't involve player agency he dislikes (e.g. crits). Fog of war and belligerent junglers are beside the point. Crits give player agency too. You can always estimate how much damage you can do on average, since your chance to crit is known. It's no different from estimating your chance of getting ganked. The point is the player can mitigate getting ganked from the fog of war, but the player cannot mitigate the sequence of crit-noncrit in a series of 5 autoattacks by the enemy AD carry. Mitigate isn't really the right word, but you know what I mean.
Sure you can.
For example, if you are playing mid lane with half health and you meet an ADC in the jungle with IE, you go "OH SHIT HE HAS IE" and run away. You just mitigated damage because you understand that building IE gives him more damage.
It only becomes a problem with your estimation and the actual damage done disagree largely with each other. But this probably does not happen as often as you think since critical strike is not a simple dice roll - it's unlikely to crit multiple times in a row to severely skew your estimate.
On October 01 2015 06:26 Goumindong wrote:Show nested quote +On October 01 2015 06:08 Sufficiency wrote:On October 01 2015 06:03 Seuss wrote:On October 01 2015 05:52 Sufficiency wrote:On October 01 2015 05:44 Goumindong wrote:
Plim(1-(1-P)^n, n->inf) =1
If it "may or may not happen" in any particular instance its more or less guaranteed to happen over time. And its not like we haven't seen crit randomness decide fights! And that was just the ones that were super obvious, which is not typically how team fight advantages go [due to how advantages stack and accrue in league]
WRT the quoted sentence: The quoted sentence means that you cannot design an algorithm that works perfectly under every situation, there must be situations where the crit algorithm fails to produce a non-advantageous result. Indeed, its more likely than not that there are situations which fail to produce a non-advantageous result. I even showed how a "perfectly non-random" solution produces significant advantage 50% of the goddamn time
We are all OK with "player created randomness". Like players randomly shoot out skillshots, or randomly dodge in certain directions or choosing the place you're going to be at any given point randomly; we are fine with that. That is a result of a strategic choice of the participants.
Crit is not. Fuck crit forever The problem is that player-created randomness is still randomness. For example, it's always possible that I will get ganked by a hell-bent jungler who wants to destroy me at level 2. It's no more interesting than something that is generated by a machine. Do you want to also remove fog of war? If you hate randomness so much you should probably go play chess. Oh wait, even in chess tournaments there are some randomness due to bracket selection and white vs black. I'm not sure what you're arguing against at the moment. Goumindong clearly has no problem with randomness so long as it's the direct result of a player choice (e.g. the player chose to dodge in a "random" direction, or ganked at a "random" time etc.). It's randomness that doesn't involve player agency he dislikes (e.g. crits). Fog of war and belligerent junglers are beside the point. Crits give player agency too. You can always estimate how much damage you can do on average, since your chance to crit is known. It's no different from estimating your chance of getting ganked. Your chance of getting ganked is a function of the strategic choice of the enemy jungler. Your chance to crit is not a strategic choice of anyone.
Yet you DO buy IE/Shiv/PD on ADC. Your strategic choice is clearly unsound.
EDIT: my point being, critical strike has tons of player agency - probably just as much as predicting a gank or every other aspect of the game. There are very few edge cases which it fails due to multiple crits in a row.
As such, I believe any argument on why crit is bad needs to first study Riot's algorithm on critical strike and see exactly how likely it is to crit (say) 3 times in a row.
|
|
|
|
|
|
EPT![[image loading]](http://i.imgur.com/W6pu8gd.png)
