How does HLTV rating work? Rating 2.0 explained
Вставка
- Опубліковано 25 вер 2022
- News, matches, stats, and more on HLTV.org:
hltv.org/
Rating 2.0 is used to measure CS:GO player performance. But how does it work? In this video, we try to explain the main idea behind the HLTV rating.
Host:
/ kingdempz
#csgo #esports - Ігри
i knew pretty much everything already except kills againts eco have same value as kills against full buy. would have liked to know more about impact rating.
I would have thought eco bashing would have reduced the impact rating (or at least not increased it) but apparently not!
useless exit kills are also counted as same impact, just loot at baitF stats
@@gandalfgass facts
@@gandalfgass well it does have direct impact on the following round since victims of exit frags have to rebuy
@@vasikanpoikanen yes, but they only chase him if they can afford it - hence he's not having an impact.
He never has impact because he's never chasing impact.
Difference is huge, he plays on their terms - you can never have a winning game playing like that, no matter the game.
HLTV: Here is rating 2.0
Everyone: How are you calculating it?
HLTV: Here is a video that lists everything we take into account
Everyone: What's the formula?
HLTV site: 0.405022* KRP - 0.657678 * DPR + 0.524246 * KAST/100 + IMPACT/5 + 0.00410341 * ADR + 0.343334
HLTV YT: Magic
Source for the value?
@@nilslorand hltv site
So you didn't tell us how to actually works.... and you identified several problems, so when is the rating 3.0 coming?
The one thing I want to see from 3.0 would be a system where killing an armourless nitr0 isn't worth as much as a kill against S1mple on a full buy and beating Faze would be worth more than beating BIG.
So changes to ratings and rankings would weigh a lot more on what you've done against who and you'd end up with like an algorithmic Web constantly altering and changing itself.
Would be a very fun project for someone smarter than me to take on.
@@jonajon91 I think you’re making a good point except for the “killing s1mple is worth more than killing nitro”. In my opinion it should be based on the players performance in the match itself than overall rating. Imagine if nitro had a 30 kill game but s1mple had a 3 kill game. Those who killed nitro would get less “points” than those who killed s1mple when he wasn’t performing.
Another thing is how much your equipment costs vs the cost of whoever you killed. That makes eco frags and awp kills less impactful
@@nicofied. BlameF in shambles.
Rating 3.0 = Leetify rating 👌👍
@@jonajon91 S1mple stats would be in shambles
Yeah, so, how does it actually work?
could have at least given the formula pdf
Well, they can't tell you how exactly it works because it is a business secret. They wouldn't want others to replicate it.
@@user-fe2oh8oj2u it’s just an arbitrary number. Other people can make their own system and it will do the same thing. So there’s no point in keeping the HLTV formula a secret.
@@user-fe2oh8oj2u but others are already capable of replicating it lol
@@user-fe2oh8oj2u A 12 year old maths geek could reverse engineer it and find it out without much effort, I just don't think anyone cares enough to bother
Impartially judging it, what do you think the strengths and weaknesses are between HLTV 2.0 and Leetify?
Supportive players and IGL's get a very raw deal with HLTV 2.0 but things such as flash assists, and good grenade usage in general do matter to successful execution and winning games, yet they're ignored by HLTV.
Leetify is much more advanced. HLTV rating is just for bigger audience, i doubt pro teams look at them that much.
Players with good communication/calls will, for obvious reasons, not get any credit in the stats.
A player with good ideas and good communication, can make every player around him play better, thus improving their stats and also the team's chances of winning the round/game. The impact of that player's comms/calls/ideas, will not be visible in his stateline/rating.
Leetify rating is bad if you are in losing team unless it's a close match
@@fica1137 so leetify rating is bad if you are bad? Yeah, that seems to sound like how it should be.
Imma give some personal experience from me here. (Faceit lvl 8, light team experience. Im usually that guy that gives a solid 1.1 ish KD, plays a lot of "bitch spots" and provides a lot of clutches/support/map control to the team.
For me, Leetify feels more accurate to my playstyle. I often get games where i feel like im playing pretty well but my KD is pretty low, for stuff like getting my teammate to trade, "kill steal" and generally not being part of most of the action. While HLTV rating totally saws me for that, leetify gives a fair, mby slightly high rating for that, which i believe is good, after all, CS is a team game.
HLTVs rating really mostly shines in showing off star players in tournaments and build hype imo.
nice video..pls show us the formula..so that community can start working on hltv 3.0 formuoa
I did see a reddit post once where someone figured it out, kind of reverse engineered it.
I'm not going to find it but it is out there.
title "how it works"
video "yeah its shit too"
Mr HLTV tech man can you add a 'consistency' stat on your website for how many on their last 100 matches fall within 15% either side
of the players average? Or something similar, it's a pain typing all the data into my excel sheet manually every time I'm curious.
We actually have an article titled "A look at inconsistency" in the back end at the moment, read it on the website soon! And the stat idea is interesting, too.
@@HLTVorg Excellent, I had someone call Valde inconsistant a few days ago and tbh I didn't really know what to say about it, is he? I've never heard that about him? So I put together the excel sheet to find out and had to type all the data in manually. I'm sure you could do a more advanced formula than what I came up with, but I look forward to the article.
i second this consistency stat suggestion
@@jonajon91 what did you find, in terms of Valde?
@@FlurryMainchin the guy that called him inconsistent was an idiot.
Are there any other statistical measures available? On HLTV you can check the percentiles as well, when you hover over someones rating 2.0. But it's confusing, because it's not distributed equally around 1.00. A rating of 0.89 puts you in the bottom 14 percentile. A rating of 1.00 put you in the 45 percentile and a rating of 1.12 equals the 82 percentile. I'd love to see the distribution chart. 1.00 doesn't equal the median nor the average, it's a bit higher. And I don't know if the scores are normally distributed and what the SD would be.
So much more could be done with these numbers.
Can you share the weightages of the individual factors contributing to the final HLTV 2.0 rating.
Didn’t ask
If you treat kills during saving and anti-eco kills as normal, any rating system sucks in my eyes.
Can you tell what kind of data is easily extractable from the matches/demos for the purpose of making a rating? Like if you'd want to for example divide multikills into different groups based on which side you are playing and how much time passes in between the kills, would that be feasible and convenient?
I mean if you get two kills in a quick succession on the T side (entry or denying a retake) it would on average probably be more impressive and impactful than on the CT side (site defense). Or if you get a kill and survive to find another kill later on, that would probably be on average more valuable on the CT(get a pick and survive, then get another on site defense or retake) than T side (1 kill on entry and another while denying a retake or just random picks).
I think the way the demos work should allow for very detailed measures of many different things if you just think it through how to consistently capture certain commonly occurring situations from the data.
Leetify rating still the GOAT
The rating should take into acount the equipment value difference beetween the players that kills/dies so that if a player onedeags an AWPer the deagle kill should be worth a lot more than vice versa. This will punish ecofraggers and punish people who die without using their grenades etc. And most importantly, nerf awp kills a bit.
Also in addition, a kill on T side would likely be worth more with their equipment being cheaper. Which is more difficult than getting a kill on CT by far
We need this to be a separate rating, maybe called econ rating, which has a big effect on rating 3.0 but doesn't completely make it up. Perhaps this "econ rating" would contribute to about 20 or 30% of the final rating 3.0
Leetify rating does this well
Great idea. Adding few more
1) T side Entry frags which get map control or unlock a site.
2) 1 v 1 duels. Like you said deagle vs awp but also the rating diff of those players.
3) CT side site holds.
4) Work rate of the player. Like how much map they cover and control. For e.g rotations , information play etc.
The fact that you said 11/0 is equal to infinity annoyed me more than it should have.
Thanks. Are you working on 3.0?
If you die in a 1v1 but they fail to close the round (Not enough time to defuse etc) does it still count as a won clutch, and does it positively impact rating?
you literally won the clutch
@@VERYR4RE I was just asking if the stats reflect that. Technically, a program could see it as a lost 1v0, not a won 1v1, hence asking.
@Syrge2k @RARE pt 2 I was just asking if the stats reflect that. Technically, a program could see it as a lost 1v0 for the other side, not a won 1v1, hence asking.
ADR shows true consistency of a player.
Does IGL calling also included in 2.0 rating?
Is it okay to use second steam account to play in hltv tournaments?
How about weighting the rounds differently to account for more? For example by how much money was spent. When the enemy had glocks a 3k really isn't special, dying to glocks is pretty bad. A 3k with glocks vs full buy is incredible, dying with a glock isn't bad, it's average. Maybe make it if the team with full buy wins, as expected, the round has very little impact, but if the eco team wins the round gets weighted more than usual.
A lot of people talking about it not differentiating against guns or ecos. In a single match, sure, it matters but in the long run those kills won't weigh as much to the point it would make a difference in someone's yearly rating. On the other hand, yes, double punching the survival (via DPR and KAST) will lead to inflating the rating but anyone who understands the indicators knows to look at more than just the Rating 2.0. You can instantly spot if a rating is inflated by an absurdly low DPR. Jame had a 1.17 rating in 2021 but had average KPR, average Impact, poor ADR and a stupidly low DPR.
Good video bro
You guys should add a players 69% (confidence interval (1 standard deviation) along with the average rating across a period.
Reckon Lettify rating is more acurate?
complexity-vs-astralis match👀👀
Rating 3.0 maybe take into acoount how much was invested so farming against eco isn't as valuable
basically awpers are only inflating stats by a bit then because their adr is the same as someone that has died (unless exit frags) but their impact is less as they arent opening the rounds with it.
So the awper topic of inflating stats isnt as bad as people make it seem
Um. Plenty of awpers get opening picks. S1mple and zywoo are perfect examples of awpers who will take the first gun fight in a round…
Awpers stats are inflated by the fact that the formula takes data on rounds survived then again takes KAST which includes rounds survived so that one stat factors into the rating more than any other. It's why Sh1ro and Jame's rating always seem a little ... inflated.
@@thchiefe thats why their ratings are so good, cause they actualyl have immense impact whereas an awper like sh1ro or broky, this is why i thgink its not inflated, if you compare the kd and stuiff, sh1ro and broky even though plus double digits will have a lower rating than a s1mple zywoo double digit kd
for example sh1ro and broky with +10 kd would be 1.40-1.59 range whereas s1mple/zywo would have 1.50-1.75 range
reason being is their openign kills, adr and impact is alot higher thats why i dont think awpers inflate it
you cna judge yhourself but isee awpers like s1hro getytying way lower rating than the average awper or rifler when the kd are same
@@sh1nenkai258 ah my bad, I thought you were saying that they were inflating their stats by not getting opening kills. I misinterpreted what you were saying
@@thchiefe nw
also rating 2.0 doesnt really apply to IGL role players that much
Add player rating per kill on a cross-gradient taking into account average rating of player killed, ranking of opposition, and level of opposition buy. I could probably sort a formula out after 12 menial hours and I’ll probably do it tonight just cos I’m so fucking OCD.
how dare you
hooxi is god, the eternal, the undisputed
he's infinitely better than papatronic
So you guys don't count economy different in impact rating?
It will be simplest thing for lowering eco frag abuse
Wonder how trades are regustered
6:05 this is key.
1:13 Dividing by 0 doesnt result in infinity though ;)
I need to know how is the "formula" to understand sh1ro and Jame ratings, theres no way they sometimes have 1.40+ rating just by saving 12 times the awp in tsites (saving gives u rating?)
survival give better HLTV stat, that's why
a save is good for the rating 2.0, for some reason even if the player make the team lose many rounds
@@arthurchung3747 yeah and i dont think thats good
11 \ 0 = Infinite😂😂😂💀
Someone find me the man who did the maths on this bc I cant believe that
Well,
The player :
We all know that 11/0 = 0. But 0 is not the
result that we want. 11 is also not the good
answer because 11 is equals to 11/1. Which
is not the performance. The performance is 11/0.
So, we have to use a number that is like 0 but not
0 to solve the problem.
The NPC wonder why he's listening this guy.
The player :
What is the number who is (almost) like 0 but not a 0 ?
The NPC is thinking...
The player :
0.1
The NPC :
Ok then what ?
The player :
And what is more close to 0 than 0.1 ?
The NPC try to think again... (But it's difficult)
The player :
0.01
The NPC :
Oh, I know ! We can try 0.000000001 then right ?
It's like 0 but not 0 ! That's the solution !
The player :
Exactly, go ahead
The NPC is typing on his calculator
11/0.000000001 and react :
WHAT ! I GOT 11.000.000.000 as a result !
The player wait
The NPC :
WAIT ! I'LL TRY WITH 0.0000000000000000
000000000000000000000001
The NPC is typing on
his calculator again :
WHAAAAAAAAAAT ! I GOT a higher number
11.000.000.000.000.000.000.000.000.000.000 !
-
WAIT ! I'll try with 0.000000000000000000
0000000000000000000000000000000000
0000000000000000000000000000000000
0000000000000000000000000000000000
0000000000000000001
The player grabs the calculator.
The player :
Don't do that. You see NPC. When you add 0s after
the decimal point, the results grows. So, the more
the value is close to 0, the more the result will
be higher. We call this +infinite.
The NPC :
Ooooooh ! I understand now !
The player :
Welcome to maths, you can be a player too if you want.
The NPC :
No thanks, I just want to rush B and tell
to my team that we are losing because of
them. Even if I'm last on the leaderboard.
- Alternative ending -
The NPC :
Thanks for this lesson.
The player
Don't worry, come short with me.
I smoke window, you smoke conector.
We play together ok ?
The NPC :
Ok, I'm with you.
- Generic -
- Hiding scene -
The director :
Hollywood would probably choose the
alternative ending. Hmmm.
The producer :
Hollywood would never have produce this sh*t.
It's boring. Go back to create magic like this video.
dempz my goat
Tbh, this could be a 3 min video
I comment to help statistics
if you are an AWP it goes your way :D Crap of ranking. It is much harder to bi rifller
HLTV doesn't know about cs go, if a team wins a tournament they are automatically the best in the world. It should come as no surprise after zyWho's top 2 steal from NiKo...
So essentially he made faces the whole video, didn't give a proper explanation of how the formula works, and at the end of the year we will see s1mple being no.1 again despite 10 other players being better than him. Because how the rating works is whatever he said in the video, and if your name is s1mple or zywoo we just add a little extra to it, despite you being worse than your rating. Smh, I swear they don't tell us the proper way of how they calculate so that we don't see how they rig the ratings. I remember 1 game back in like 2018 I think where Electronic had like 30 kills and s1mple only like 20, electronic did multi kills, most clutches in that game and everything, but s1mple still got the higher rating, and then at the end of the year of course this adds up.
Meanwhile papatronic..
Leetify rating better
>>>HLTV: Let's weigh down anti-eco and exit kills in the rating calculations
>>>BaitF's rating plummets below 0.9
m4a1s is overpowered and csgo devs don't care about balance therefore it's a waste of time to play csgo.
heych el tea v
I think a system that evaluate players like how IQ have been doing over 100 years is a good idea for rating 3.0. Gonna be sick
Do one of these videos on how your ranking system works so we can link it to the brain dead people that think it's a power ranking.
'My favorite team went down in points, but they didn't even play waaahhh'
Instead of average use median
it will be a lot more accurate
They also said the average is 1.00 when people figured out almost instantly that the average hovers around 1.06.
you cant even calculate teams let alone players
they should just use elo
2:20 you gave gigachad low skill? You are low skilled, HooXi solo you all
HLTV ranking sucks still, Movistar Riders above Vitality currently, like clearly it is broken and you need to update
There's nothing broken dude. The ranking is totally fine as it is. It will reupdate when there will be the finals at the end of this week
Sorry but I can't hear this accent
Lets hear them explain this bullshit
11/0 =/= infinite
karrigan 0.93 hahahah fattigan is so garbage agahahahah trash player, t4 player overrated igl
Useless video explains nothing we don't know.
Video is actually useless
Leetify ranking may have it's flaws but it is so much better than this 😅