@@wardog9959nah I know plenty of people who solo queued to high elo, myself included. You’re just bad, unfortunately. One person can absolutely carry games consistently with the right playstyle
@@imwawodid you watch the video? Do you have multiple leaning disabilities? I’m genuinely asking… he proved that it’s possible for that happen with the elo system that was in place for the longest time. It’s like you literally just want to argue
A big issue is that the top fragger might not be a good player. He might just be sacking team m8's for kills. Maybe he doesn't do that and instead just doesn't give call outs, so he has all the infromation from the other players, but no one else has any info from him, which means that he has a better chance to top frag.
Yeah it is very obvious that it would be better to take performance into account *If* there was an objective measure of performance. The problem is that there are not really any good measures. For example the rating based on fantasy esports seems absolutely terrible
You don’t need to be a fragger to be good at this game. Look at any pro team and they have players that do not frag out unless on rare occasions, they take on roles that sacrifice their stats for the win. Aleksib is a good example of a player that changes the outcome of fights just by being there with his utility and sacrifices often to set up his players. As long as your team is capable of trading each other well you can always perform. While there are better players at lower elos than can change the outcome of the entire match; I can guarantee if you look at your past games you’ve lost or lost several rounds due to your team or yourself not playing for the team. Swinging things you shouldn’t be swinging to get a kill and dying or randomly pushing somewhere without a team mate nearby to trade. While sometimes sure the person trying to trade will die too. But most of the time the positioning and crosshair placement of the individual who just killed your team mate is off and they’re open to being killed especially if you’re playing a crossfire. This game is about trading, multifrags happen but you shouldn’t be hunting for them. Your job is to take site or defend site and buy time for rotates. If you lose site within seconds even if you get one or two it’s pretty much over if the enemy has util to defend site. People need to think more tactically with this game, it’s not cod.
I feel the biggest inflator of kds is how many 1vxs they get into. If you get 3 kills but they where in a 1v5 those 3 kills matter very little compared to the two kills your teammates might have gotten the round prior and won because a save.
For how much shit talk chess and the system get in this video, they are perfect for each other - chess is a fair 1v1 game, in which the 3 way result, and the relative strength of the players are the only things that should matter when evaluating their ratings.
@@sauceinmyface9302It's not that simple... There are openings, every opening has many lines, and every unique variation can have a different average win % for each side. White does move first, but chess played perfectly is a draw, since for every perfect attack there is the perfect defense, and at most levels, the color doesn't make much of a difference...
@@1GMitzy chess played perfectly is not a draw that’s a misconception based on our current technology’s ability to play chess being drawish. We have no idea what chess played perfectly would look like or its outcome because the chess engines of today are primitive in comparison to the futuristic and gargantuan computational power needed to solve chess, and only then would we know what perfect play would look like. Currently stock fish and alpha zero are just our best guesses with the technology we have
@@1GMitzy You clearly don't play chess at a high level then at all. White absolutely has an inherent advantage by going first. This is why tournament games give each player an equal chance to play on both white and black.
There's simply no way to truly measure individual performance in this game. Individual performance based system will encourage baiting and not dropping weapons to your teammates. Let me explain on seemingly ridiculous example. If there would be a CS2 team that wins all the majors/tournaments in which one player is not even sitting on a chair but gives magical feet massage to his teammates which makes them insanely good so that they win games, he would be one of the top 5 CS2 players in the world. It does not matter what he's doing, all that matters is that he's doing something that is winning games. There simply cannot be other accurate indicator of who is a better player as winning games, it doesn't matter how it's done. Rather than individual performance, what I think would be a good idea is to make elo gain/loss adjustment based on rounds won/lost. Like there would be a significant difference between 0-13 game and 11-13 game. It reduces variance but doesn't destroy the objective of the game which is winning rounds/games together with a team
While I think that round based system could be a good idea it might just make it even scarier for solo queueing. 5 stacks it’ll obviously work bc the other team is better but for solo queueing you would be scared shitless to even queue for the fear of getting bad teammates and losing 4 times the elo you gained in a legitimately enjoyable hard though match with nice teammates that you won 13:11. It’ll make the negatives so much more potent and the positives weaker.
I'm not saying it should be only round based. Most or at least half of elo gain/loss should still depend of just winning or losing a game. Let's say that 60% elo gain/loss depends only on winning or losing, and then 40% is bonus dependent on how close the match was
I agree with this. Basically if you tried to incorporate every little thing a player does that has impact on the game into their elo gain/loss you'd probably just boil it down to did the team win or lose? I like your suggestion of making it smaller or larger gain/loss depending on how the game went for sure.
@@KMurtheGamer But it's not about winning or losing, you can perform very well, carry your team, and still lose. Yes, there are games where someone is absolutely backpacking and they drag their team, kicking and screaming to a win, but the rest of the team does statistically poorly. You cannot look me in the face and tell me that a team that went 13-11, with stats lines looking like, 28-17, 19-19, 21-19, 5-21,16-14 and a team that went 13-0 with stat lines all in similar ranges should get the same elo upon their game finishing. That just logically doesn't make sense, and while it may seem that just looking at wins and loses provides a valuable measure of a team's skill(it doesn't, that's why win/loss %'s are flawed) but your rank isn't a measure of your team's skill, it's a measure of YOUR skill. Plus, you can absolutely track things like drops and team economy, and factor those things into the rating. Average everyone's cash, and if everyone has within +/- $500 of that value in equipment at the start of the round, then boom, they've dropped. Or, just penalize people with massive eco differences when they're actually in the game and playing. If you're just looking at wins and losses you're communicating that you care about nothing else, as long as you win you get those points. That can lead to situations where teams kick their under performing members. Before they were removed from competitive play, they acted as another life for a person on your team. So, kick your bottom fragger, and now you have an extra life for your top-frag, allowing them to just throw themselves in an entry. That was such a concern, even after they removed bots from comp, that they don't allow 4 stacks to kick the random in CS2. There are so many things in this game that SCREAM that it would be a better experience if person performance was evaluated. It would make solo-queue less of a fucking slog, for one, as I'm no longer concerned about how my teammates are performing on a macro scale. If they lose the round but I did well, I can take solace in the fact that I won't be penalized as much because the flawed matchmaking system put me in a game where I'm significantly better than my teammates. And conversely, if I get into a game an I'm not performing well and team loses as a result, but everyone else did well, I can take solace in the fact that my poor performance isn't penalizing them as much. Valuing just wins and losses creates a such a toxic team environment because everyone has to perform at their peak, there can be no dips, otherwise it *IS* your fault.
In my opinion. U need 4 other people who like to play to improve and want to win and create strats and not just random q. Hardest part is finding people who are like that.
I mean, that is what the matchmaking system basically is. It's just that the skill rating falls below ELO, which is the main problem with any ranked game you could find
Firstly, this video has exceptional quality. People have pointed this out and I thoroughly agree. Unfortunately, the actual content is a bit of an issue, as numerous people have pointed out. The first simulation has a bunch of problems and is clearly not an accurate representation of CS matchmaking. Since your entire argument is based on this simulation showing a negative result, your entire argument is flawed. However, more fundamentally, the video (even if everything you said were true!!) doesn't actually mean much regarding fixing elo hell. Players would be more accurately ranked, maybe, but they would still be hardstuck and would still blame that on everything that isn't themselves. In the current system, if you are struggling to rank up, it is because you aren't winning games, and thus it is a skill issue. People complain about elo hell anyway. In your system, people would struggle to rank up because of personal perfomance and would complain about elo hell anyway (perhaps the metrics and figures are wrong, perhaps the game doesn't weigh wins enough, perhaps they are still getting god awful teammates every game!). Elo hell is a myth based almost entirely around people perceiving their skill level to be higher than it actually is, and this system doesn't fix that underlying issue, so people would still feel bad when they are stuck at the same rating for six months.
bro your video making and programming skills just shows me that CS is played by very smart and capable People. i feel honored to be a part of this community, BLYAD!!!
In theory using wins and losses alone is the best system, because it does not create any perverse incentives where a player could boost their own ranking at the cost of making their team more likely to lose. The problem is that in a team game there is so much variance (random factors outside of a player's control) that it takes too long for a pure win/loss system to converge.
I think the modern way is to use machine learning to identify the features that mostly contribute to the outcome, and scale the lost/won points based on that.
its funny cuz we've had HLTV ratings for over a decade and leetify for years now, both of which are specifically designed to gauge individual performance (impact), and leetify actually compares your stats across multiple skill groups. and yet valve decided to ditch their original comp ranking system for a slightly better yet still terrible one.
It's very easy to make an unrealistic simulation, which I think is all that has been demonstrated here. There are a couple decisions which are highly suspect: 1) The skill distribution of players in the simulation is flat, actual player skill probably fits closer to a normal distribution, and most player rating systems have this assumption baked in, so you are likely warping the results by doing this. 2) The description of matchmaking does not make sense "one player is chosen at random, then random players with in 100 ranks are chosen to form a match", and "each player has exactly 1000 matches" are not compatible statements, you would expect some variation in # of games played if matches are forming randomly, if you are removing people from the pool as they hit 1k games then you are messing with skill distribution of players, this may explain why many of the ranking graphs have huge swings for the last 100 or so games. I would not be surprised if the system actually looked like it was performing better while the average player is only at ~700 games. 3)The matchmaking system as described is much less likely to form matches with top rated and bottom rated players since there are fewer player within 100 rating of them, this likely makes it harder to 'decompress' skill ratings compared to player skill. It also biases the 2nd team to lose since it is definitely giving them worse players (the 1st player should be better than the 2nd player, the 3rd player should be better than the 4th player, and so on) these are both likely small effects, but it's an additional layer of bias. It's very important to make sure a simulation actually works before trying to use it to draw inferences
jesus you guys are expecting paper level of simulation from a youtuber. everybody knows elo is a bad system to rank ability in modern gaming. these 3 points you made would not alter the point of the video
@@JPClow3 No, they're expecting if you want to bring simulations and statistics into it that you use very basic high school level statistics and the stated rules of the system. This is a shoddy simulations which doesn't understand the system (players are selected from within a certain bound of the current level is a fundamental rule of matchmaking) and a normal distribution is so common it is ASSUMED that a normal distribution describes most things as an approximation until demonstration shows that it doesn't. Most people are average. Average intelligence, average height, etc. The people who are extremely smart, tall, whatever, are vanishingly small. A bell curve normal distribution has the majority of people around the middle and more accurately describes the situation. If you want to approximate things with a flat distribution, go ahead. It's equally likely to find someone of perfectly average height and the tallest man alive, as well as the shortest. You're just as likely to be cut off in traffic by a drooling moron, as the shop attendent to literally be the rebirth of Einstein or Euler.
39:04 i would say restricitng player gameplay is a really bad take. If a player actually can succeed with weapons then they should be allowed to. Punishing players for not playing "standardly" gives matches where at high levels you get stagnant games of exact matchups. If someone wants to blitz with a shotgun and dies, the kill penalty already applies. Also, if they still manage to win with a technically "worse" gun, they still deserve points since mechanical skill should also be rewarded just as much as technical. Secondly, the flash mechanic is kinda bad, as there may be situations where a flash may be used to blindly enter a site while the opponent cant see. Honestly, if you want to make this affect elo, make it so that when someone flashes a teammate or gets flashed they get elo loss through a vote system or something
That's the problem with any subjective rating system. The main strength of Elo and Glicko systems is that they assume only one thing: a player's performance in a single game is govern by a normal distribution, the rest is just pure probability math - that's why they are totally objective.
Im at 5:20 atm. old Fastcup on cs 1.6 had this system. They watch how you played and against what skills. Then, if you are in minus elo but won, then you get your minus elo /3. If you are minus elo and lost, you lose full. If you are plus elo and win, you get all elo, if you are plus elo and lost, you get your plus elo/3. This is the problem about everything that isn't impact - people will bait, play for kills (no all, but many) etc. etc.
31:00 one small mistake, at least I see it as one, you should say the name "Fantasy esports" for people that don't watch but listen or something. felt a bit off u not saying it and going insta with things it provides.
Elo hell is just solo q variance. You are only 20% of the team, and this means you can get very unlucky with teammates for a very long time. If you truly are better, eventually, the law of large numbers will drag you up. Itll take 1000 games, which is unrealistic for most players who get frustrated and quit. Some tips to help mitigate the variance, stack with good players whenever possible, play at different times(like in the morning, i had much better teammates), and force yourself to stop after 1 or 2 loses to reset your mental Changing the elo system to account for individual performance is a bad idea because it is exploitable and has unintended side effects
Did you watch the video? "Elo hell is just solo q variance" is literally the point of the video. The idea is to limit the amount of variance and difference between actual skill (impact on the game) and displayed skill (rank). Accounting for individual performance being bad is just cope, the data is available and could be used to generate some sort of impact score, valve has the data to correlate every action from position to util usage to damage from positions etc. Anyone who says otherwise is coping. The exploit would be what exactly? Playing "optimally"?
@biemsu7426 yea impact score can be gamed. If you make the system care about stuff like impact instead of wins then things like entries and flashes might get tagged as "high impact" and then you have everyone buying 1 flash every round and holding w.
@@TheGingerjames123 But we are talking about using the data that WINS games as a way to evaluate impact If buying flashes and holding W won games then it would be rewarded. Going one for one on T side increases the round winrate. Going one for one on CT decreases the round winrate. You aren't understanding that whatever example you use is an issue with what is weighted rather than the system itself. With the data that valve has and the amount of processing power they have then they can find the most accurate weights for actions that positively and negative impact win rate.
@biemsu7426 I am by no means an expert but this has already been tried 1000 times and if you look for why it doesn't work I'd bet you'd find a better answer then I can give. With the entry example, a good entry might have big impact but you don't want 5 players trying to entry
The main problem I've had with the Premier system (other than just being based on W/L instead of incorporating individual performance), is the *magnitude* of the points that are awarded or penalized. This is predetermined as you enter a match, and has seemingly has nothing to do with the relative ranks of the players in the match. As the lowest ranked player in the match, I have lost the most points on a loss; and vice versa. It's furthermore impacted somewhat by the "streakiness" of your recent wins/losses, which adds to the feeling of unfairness on both up and down. Winning multiple matches in a row because you were lucky enough to play against opponents with network connectivity issues resulting in a 5v4 scenario for most of the match should not exponentially help your way up the points ladder.
problem is with adjusting elo to performance it quantifying and tracking those metrics which depend on playstyle, for example in csgo i play a more supportive igl role. i have plenty of games where i top frag but i prefer to play for controlling space and winning rounds through that so if you were to ajust based of damage or frags i would get a lower rating. The complexities of the different roles, mechanics and situational functions make it difficult however it doesnt mean the current win loss system is any good at all as shown by my league rank being so wildly inconsistent between different accounts, hard stuck iron on some and hard stuck gold on others
I don't know if distance from expected ranking is actually a valuable metric. Elo is not designed to rank players, it is designed to rate them. To try to assign a number to a player that, in the context of the system, approximately describes a player's skill. I could spoil your distance from true rank metric with a player who is rated 500, who is actually slightly more skilled than the 20000 players rated 501, and is not more skilled than any player rated above 502. The system says this player is 2 points from a "perfectly accurate rating", which feels close enough to me. The question "what score should you have for a given skill level" is, of course, not a trivial thing to answer (that's why we're making ratings systems, lol). But you want to be careful you're not using a hammer on a screw by making new statistics that ultimately confuse what their underlying numbers are representing in the first place. i.e., the Elo system is not designed to sort players by skill, but to approximate a player's skill with a number. It's also suspicious to me that all your examples of Elo ratings end in a steep climb or nosedive. Maybe that's what you were trying to point out. I'm not totally sure. It indicates to me that there's high variance. Because the deciding variable for a player being at the most extreme ratings when you stopped your simulation was luck. Supporting your argument that Elo is incapable of accurately rating players in a team-based game. I suppose that's your point ultimately. A rating system is meant to tease out skill from randomness, and in this case, for your simulation, it couldn't do it. I just wasn't seeing the statistical analysis I was expecting.
In a sample that big there will be instances of "unlucky" players who just lost 5 games in a row just before the end of simulation - those were used as an examples proving how bad Elo is. In this modified version those "unlucky" players' results were flatten by this magic skill correction which was based on *actual* skill value (something that doesn't exist in reality). What's funniest about this analysis is that you can omit wins and loses entirely from this simulation and it will work even better* because... if you know the actual skill value then why would you even bother with tracking match results? *) at this point the whole simulation is just a bit ineffective sorting algorithm
This explains why i get so many skill deviations in my teammates over the 120 premier matches i played 😳 Literally 14k points players with 13k hours and links to their 3.5k elo faceit accounts, and in the same game, a guy with 19k points with 1k hours, playing like a literal bot...
really great idea, but i do however disagree with your “fantasy” ranking catagories it incentivises players to farm flashes which wouldnt mean anything if it doesnt result in kills (edit: i hope this makes sense, idk how to say it lol)
exactly, you could throw as many useless flashes over bomb sites as you want and farm flashes, as well as -1 point for getting team flashed is ridiculous, how am i going to lose points if a teammate is griefing me and is flashing me on purpose
@@cstinybaby could be turned on its head, too, if a teammate is griefing and intentionally looking at all ur flashes. If flashes were used as a metric, they certainly shouldnt be counting teamflashes in any major capacity.
In Rocket League, if you lose an extremely close game (5 minute long OT for example) you still derank by the same amount you would have if you last 0-10. It's barbaric.
This is such an insanely in depth video its crazy! The production value and effort that has gone into the research and creation of this is just fucking crazy. You have earned my sub 1000x over, sharing this video everywhere I can.
This is a really good video - entartaining and well explained. I've been loving these types of data driven investigations into interesting things recently.
Shouldn't the skill of players in the simulation be normally distributed instead of uniformly distributed? The uniform distribution might be a big culprit in causing issues with your average rank deviation.
@@nikalasguadagnino1296 In the video they assigned skill incrementally by 0.01. Which means the number of players with the skill level between 700-800 would be the same as 1400-1500. Normal distribution visually is the bell curve graph. Most players are average performers while only few are high performers.
It should be, and that is not the only thing wrong with this simulation. His simulation is basically going to regress to a mean despite his input being garbage, and it doesn't accurately model the system
Really well presented and structured vid, I didn't really get some of it & the initial breakdown felt like it was in 1.5x speed, but it was well explained & I'm happy to say you earnt my sub & respect ❤
Yes, a System based only on wins can be influenced by bad luck with teammates, but given enough matches this luck will matter less and less until it averages out at having no influence at all. So the disadvantage is alot of noise, that averages out to zero over time. Other Systems that include Player statistics are all flawed because player statistics can never accurately portray how important you where to your win or loss, instead it introduces a range of stats which may or may not help your team win, but WILL help you rank up, therefore leading to a playstyle that isnt only focused on winning, but also focuses on something like kills or whatever you choose. Imo a ranking system should NEVER prevent the player solely focusing on winning the game, so only systems which only take win/loss into account can ever be considered. Imagine your teammate not calling out an enemy because he wants to get those kill points, or buying useless smokes and throwing them randomly to gain support points etc. You can always refine such a system, but you can never make it perfect, so you will always influence playstyle, which can never be accepted. If you are stuck in an elo for long enozgh, you simply arent good enough to rank up, everything else is just cope.
Fully agree. It's impossible to accurately quantify a player's impact on the match anyways. A top fragger who makes their teammates throw the match is many times worse than someone who does nothing but brings out the best in their teammates
This implies that all players always play at their best, there is no form taken into consideration, no trolls, bad moods and so on. The human factor is what ruins any theoretical statistics. Because of this, it makes sense that the win is not the only factor in TEAM GAMES. If I solo win 11 rounds but lose 13 because my team is that bad, am I deserving of a punishment? I know these are outliers, but as long as we do not have a percentage for these outliers, we can only presume their impact. I would confidently say at least 33% of games have a human factor upsetting the statistics. These games still count just the same as one of the non-outlier games.
@@Sp33dyJohn It might suck to hear, but yes, you still deserve to be punished for losing. Being the winningest loser is unimportant, all that matters is winning the match
@@Sp33dyJohn I don't think you really understand how the statistics of matchmaking work. Saying that 33% of games have a human factor upsetting the statistics is extremely ambiguous, but assuming that some of these cases have a big importance and they are somewhat common, you should easily achieve a higher winrate than other people of your skill if you don't troll games. I can't do the math because you didn't provide enough numbers, but assuming some games are an automatic loss because of a teammate trolling, then if you NEVER troll (you can EASILY control this), your team should have less chances of someone trolling than the other team. The conclusion should be obvious, if some people throw games and you don't, your winrate should be even better than the average for your skill. Obviously this doesn't directly correlate to reality, but if you don't believe that not contributing to these outlier games (doing things such as just taking a break when frustrated or putting an effort towards shooting enemies instead of allies), then please tell me how common these outliers are and how impactful they are towards the winrate of a team. This isn't even a crazy revelation, a lot of guides tell you to take breaks and stop losing streaks because it's really effective, you are literally just boosting your winrate by a few percentages for FREE. If you truly deserve to climb then your probability of winning should be above average to start with, therefore if you don't lose intentionally and you only play in your good moments your winrate will be high enough to guarantee you will climb if you play a decent amount of games.
Such a fantastic video, was going to make a video on elo inflation across in the real world but this video blows whatever i had planned out of the water.
@@-_-NiX-_-99-_- that's the fun part, valve's system makes absolutely no sense and nobody knows how it works, but it sure as hell isn't Elo because Elo makes sense
depends if you think their frags and other in-game score metrics actually truly represent higher/lower skill. i personally dont think its MORE reliable then the win/loss metric. although standard metric is bad, I think frag based will result in something slightly more inaccurate actually. considering thet only variable which matters about the game is the win/loss, i think it makes sense thats the only tested factor. it just means that realistically a win/loss shouldn't change their elo much (very little elo change per game = good) , this system relies on law of large numbers i suppose
Now that's an elaborate way of saying i suck and want to blame my teammates for not ranking up 😂 Jokes aside: i really respect your work and effort put in. Still it's so flawed that i cannot agree. Your entire simulation is based on the premise that skill is a one dimensional thing without fluctuation and everyone always just solo queueing. Yes in theory performance based ranking would be much better. In reality, especially in valves premier, there's so many factors that you can never take into account. Different maps, different playstyles clashing, different teammates matching well or not well together, personal preferences, personal Situation like form, health, alertness the list goes on. For example yesterday i was mad, lost 3 games as bottom fragger. Comeback 2hrs later cooled down and win 2 matches as top fragger. Yeah i would lose some more or less elo but it doesn't solve the problem of someone not performing to their ranks standard.
So no i don't want any Performance based ranking that i don't understand and therefore makes me mad and i don't want a Performance based skillrating ppl do understand so they can just play for points. In the end win or lose is what counts. Everyone knows that so everyone can decide to do whatever it takes to win and therefore rank up. Yes it sucks sometimes to play with ppl you don't feel like belong in your skillbase. But that's just a reality Performance based ranking cannot change in my opinion
bro is doing gawd's work over here. I've ranted, raved, sweated through so many ELO bounces, all the while saying "it can't just be based on wins and losses when some players are contributing more than others." I hope this gets Valve's attention, as their leaderboard is probably one of the least optimized features they've added to the game (all things subtick aside), while simultaneously having so much potential for success if done right. thanks for a great watch!
we should also be able to see more stats about players we're queued with -- at the very least their basic W/L streak and overall player rating. instead of having to check leetify etc. etc.
I have played around 500 CS games and I remember one game where I had a teammate who played with only Deagle and bought nothing else, not even armor. He was not technically a smurf as that was his main account and he just like playing against people with *far* worse game mechanics than we had. We even won that game. It was also one of the games that made me stop playing.
So in the fantasy portion of the video when making points for certain guns, it would make sense to base the points off of the money you receive from a kill. Typically certain guns reward higher money per kill if they’re harder/off meta etc. Edit: Also Leetify tracks trade attempts etc. so that is a stat that is currently tracked by some third party software
this is a really amazing quality video, and you definitely went through a lot to make it and it required a lot of dedication. The only thing I would maybe suggest is changing the thumbnail so it would be more clickable/get more views. There's a chance the thumbnail is fine, it's just a suggestion
I had this problem with Overwatch for so long back in the day-- unless you were literally like a grandmaster, top 500 God at the game it was IMPOSSIBLE to single-handedly carry a team to a W. and lo and behold, OW elo hell (gold-diamond) would place you in braindead teams constantly. i'd main DPS and have gold elims, damage, assists, captures, defenses, etc EVERY single game and still lose 8 in a row, keeping my elo in the dumpster despite being pretty fkn decent at the game.
That's literally because you're focusing on wrong things Gold medals dont mean anything unless you actually win and this highlights what this video showed in the end - there isn't an adequate way to judge someone's performance except for the outcome being win or loss. There is a ton of stuff you need to account in cs and because of the nature of the game, even more in ow. Also, you don't have to single-handedly carry games, and there is no need to be a god to win games. Been playing since 2017, got stuck in diamond, improved, got masters, played scrims primarily, improved, got gm, lost a few to lose gm and then got a 14 losses streak. It happens. Went back to it, played alt accounts to boost skill in one-tricking or to get to masters and sell, got back on main, kept up with masters while somewhat chilling, got gm for several seasons in ow2. It's all a process and dedication and actually understanding the nuances help a lot. Most recent achievement I got is one-tricking as widow to masters. Took nearly 600 games because even in gold people try to counter you as sombra dva winston, but I got there anyway. Nothing crazy about my skill, just determination and practice. Not even VOD reviewing, although it could be very valuable at some points.
@@YTHandlesWereAMistake I’m not gonna respond to all that but I’ll say this- you don’t need to single handedly carry games to win, but you need your team to be filling their roles at the very least. That’s my point, not that it’s impossible to get to GM or whatever you think I said. In order to win a game with an incompetent team you have to go far above and beyond what your role or your elo requires you to do in OW, or at least that’s how it was back in the day and in elo hell, you got an incompetent team nearly every other game. Winning in OW requires far more team cooperation than in CS, which leads back to my point about the individualized elo system working even less. I played OW religiously since the beta until around 2018, and still have like 1200 in game hours. Cruised in diamond for a little while when I did grind comp but I’d still shit on master lobbies.
@@CoolDrifty you only need to win more than you lose. Even 50.5% winrate is enough, and that's been the same way in the days you played, idk why call that elo hell
@@YTHandlesWereAMistake I actually had a slightly positive win ratio over my playtime and yeah lol, with a 50.5% win rate it’d only take you like 1000 games to rank up, I wonder why it’s called elo hell
Tradeable death cooooullld be tracked via "crosshair-on-target time", which would start each time there is an exchange and clear LOS between two players, so no wallbangs. So in essence, if both players aim at eachother and once the first bullet is fired, it would check if the other fired back during being on target. It is a bit hard, since there is recoil patterns, or just pray and spray, but maybe it would balance itself out in the grand scheme, especially with +-1 points for kills and deaths. Comms tracker is impossible since premades just use external options and it would be easily abusable.
Your assuming you know what the best way to play cs is and modeling your skill rating after that, which is the exact reason a system like this wouldn't work. With your system an entry fragger gain less elo then the second or third anker even though their impact on the game might be higher. A lurker in a loosing game would be highly favored since the enemies often will start running around chasing for the kill and are much easier to kill giving the lurker many more easy kills then the entry fragger even though their impact on the game is less. Even if you find a skill rating system that works it will most likely only work at a certain ranked range since what you need to do to win a match in 500 elo is completely different from 1500 elo which is different from 3000 elo, also even if you find a way to make a system that accounts all of this the system will likely break overtime as people change how they play the game. That or you make the game stale forcing high elo players into one way of playing the game, punishing anyone that tries something new that might be better for the team but worse for their skill rating. Also your simulation is flawed, a person with 63% win rate is not going to end up at a lower rank then they started at.
Yeah this. Frankly im not even sure if elo hell actually exists, and if it does exist it certainly isnt something that can be fixed realistically in a good way. I mean really, if youre actually so much better but getting pulled down by shit teammates then eventually youre going to be at a level where youre just way better than everyone else and can do way more than 20% of the work of your team consistently, and because youre better than everyone else you should rank back up. If you consistently fail to rank up it sounds more like you have stopped improving or are improving at the wrong things and just want an excuse for why you aren't good enough for higher level play.
@@andrek6920 I saw a great comment about elo hell some time ago. "Look at any online game community, and you'll see people complaining about "the trench". The trench is a mystical point where it's impossible to keep climbing the rank ladder. Nobody can ever agree on what rank the trench resides at, or how large the trench is, but much of the community agrees that it exists and that they're stuck in it."
His simulation is straight trash. It's like, wow. You're as good at maths as you are at CS. Terrible. Skill is normally distributed. Reaction times, intelligence, manual dexterity, the things that make up playing, all are better approximated normally than uniformly. So his data set is garbage. His rules also use random selection. Matchmaking is matchmaking. Not random. This is a classic case of selection bias: he has assumed that this is random, when matchmaking follows some rules. It's bad and dumb. Finally, if those flashes are contributing to winning, they are included in the score. Because they WON THE GAME. I hate the algorithm CS2 uses. It blatantly has poor bounds on the matches it makes, giving some teams a higher team elo, or a much higher elo player in their group, which is bad for match fairness. This also has almost no impact on the points given by the system: it is mostly based on match history, start losing more than winning, start losing points at a higher rate. Start winning more than losing, start gaining points at a higher rate, with a larger potential loss limit. However, the problem is nothing to do with "ELO HELL". CS2 does not have an elo hell. It has a bad and unstable matchmaking system, where you can finally end up at a rank where you're having games you enjoy, with enemies that challenge you but you can overcome, with teammates of a similar skill and knowledge, and then you lose two games and the next loss will take you down 400 elo and put you back more than 8 hours. It is not "Elo hell" it is an unstable system. It's so unstable that I've played at 12k, I've played at 6k. I'm around a 10k player, but whenever I play, wooooaahhh, going up, wooooaaaahh going down. The system isn't designed to settle me into 10k matches I enjoy. It's designed to put me into a 12k match when I reach 10k, knock me back down, get streaky, get down to 6k, where I can't help but dominate the non-aimers down at that elo, ruining the game for them, until I'm back around that 8-10k range I do best in. I could win a 3v5 game against a group of 5-6k players. It is really that simple, they do not have the gamesense or the aim. I'm going to kill 2-3 of them every single round and win every single clutch they throw at me. Elo hell simply does not exist, you eventually fall to a level where you will dominate the enemy, and if you never do, then you are bad and should try to improve by thinking about how you do things a little more. I am not a "good" player. I am an entirely average player. It is not a flex, it's a simple fact that when you put a 50th percentile player into a game with the bottom 25% of players in the world, he is going to farm them. That's what's wrong with the system. The chess ELO system would ensure I stayed there until I started to improve or get worse. The CS2 system instead forces me to ruin games for people who haven't even worked out that forcebuying the nova every round is a bad idea. It makes me spend more time grinding against teams that are absolutely not capable of dealing with me, only losing when I have a truly awful team or griefers (because CS2 punishers griefers with elo penalties, not game bans), which only slows me down. Instead of giving me balanced matches I enjoy, it gives me homework to get there. Also, HLTV and Leetify already do these things, like he mentioned. And they suck. HLTV will tell you if you were killing or assisting more than dying badly. Leetify will bloat that out by applying a silly algorithm to it that says "The opening duel matters, not dying mid round matters, and kills in the endgame matter" which rewards baiting and lurking, and punishes entries who don't frag out, punishes players who take objectives, and rewards players who capitalise off their teammates to win-at the end of the day, nobody cares as long as it's a win-only an idiot cares that an entry who's distraction and calls got his team two kills and a plant got shot in the process, but individual systems and KPIs DO NOT WORK. If I go in last every time, flash for the team, get plant, finish off 2-3 players, or if I lurk unders on mirage and get a frag on the con guy when he goes chasing kills every round, positions which are low impact but high reward, leetify will say I am awesome. If I go in first and tag two people, call them out to my team, and get them the plant, leetify will say I am bad because I am "losing opening duels" and have a "poor mid round K/D" but we will win and my elo will increase. You want to bring maths into it. Fine. How about you look up what happens when companies introduce KPI metrics to their jobs to measure individual performance. I'll wait. What usually happens is people focus on the element that increases their KPI and ignore everything else, reducing overall effeciency. You measure completed units? Ok. Nobody is doing the prep work anymore they want to sign off on the finished one. You measure time on the clock? Ok. People work slower to get more time in. You measure how quick they do a task? They'll do the simplest tasks with the least complications that take the shortest time to increase their average time under allocation. Holistic approaches work, that doesn't. If you want to track flashes for rank xp, fine. People will throw more useless flashes. You want to track smokes. Fine. People will throw more useless smokes. The judge of how good those flashes or smokes are is WHETHER OR NOT THEY MAKE YOUR TEAM WIN. The holistic measure is important. Online matchmaking needs improving, but it needs it in the algorithms it uses to design a suitable match-these matches need closer matchups, and it needs it in the rewards-the games should give close matches and small elo gains and losses. Drastic gains and losses should only be used in the placement algorithm used in the games that are played to unlock a matchmaking rank.
Imagine that points are calculated based on kills. All players will try to get as many kills as possible and will forget the main purpose of playing a tactical shooter.
Got 1.5k games just in premier alone. You're out here doing Gods work. Also got a 35 week streak just because I don't want to let it go. Been going since they released it. Man we need new content and an updated ranking system would be cool.
The ranking system on premier is actually not elo based system. If you stack with higher ranks, you'll be getting like +365/-105 on average and you can climb with a ridiculously low winrate like 33% (when average enemy's team rating vs your average team's rating are the same). On premier it's actually way more important the ranks you stack with, for your rank, than how good you play/how much value over time you give to the team
I know that I'm only at 10:10 , but I feel like I would analyze this problem with a Montecarlo simulation that you can do in python, with a simple RNN to create the function model to produce the desired ranking function.
A rating for personal perfomance would be great because we have so many parameters in cs to show your performance in a match: kdr, adr, nate damage, entry kill, trading and many more. the worst part about only counting the team performance is, you give trolls and afks so much power about your elo rating. no matter how good you play, if one guy decides to be afk or to troll, the match is almost lost already, and you have nothing to fight for because your performance is not honored. This current elo system is so teriible, i stoped caring and just play
I agree with everything except the point system. I buy SMGs exclusively because it can be so disruptive to the enemy economy when they are on CT side and it can be a big boon to us when we are on CT side. Dictating the wider teamplay-style to some degree is okay but trying to enforce individual play styles by punishing players who do interesting things with the game's meta is toxic. Your system rewards me for out-performing the enemy with inferior weapons but punishes me before I even leave spawn, I understand that you acknowledge that it might have flaws but to even suggest this means the system is too married to the current meta. Having 1 Mac-10 player on T side means CTs spend 350 extra each to not get clowned on by me, having an MP9 on CT side means I can buy a gun and nade every round regardless and farm money, when we win on CT, I get to drop a free gun every 3rd round (an still buy, assuming an average game).
The true cause of ELO hell is that ELO exists in the first place. It's a tool designed to create false engagement with a game, an interest in the status of number go up, not of actually becoming skilled at the game.
Elo rating works well for chess. It also works decently well for fighting games. But for team games, no matter which, there isn't just your rating at play, but also the rating of your teammates. And how well you act as a team, something completely independent from your individual ratings. And a group of worse players with good coordination and teamplay can easily win against a group of better players that all play alone. Their individual rating might be worse, but their team rating is higher.
There is no fix for this problem. Teamshooters cant be rated with a elosystem because mathematically you are responsible for 20% of the Teams performance, if you consider 5v5. So you could say "Lets rate the Player upon his own performance like K/D or ADR for example". This would result in people baiting 24/7, wich isnt a good idea. So yeah... there could be a fix in making a personal ID mandatory for playing FaceIt for example, wich makes every Player personally attached to his Ingame-Persona. Meaning that behaviour and performance are important, wich would result in way better Communications and Teamplay. Dont forget were talking Teamshooters here, individual performance is important but communications and teamplay are crucial to climb elo, thats why mostly 2-3 Player lobbys grind effectively in Faceit. The key is locking ID to a Players account so that Cheaters cant continue after getting banned and vocally violent and crazy people would face serious repercussions too. Thats the only logical and doable solution if youre not playing a 1on1 Game like in the old days. If 1on1 games would be a thing we could implement the strict elo win/loose rules again.
But your rating will also be heavily influenced by flash assists, trades and potentially successful entries. The whole point of the video is that the current elo system doesn't give any weight to individual performance in any way, which results in huge disparities of deserved ELO and actual ELO. No system is perfect, but a system that rewards different aspects of individual performance is more accurate, as per video's conclusion
@@Dapplication Yeah, totally agree with you. I was just venturing off in a deeper problem with anonymous Players in a team-based game. Im pretty sure that we should work on the root of problems before considering tweaking individual elo, as a lot could be instantly fixed by making people accountable due to permanent ID lock on accounts. Especially in a Teamgame.
I dont even think valve, riot, and literally every comp game uses ELO system. The system they are using was made to keep you playing in hopes of winning and climbing. There is no fing way they are using ELO system, it was not created for comp games. So basically you need to be insanely good at the game to be able to climb or have duo or trio to win more games. Or grind ur life away to reach high-elo
@@fica1137 Same for Valorant and league of legends. Am talking about big comp games, They all do the same thing. Nothing is fair about their rank system.
Remember, kids, it's not coping if you're justified and you're able to articulate yourself
Explain how some people rank up then 💀
@@imwawo 3/4/5 queue stacking every game. most people suck solo by themselves with random teams.
@@wardog9959nah I know plenty of people who solo queued to high elo, myself included. You’re just bad, unfortunately. One person can absolutely carry games consistently with the right playstyle
@@imwawo better players still rank higher in the current elo system
@@imwawodid you watch the video? Do you have multiple leaning disabilities? I’m genuinely asking… he proved that it’s possible for that happen with the elo system that was in place for the longest time. It’s like you literally just want to argue
that VONDAS guy is terrible wtf
Colombian Vondas is goated
vondas PEEK
You're stuck in Elo hell? Welcome to Elo hell 2.0 "a prison of your own creation".
The quality of this video is that of a channel with 1m+ subscribers, I genuinely thought it said 1.4m subscribers, not 1400. Insanely good video
strongly agree, well done mate
agree
Wildly good. Instant sub. He could use a new mic, though.
Absolutely. This guy can expect to grow quickly.
Quality is related to subscribers?
I guess not…
The WORK behind this is just crazy
Happy to help :)
Legend!
thanks
Madlad! 🤙
A big issue is that the top fragger might not be a good player. He might just be sacking team m8's for kills. Maybe he doesn't do that and instead just doesn't give call outs, so he has all the infromation from the other players, but no one else has any info from him, which means that he has a better chance to top frag.
Yeah it is very obvious that it would be better to take performance into account *If* there was an objective measure of performance. The problem is that there are not really any good measures. For example the rating based on fantasy esports seems absolutely terrible
If they are top frag they are better. You’re just coping
Nothing worse than a teammate never giving info yet thinking they are top shit for getting frags while loosing
You don’t need to be a fragger to be good at this game. Look at any pro team and they have players that do not frag out unless on rare occasions, they take on roles that sacrifice their stats for the win. Aleksib is a good example of a player that changes the outcome of fights just by being there with his utility and sacrifices often to set up his players. As long as your team is capable of trading each other well you can always perform. While there are better players at lower elos than can change the outcome of the entire match; I can guarantee if you look at your past games you’ve lost or lost several rounds due to your team or yourself not playing for the team. Swinging things you shouldn’t be swinging to get a kill and dying or randomly pushing somewhere without a team mate nearby to trade. While sometimes sure the person trying to trade will die too. But most of the time the positioning and crosshair placement of the individual who just killed your team mate is off and they’re open to being killed especially if you’re playing a crossfire. This game is about trading, multifrags happen but you shouldn’t be hunting for them. Your job is to take site or defend site and buy time for rotates. If you lose site within seconds even if you get one or two it’s pretty much over if the enemy has util to defend site. People need to think more tactically with this game, it’s not cod.
I feel the biggest inflator of kds is how many 1vxs they get into. If you get 3 kills but they where in a 1v5 those 3 kills matter very little compared to the two kills your teammates might have gotten the round prior and won because a save.
Clearly there's no problem with the simulation, since I'm the best player.
Great editing, pacing, narration - should have more views!
For how much shit talk chess and the system get in this video, they are perfect for each other - chess is a fair 1v1 game, in which the 3 way result, and the relative strength of the players are the only things that should matter when evaluating their ratings.
doesnt white win like 10% more often
@@sauceinmyface9302It's not that simple...
There are openings, every opening has many lines, and every unique variation can have a different average win % for each side.
White does move first, but chess played perfectly is a draw, since for every perfect attack there is the perfect defense, and at most levels, the color doesn't make much of a difference...
@@1GMitzy chess played perfectly is not a draw that’s a misconception based on our current technology’s ability to play chess being drawish. We have no idea what chess played perfectly would look like or its outcome because the chess engines of today are primitive in comparison to the futuristic and gargantuan computational power needed to solve chess, and only then would we know what perfect play would look like. Currently stock fish and alpha zero are just our best guesses with the technology we have
@@1GMitzy Ok but white still wins more, its not a symmetrical game like that
@@1GMitzy You clearly don't play chess at a high level then at all. White absolutely has an inherent advantage by going first. This is why tournament games give each player an equal chance to play on both white and black.
There's simply no way to truly measure individual performance in this game. Individual performance based system will encourage baiting and not dropping weapons to your teammates.
Let me explain on seemingly ridiculous example. If there would be a CS2 team that wins all the majors/tournaments in which one player is not even sitting on a chair but gives magical feet massage to his teammates which makes them insanely good so that they win games, he would be one of the top 5 CS2 players in the world. It does not matter what he's doing, all that matters is that he's doing something that is winning games. There simply cannot be other accurate indicator of who is a better player as winning games, it doesn't matter how it's done.
Rather than individual performance, what I think would be a good idea is to make elo gain/loss adjustment based on rounds won/lost. Like there would be a significant difference between 0-13 game and 11-13 game. It reduces variance but doesn't destroy the objective of the game which is winning rounds/games together with a team
Might be the dumbest shit I’ve ever read.
While I think that round based system could be a good idea it might just make it even scarier for solo queueing. 5 stacks it’ll obviously work bc the other team is better but for solo queueing you would be scared shitless to even queue for the fear of getting bad teammates and losing 4 times the elo you gained in a legitimately enjoyable hard though match with nice teammates that you won 13:11. It’ll make the negatives so much more potent and the positives weaker.
I'm not saying it should be only round based. Most or at least half of elo gain/loss should still depend of just winning or losing a game. Let's say that 60% elo gain/loss depends only on winning or losing, and then 40% is bonus dependent on how close the match was
I agree with this. Basically if you tried to incorporate every little thing a player does that has impact on the game into their elo gain/loss you'd probably just boil it down to did the team win or lose?
I like your suggestion of making it smaller or larger gain/loss depending on how the game went for sure.
@@KMurtheGamer But it's not about winning or losing, you can perform very well, carry your team, and still lose. Yes, there are games where someone is absolutely backpacking and they drag their team, kicking and screaming to a win, but the rest of the team does statistically poorly. You cannot look me in the face and tell me that a team that went 13-11, with stats lines looking like, 28-17, 19-19, 21-19, 5-21,16-14 and a team that went 13-0 with stat lines all in similar ranges should get the same elo upon their game finishing. That just logically doesn't make sense, and while it may seem that just looking at wins and loses provides a valuable measure of a team's skill(it doesn't, that's why win/loss %'s are flawed) but your rank isn't a measure of your team's skill, it's a measure of YOUR skill.
Plus, you can absolutely track things like drops and team economy, and factor those things into the rating. Average everyone's cash, and if everyone has within +/- $500 of that value in equipment at the start of the round, then boom, they've dropped. Or, just penalize people with massive eco differences when they're actually in the game and playing.
If you're just looking at wins and losses you're communicating that you care about nothing else, as long as you win you get those points. That can lead to situations where teams kick their under performing members. Before they were removed from competitive play, they acted as another life for a person on your team. So, kick your bottom fragger, and now you have an extra life for your top-frag, allowing them to just throw themselves in an entry. That was such a concern, even after they removed bots from comp, that they don't allow 4 stacks to kick the random in CS2.
There are so many things in this game that SCREAM that it would be a better experience if person performance was evaluated. It would make solo-queue less of a fucking slog, for one, as I'm no longer concerned about how my teammates are performing on a macro scale. If they lose the round but I did well, I can take solace in the fact that I won't be penalized as much because the flawed matchmaking system put me in a game where I'm significantly better than my teammates. And conversely, if I get into a game an I'm not performing well and team loses as a result, but everyone else did well, I can take solace in the fact that my poor performance isn't penalizing them as much. Valuing just wins and losses creates a such a toxic team environment because everyone has to perform at their peak, there can be no dips, otherwise it *IS* your fault.
In my opinion. U need 4 other people who like to play to improve and want to win and create strats and not just random q. Hardest part is finding people who are like that.
I mean, that is what the matchmaking system basically is. It's just that the skill rating falls below ELO, which is the main problem with any ranked game you could find
Nah, you need to focus on yourself and your impact on the game
This is easy to fix, there are websites for this.
@@OGfisk name me some? I personally found 1-2 players but they all either stop improving or stop playing the game.
A+ tier content from someone with less than 5.000 subs, you don't see that often! Props!
Firstly, this video has exceptional quality. People have pointed this out and I thoroughly agree.
Unfortunately, the actual content is a bit of an issue, as numerous people have pointed out. The first simulation has a bunch of problems and is clearly not an accurate representation of CS matchmaking. Since your entire argument is based on this simulation showing a negative result, your entire argument is flawed.
However, more fundamentally, the video (even if everything you said were true!!) doesn't actually mean much regarding fixing elo hell. Players would be more accurately ranked, maybe, but they would still be hardstuck and would still blame that on everything that isn't themselves. In the current system, if you are struggling to rank up, it is because you aren't winning games, and thus it is a skill issue. People complain about elo hell anyway. In your system, people would struggle to rank up because of personal perfomance and would complain about elo hell anyway (perhaps the metrics and figures are wrong, perhaps the game doesn't weigh wins enough, perhaps they are still getting god awful teammates every game!). Elo hell is a myth based almost entirely around people perceiving their skill level to be higher than it actually is, and this system doesn't fix that underlying issue, so people would still feel bad when they are stuck at the same rating for six months.
bro your video making and programming skills just shows me that CS is played by very smart and capable People. i feel honored to be a part of this community, BLYAD!!!
In theory using wins and losses alone is the best system, because it does not create any perverse incentives where a player could boost their own ranking at the cost of making their team more likely to lose. The problem is that in a team game there is so much variance (random factors outside of a player's control) that it takes too long for a pure win/loss system to converge.
this video should pop off wtf why is it stuck in elo hell
I think the modern way is to use machine learning to identify the features that mostly contribute to the outcome, and scale the lost/won points based on that.
its funny cuz we've had HLTV ratings for over a decade and leetify for years now, both of which are specifically designed to gauge individual performance (impact), and leetify actually compares your stats across multiple skill groups. and yet valve decided to ditch their original comp ranking system for a slightly better yet still terrible one.
It's very easy to make an unrealistic simulation, which I think is all that has been demonstrated here. There are a couple decisions which are highly suspect:
1) The skill distribution of players in the simulation is flat, actual player skill probably fits closer to a normal distribution, and most player rating systems have this assumption baked in, so you are likely warping the results by doing this.
2) The description of matchmaking does not make sense "one player is chosen at random, then random players with in 100 ranks are chosen to form a match", and "each player has exactly 1000 matches" are not compatible statements, you would expect some variation in # of games played if matches are forming randomly, if you are removing people from the pool as they hit 1k games then you are messing with skill distribution of players, this may explain why many of the ranking graphs have huge swings for the last 100 or so games. I would not be surprised if the system actually looked like it was performing better while the average player is only at ~700 games.
3)The matchmaking system as described is much less likely to form matches with top rated and bottom rated players since there are fewer player within 100 rating of them, this likely makes it harder to 'decompress' skill ratings compared to player skill. It also biases the 2nd team to lose since it is definitely giving them worse players (the 1st player should be better than the 2nd player, the 3rd player should be better than the 4th player, and so on) these are both likely small effects, but it's an additional layer of bias.
It's very important to make sure a simulation actually works before trying to use it to draw inferences
I'm amazed there aren't more posts highlighting these issues.
Definitely wondered about the normal distribution coming into this
jesus you guys are expecting paper level of simulation from a youtuber.
everybody knows elo is a bad system to rank ability in modern gaming.
these 3 points you made would not alter the point of the video
@@JPClow3 except point 2, if you look at the graphs, would
@@JPClow3 No, they're expecting if you want to bring simulations and statistics into it that you use very basic high school level statistics and the stated rules of the system.
This is a shoddy simulations which doesn't understand the system (players are selected from within a certain bound of the current level is a fundamental rule of matchmaking) and a normal distribution is so common it is ASSUMED that a normal distribution describes most things as an approximation until demonstration shows that it doesn't.
Most people are average. Average intelligence, average height, etc. The people who are extremely smart, tall, whatever, are vanishingly small. A bell curve normal distribution has the majority of people around the middle and more accurately describes the situation.
If you want to approximate things with a flat distribution, go ahead. It's equally likely to find someone of perfectly average height and the tallest man alive, as well as the shortest. You're just as likely to be cut off in traffic by a drooling moron, as the shop attendent to literally be the rebirth of Einstein or Euler.
39:04 i would say restricitng player gameplay is a really bad take. If a player actually can succeed with weapons then they should be allowed to. Punishing players for not playing "standardly" gives matches where at high levels you get stagnant games of exact matchups. If someone wants to blitz with a shotgun and dies, the kill penalty already applies. Also, if they still manage to win with a technically "worse" gun, they still deserve points since mechanical skill should also be rewarded just as much as technical.
Secondly, the flash mechanic is kinda bad, as there may be situations where a flash may be used to blindly enter a site while the opponent cant see.
Honestly, if you want to make this affect elo, make it so that when someone flashes a teammate or gets flashed they get elo loss through a vote system or something
That's the problem with any subjective rating system. The main strength of Elo and Glicko systems is that they assume only one thing: a player's performance in a single game is govern by a normal distribution, the rest is just pure probability math - that's why they are totally objective.
Im at 5:20 atm. old Fastcup on cs 1.6 had this system. They watch how you played and against what skills. Then, if you are in minus elo but won, then you get your minus elo /3. If you are minus elo and lost, you lose full. If you are plus elo and win, you get all elo, if you are plus elo and lost, you get your plus elo/3. This is the problem about everything that isn't impact - people will bait, play for kills (no all, but many) etc. etc.
31:00 one small mistake, at least I see it as one, you should say the name "Fantasy esports" for people that don't watch but listen or something. felt a bit off u not saying it and going insta with things it provides.
Elo hell is just solo q variance. You are only 20% of the team, and this means you can get very unlucky with teammates for a very long time. If you truly are better, eventually, the law of large numbers will drag you up. Itll take 1000 games, which is unrealistic for most players who get frustrated and quit. Some tips to help mitigate the variance, stack with good players whenever possible, play at different times(like in the morning, i had much better teammates), and force yourself to stop after 1 or 2 loses to reset your mental
Changing the elo system to account for individual performance is a bad idea because it is exploitable and has unintended side effects
Bro I swear the people playing during the day are much more sane 😅
Did you watch the video? "Elo hell is just solo q variance" is literally the point of the video.
The idea is to limit the amount of variance and difference between actual skill (impact on the game) and displayed skill (rank).
Accounting for individual performance being bad is just cope, the data is available and could be used to generate some sort of impact score, valve has the data to correlate every action from position to util usage to damage from positions etc. Anyone who says otherwise is coping.
The exploit would be what exactly? Playing "optimally"?
@biemsu7426 yea impact score can be gamed. If you make the system care about stuff like impact instead of wins then things like entries and flashes might get tagged as "high impact" and then you have everyone buying 1 flash every round and holding w.
@@TheGingerjames123 But we are talking about using the data that WINS games as a way to evaluate impact
If buying flashes and holding W won games then it would be rewarded.
Going one for one on T side increases the round winrate. Going one for one on CT decreases the round winrate.
You aren't understanding that whatever example you use is an issue with what is weighted rather than the system itself. With the data that valve has and the amount of processing power they have then they can find the most accurate weights for actions that positively and negative impact win rate.
@biemsu7426 I am by no means an expert but this has already been tried 1000 times and if you look for why it doesn't work I'd bet you'd find a better answer then I can give. With the entry example, a good entry might have big impact but you don't want 5 players trying to entry
The main problem I've had with the Premier system (other than just being based on W/L instead of incorporating individual performance), is the *magnitude* of the points that are awarded or penalized. This is predetermined as you enter a match, and has seemingly has nothing to do with the relative ranks of the players in the match. As the lowest ranked player in the match, I have lost the most points on a loss; and vice versa.
It's furthermore impacted somewhat by the "streakiness" of your recent wins/losses, which adds to the feeling of unfairness on both up and down. Winning multiple matches in a row because you were lucky enough to play against opponents with network connectivity issues resulting in a 5v4 scenario for most of the match should not exponentially help your way up the points ladder.
I think it's almost entirely based on win and loss streaks from my experience. Although I don't really touch premiere, faceit is so much better.
problem is with adjusting elo to performance it quantifying and tracking those metrics which depend on playstyle, for example in csgo i play a more supportive igl role. i have plenty of games where i top frag but i prefer to play for controlling space and winning rounds through that so if you were to ajust based of damage or frags i would get a lower rating. The complexities of the different roles, mechanics and situational functions make it difficult however it doesnt mean the current win loss system is any good at all as shown by my league rank being so wildly inconsistent between different accounts, hard stuck iron on some and hard stuck gold on others
The unmanageable pain wasn't from the surgery. It was your body trying to push out the poison from the Python bite you had suffered.
I don't know if distance from expected ranking is actually a valuable metric. Elo is not designed to rank players, it is designed to rate them. To try to assign a number to a player that, in the context of the system, approximately describes a player's skill. I could spoil your distance from true rank metric with a player who is rated 500, who is actually slightly more skilled than the 20000 players rated 501, and is not more skilled than any player rated above 502. The system says this player is 2 points from a "perfectly accurate rating", which feels close enough to me. The question "what score should you have for a given skill level" is, of course, not a trivial thing to answer (that's why we're making ratings systems, lol). But you want to be careful you're not using a hammer on a screw by making new statistics that ultimately confuse what their underlying numbers are representing in the first place. i.e., the Elo system is not designed to sort players by skill, but to approximate a player's skill with a number.
It's also suspicious to me that all your examples of Elo ratings end in a steep climb or nosedive. Maybe that's what you were trying to point out. I'm not totally sure. It indicates to me that there's high variance. Because the deciding variable for a player being at the most extreme ratings when you stopped your simulation was luck. Supporting your argument that Elo is incapable of accurately rating players in a team-based game. I suppose that's your point ultimately. A rating system is meant to tease out skill from randomness, and in this case, for your simulation, it couldn't do it. I just wasn't seeing the statistical analysis I was expecting.
In a sample that big there will be instances of "unlucky" players who just lost 5 games in a row just before the end of simulation - those were used as an examples proving how bad Elo is. In this modified version those "unlucky" players' results were flatten by this magic skill correction which was based on *actual* skill value (something that doesn't exist in reality). What's funniest about this analysis is that you can omit wins and loses entirely from this simulation and it will work even better* because... if you know the actual skill value then why would you even bother with tracking match results?
*) at this point the whole simulation is just a bit ineffective sorting algorithm
This explains why i get so many skill deviations in my teammates over the 120 premier matches i played 😳
Literally 14k points players with 13k hours and links to their 3.5k elo faceit accounts, and in the same game, a guy with 19k points with 1k hours, playing like a literal bot...
they are 13k cause they get reported too much, get cheaters and just leave the game so they lose 1k ELO very often, I call that trustfactor hell
really great idea, but i do however disagree with your “fantasy” ranking catagories
it incentivises players to farm flashes which wouldnt mean anything if it doesnt result in kills
(edit: i hope this makes sense, idk how to say it lol)
exactly, you could throw as many useless flashes over bomb sites as you want and farm flashes, as well as -1 point for getting team flashed is ridiculous, how am i going to lose points if a teammate is griefing me and is flashing me on purpose
@@cstinybaby could be turned on its head, too, if a teammate is griefing and intentionally looking at all ur flashes.
If flashes were used as a metric, they certainly shouldnt be counting teamflashes in any major capacity.
Great video. An insane amount of work must've gone into this. Much appreciated.
Been stuck in Silver since 2014, got Silver Elite Master before Valve reshuffled the ranks and I got booted back to Silver 3.
😂
this video is insane in quality and editing. A+
In Rocket League, if you lose an extremely close game (5 minute long OT for example) you still derank by the same amount you would have if you last 0-10. It's barbaric.
This is such an insanely in depth video its crazy! The production value and effort that has gone into the research and creation of this is just fucking crazy. You have earned my sub 1000x over, sharing this video everywhere I can.
I like Python well enough, but the moment I saw Tkinter my soul was crushed
The DAMAGE part of the scoreboard does not update till the message that confirms round ends shows up
This is a really good video - entartaining and well explained. I've been loving these types of data driven investigations into interesting things recently.
Shouldn't the skill of players in the simulation be normally distributed instead of uniformly distributed? The uniform distribution might be a big culprit in causing issues with your average rank deviation.
Wouldn't that place them in different brackets?
@@nikalasguadagnino1296 In the video they assigned skill incrementally by 0.01. Which means the number of players with the skill level between 700-800 would be the same as 1400-1500. Normal distribution visually is the bell curve graph. Most players are average performers while only few are high performers.
It should be, and that is not the only thing wrong with this simulation.
His simulation is basically going to regress to a mean despite his input being garbage, and it doesn't accurately model the system
Great thumbnail choice this ones gonna do numbers👍
Really well presented and structured vid, I didn't really get some of it & the initial breakdown felt like it was in 1.5x speed, but it was well explained & I'm happy to say you earnt my sub & respect ❤
Yes, a System based only on wins can be influenced by bad luck with teammates, but given enough matches this luck will matter less and less until it averages out at having no influence at all. So the disadvantage is alot of noise, that averages out to zero over time.
Other Systems that include Player statistics are all flawed because player statistics can never accurately portray how important you where to your win or loss, instead it introduces a range of stats which may or may not help your team win, but WILL help you rank up, therefore leading to a playstyle that isnt only focused on winning, but also focuses on something like kills or whatever you choose.
Imo a ranking system should NEVER prevent the player solely focusing on winning the game, so only systems which only take win/loss into account can ever be considered. Imagine your teammate not calling out an enemy because he wants to get those kill points, or buying useless smokes and throwing them randomly to gain support points etc. You can always refine such a system, but you can never make it perfect, so you will always influence playstyle, which can never be accepted.
If you are stuck in an elo for long enozgh, you simply arent good enough to rank up, everything else is just cope.
Fully agree. It's impossible to accurately quantify a player's impact on the match anyways. A top fragger who makes their teammates throw the match is many times worse than someone who does nothing but brings out the best in their teammates
This implies that all players always play at their best, there is no form taken into consideration, no trolls, bad moods and so on. The human factor is what ruins any theoretical statistics. Because of this, it makes sense that the win is not the only factor in TEAM GAMES. If I solo win 11 rounds but lose 13 because my team is that bad, am I deserving of a punishment? I know these are outliers, but as long as we do not have a percentage for these outliers, we can only presume their impact. I would confidently say at least 33% of games have a human factor upsetting the statistics. These games still count just the same as one of the non-outlier games.
@@Sp33dyJohn It might suck to hear, but yes, you still deserve to be punished for losing. Being the winningest loser is unimportant, all that matters is winning the match
@@Sp33dyJohn I don't think you really understand how the statistics of matchmaking work. Saying that 33% of games have a human factor upsetting the statistics is extremely ambiguous, but assuming that some of these cases have a big importance and they are somewhat common, you should easily achieve a higher winrate than other people of your skill if you don't troll games.
I can't do the math because you didn't provide enough numbers, but assuming some games are an automatic loss because of a teammate trolling, then if you NEVER troll (you can EASILY control this), your team should have less chances of someone trolling than the other team. The conclusion should be obvious, if some people throw games and you don't, your winrate should be even better than the average for your skill.
Obviously this doesn't directly correlate to reality, but if you don't believe that not contributing to these outlier games (doing things such as just taking a break when frustrated or putting an effort towards shooting enemies instead of allies), then please tell me how common these outliers are and how impactful they are towards the winrate of a team. This isn't even a crazy revelation, a lot of guides tell you to take breaks and stop losing streaks because it's really effective, you are literally just boosting your winrate by a few percentages for FREE.
If you truly deserve to climb then your probability of winning should be above average to start with, therefore if you don't lose intentionally and you only play in your good moments your winrate will be high enough to guarantee you will climb if you play a decent amount of games.
Absolutely baller video my guy, insane brain work, and an insane product!
Such a fantastic video, was going to make a video on elo inflation across in the real world but this video blows whatever i had planned out of the water.
@@1bird_d The more understanding of elo the better. Feel free to make whatever video you want, use our simulation code if you want to.
Only problem is cs2 premier doesn't use Elo :/
Oh man I didn’t know this. How does it actually work? Sometimes I lose 100 and or win 300 or lose 300 and win only 100.
@@-_-NiX-_-99-_- that's the fun part, valve's system makes absolutely no sense and nobody knows how it works, but it sure as hell isn't Elo because Elo makes sense
Ah, another victim of the global interpreter lock 😔🙏
Honestly love this video. Rather the facts and journalism
Bro.. amazing video! This obviously took a looong time to do. Subbed, liked!
depends if you think their frags and other in-game score metrics actually truly represent higher/lower skill. i personally dont think its MORE reliable then the win/loss metric. although standard metric is bad, I think frag based will result in something slightly more inaccurate actually. considering thet only variable which matters about the game is the win/loss, i think it makes sense thats the only tested factor. it just means that realistically a win/loss shouldn't change their elo much (very little elo change per game = good) , this system relies on law of large numbers i suppose
I’m 12 minutes in and I’m astonished on how well this video is done already. This needs to go viral
Now that's an elaborate way of saying i suck and want to blame my teammates for not ranking up 😂
Jokes aside: i really respect your work and effort put in. Still it's so flawed that i cannot agree.
Your entire simulation is based on the premise that skill is a one dimensional thing without fluctuation and everyone always just solo queueing.
Yes in theory performance based ranking would be much better. In reality, especially in valves premier, there's so many factors that you can never take into account.
Different maps, different playstyles clashing, different teammates matching well or not well together, personal preferences, personal Situation like form, health, alertness the list goes on.
For example yesterday i was mad, lost 3 games as bottom fragger. Comeback 2hrs later cooled down and win 2 matches as top fragger.
Yeah i would lose some more or less elo but it doesn't solve the problem of someone not performing to their ranks standard.
So no i don't want any Performance based ranking that i don't understand and therefore makes me mad and i don't want a Performance based skillrating ppl do understand so they can just play for points.
In the end win or lose is what counts. Everyone knows that so everyone can decide to do whatever it takes to win and therefore rank up.
Yes it sucks sometimes to play with ppl you don't feel like belong in your skillbase. But that's just a reality Performance based ranking cannot change in my opinion
bro is doing gawd's work over here. I've ranted, raved, sweated through so many ELO bounces, all the while saying "it can't just be based on wins and losses when some players are contributing more than others." I hope this gets Valve's attention, as their leaderboard is probably one of the least optimized features they've added to the game (all things subtick aside), while simultaneously having so much potential for success if done right.
thanks for a great watch!
we should also be able to see more stats about players we're queued with -- at the very least their basic W/L streak and overall player rating. instead of having to check leetify etc. etc.
More vids like this and there is no way you don’t blow up on TH-cam! Great video.
just answered the survey chief! thank you for the work
I am in elo hell level 7 😢 +1.70 leetify… 😢
Such a good quality video. Tried to give you quite indepth answers for your faceit poll ;) Gained a sub
imagine being VONDAS_
VONDAS_ PEEK th-cam.com/video/m2WneNUWxug/w-d-xo.html
a fate worse than death
@@MF_VONDAS💀
Pouring out a peek for vondas th-cam.com/video/m2WneNUWxug/w-d-xo.html
It's hard man
I run MP7 only. Farm money and buy stuff for team only. This system would be fun and crazy to try. I would definitely learn flashes more too.
I have played around 500 CS games and I remember one game where I had a teammate who played with only Deagle and bought nothing else, not even armor. He was not technically a smurf as that was his main account and he just like playing against people with *far* worse game mechanics than we had. We even won that game. It was also one of the games that made me stop playing.
6min in is an explanation of how old MM ratings worked.
Insane video. So fkn good man. Super impressed
Insane Video With a nice topic!
Keep it going.
Bro the efort behind this is crazy much respect
Damn what a Video, especially for the size of your Channel. Hope you blow up.
Earned a sub. This was highly interesting to watch. Thanks.
Open aource your code and in 1 week you gonna have a rust version running the 15 millions games in 15 seconds
The code is open source, linked in description
So in the fantasy portion of the video when making points for certain guns, it would make sense to base the points off of the money you receive from a kill. Typically certain guns reward higher money per kill if they’re harder/off meta etc.
Edit:
Also Leetify tracks trade attempts etc. so that is a stat that is currently tracked by some third party software
this is a really amazing quality video, and you definitely went through a lot to make it and it required a lot of dedication. The only thing I would maybe suggest is changing the thumbnail so it would be more clickable/get more views. There's a chance the thumbnail is fine, it's just a suggestion
"Satan himself was born in an burning orphanage" I'm stealing that whenever I have a mild inconvenience
This man randomly decides to sound like WarOwl in the middle of sentences
So true lmao
6:25 and after is crazy
And yet, "Sounds like a skill issue to me" is that they'll say
Life is a big ass elo hell - Sun Tzu
I had this problem with Overwatch for so long back in the day--
unless you were literally like a grandmaster, top 500 God at the game it was IMPOSSIBLE to single-handedly carry a team to a W. and lo and behold, OW elo hell (gold-diamond) would place you in braindead teams constantly. i'd main DPS and have gold elims, damage, assists, captures, defenses, etc EVERY single game and still lose 8 in a row, keeping my elo in the dumpster despite being pretty fkn decent at the game.
also, holy shit this video deserves 100x more than 24k views
That's literally because you're focusing on wrong things
Gold medals dont mean anything unless you actually win and this highlights what this video showed in the end - there isn't an adequate way to judge someone's performance except for the outcome being win or loss.
There is a ton of stuff you need to account in cs and because of the nature of the game, even more in ow.
Also, you don't have to single-handedly carry games, and there is no need to be a god to win games.
Been playing since 2017, got stuck in diamond, improved, got masters, played scrims primarily, improved, got gm, lost a few to lose gm and then got a 14 losses streak. It happens. Went back to it, played alt accounts to boost skill in one-tricking or to get to masters and sell, got back on main, kept up with masters while somewhat chilling, got gm for several seasons in ow2. It's all a process and dedication and actually understanding the nuances help a lot. Most recent achievement I got is one-tricking as widow to masters. Took nearly 600 games because even in gold people try to counter you as sombra dva winston, but I got there anyway. Nothing crazy about my skill, just determination and practice. Not even VOD reviewing, although it could be very valuable at some points.
@@YTHandlesWereAMistake I’m not gonna respond to all that but I’ll say this- you don’t need to single handedly carry games to win, but you need your team to be filling their roles at the very least. That’s my point, not that it’s impossible to get to GM or whatever you think I said. In order to win a game with an incompetent team you have to go far above and beyond what your role or your elo requires you to do in OW, or at least that’s how it was back in the day and in elo hell, you got an incompetent team nearly every other game. Winning in OW requires far more team cooperation than in CS, which leads back to my point about the individualized elo system working even less. I played OW religiously since the beta until around 2018, and still have like 1200 in game hours. Cruised in diamond for a little while when I did grind comp but I’d still shit on master lobbies.
@@CoolDrifty you only need to win more than you lose. Even 50.5% winrate is enough, and that's been the same way in the days you played, idk why call that elo hell
@@YTHandlesWereAMistake I actually had a slightly positive win ratio over my playtime and yeah lol, with a 50.5% win rate it’d only take you like 1000 games to rank up, I wonder why it’s called elo hell
Tradeable death cooooullld be tracked via "crosshair-on-target time", which would start each time there is an exchange and clear LOS between two players, so no wallbangs. So in essence, if both players aim at eachother and once the first bullet is fired, it would check if the other fired back during being on target. It is a bit hard, since there is recoil patterns, or just pray and spray, but maybe it would balance itself out in the grand scheme, especially with +-1 points for kills and deaths.
Comms tracker is impossible since premades just use external options and it would be easily abusable.
this video is so well made that I just HAD to subscribe. Amazing job
Your assuming you know what the best way to play cs is and modeling your skill rating after that, which is the exact reason a system like this wouldn't work. With your system an entry fragger gain less elo then the second or third anker even though their impact on the game might be higher. A lurker in a loosing game would be highly favored since the enemies often will start running around chasing for the kill and are much easier to kill giving the lurker many more easy kills then the entry fragger even though their impact on the game is less.
Even if you find a skill rating system that works it will most likely only work at a certain ranked range since what you need to do to win a match in 500 elo is completely different from 1500 elo which is different from 3000 elo, also even if you find a way to make a system that accounts all of this the system will likely break overtime as people change how they play the game. That or you make the game stale forcing high elo players into one way of playing the game, punishing anyone that tries something new that might be better for the team but worse for their skill rating.
Also your simulation is flawed, a person with 63% win rate is not going to end up at a lower rank then they started at.
Yeah this. Frankly im not even sure if elo hell actually exists, and if it does exist it certainly isnt something that can be fixed realistically in a good way.
I mean really, if youre actually so much better but getting pulled down by shit teammates then eventually youre going to be at a level where youre just way better than everyone else and can do way more than 20% of the work of your team consistently, and because youre better than everyone else you should rank back up. If you consistently fail to rank up it sounds more like you have stopped improving or are improving at the wrong things and just want an excuse for why you aren't good enough for higher level play.
He's also assuming every skill level has the same variable/ skill ceiling and floor this video is mad cope
@@andrek6920 I saw a great comment about elo hell some time ago. "Look at any online game community, and you'll see people complaining about "the trench". The trench is a mystical point where it's impossible to keep climbing the rank ladder. Nobody can ever agree on what rank the trench resides at, or how large the trench is, but much of the community agrees that it exists and that they're stuck in it."
His simulation is straight trash. It's like, wow. You're as good at maths as you are at CS. Terrible.
Skill is normally distributed. Reaction times, intelligence, manual dexterity, the things that make up playing, all are better approximated normally than uniformly. So his data set is garbage. His rules also use random selection. Matchmaking is matchmaking. Not random. This is a classic case of selection bias: he has assumed that this is random, when matchmaking follows some rules. It's bad and dumb.
Finally, if those flashes are contributing to winning, they are included in the score. Because they WON THE GAME.
I hate the algorithm CS2 uses. It blatantly has poor bounds on the matches it makes, giving some teams a higher team elo, or a much higher elo player in their group, which is bad for match fairness. This also has almost no impact on the points given by the system: it is mostly based on match history, start losing more than winning, start losing points at a higher rate. Start winning more than losing, start gaining points at a higher rate, with a larger potential loss limit. However, the problem is nothing to do with "ELO HELL". CS2 does not have an elo hell. It has a bad and unstable matchmaking system, where you can finally end up at a rank where you're having games you enjoy, with enemies that challenge you but you can overcome, with teammates of a similar skill and knowledge, and then you lose two games and the next loss will take you down 400 elo and put you back more than 8 hours. It is not "Elo hell" it is an unstable system.
It's so unstable that I've played at 12k, I've played at 6k. I'm around a 10k player, but whenever I play, wooooaahhh, going up, wooooaaaahh going down. The system isn't designed to settle me into 10k matches I enjoy. It's designed to put me into a 12k match when I reach 10k, knock me back down, get streaky, get down to 6k, where I can't help but dominate the non-aimers down at that elo, ruining the game for them, until I'm back around that 8-10k range I do best in. I could win a 3v5 game against a group of 5-6k players. It is really that simple, they do not have the gamesense or the aim. I'm going to kill 2-3 of them every single round and win every single clutch they throw at me. Elo hell simply does not exist, you eventually fall to a level where you will dominate the enemy, and if you never do, then you are bad and should try to improve by thinking about how you do things a little more. I am not a "good" player. I am an entirely average player. It is not a flex, it's a simple fact that when you put a 50th percentile player into a game with the bottom 25% of players in the world, he is going to farm them. That's what's wrong with the system.
The chess ELO system would ensure I stayed there until I started to improve or get worse. The CS2 system instead forces me to ruin games for people who haven't even worked out that forcebuying the nova every round is a bad idea. It makes me spend more time grinding against teams that are absolutely not capable of dealing with me, only losing when I have a truly awful team or griefers (because CS2 punishers griefers with elo penalties, not game bans), which only slows me down. Instead of giving me balanced matches I enjoy, it gives me homework to get there.
Also, HLTV and Leetify already do these things, like he mentioned. And they suck. HLTV will tell you if you were killing or assisting more than dying badly. Leetify will bloat that out by applying a silly algorithm to it that says "The opening duel matters, not dying mid round matters, and kills in the endgame matter" which rewards baiting and lurking, and punishes entries who don't frag out, punishes players who take objectives, and rewards players who capitalise off their teammates to win-at the end of the day, nobody cares as long as it's a win-only an idiot cares that an entry who's distraction and calls got his team two kills and a plant got shot in the process, but individual systems and KPIs DO NOT WORK. If I go in last every time, flash for the team, get plant, finish off 2-3 players, or if I lurk unders on mirage and get a frag on the con guy when he goes chasing kills every round, positions which are low impact but high reward, leetify will say I am awesome. If I go in first and tag two people, call them out to my team, and get them the plant, leetify will say I am bad because I am "losing opening duels" and have a "poor mid round K/D" but we will win and my elo will increase.
You want to bring maths into it. Fine. How about you look up what happens when companies introduce KPI metrics to their jobs to measure individual performance. I'll wait. What usually happens is people focus on the element that increases their KPI and ignore everything else, reducing overall effeciency. You measure completed units? Ok. Nobody is doing the prep work anymore they want to sign off on the finished one. You measure time on the clock? Ok. People work slower to get more time in. You measure how quick they do a task? They'll do the simplest tasks with the least complications that take the shortest time to increase their average time under allocation. Holistic approaches work, that doesn't. If you want to track flashes for rank xp, fine. People will throw more useless flashes. You want to track smokes. Fine. People will throw more useless smokes.
The judge of how good those flashes or smokes are is WHETHER OR NOT THEY MAKE YOUR TEAM WIN. The holistic measure is important. Online matchmaking needs improving, but it needs it in the algorithms it uses to design a suitable match-these matches need closer matchups, and it needs it in the rewards-the games should give close matches and small elo gains and losses. Drastic gains and losses should only be used in the placement algorithm used in the games that are played to unlock a matchmaking rank.
Great video! I really enjoyed it and subscribed.
hope to see more from you in the future!
The subtle humor in this video is crazy
Imagine that points are calculated based on kills. All players will try to get as many kills as possible and will forget the main purpose of playing a tactical shooter.
Got 1.5k games just in premier alone. You're out here doing Gods work.
Also got a 35 week streak just because I don't want to let it go. Been going since they released it. Man we need new content and an updated ranking system would be cool.
0:30 epilepsy warning
Damn can something as basic as that cause a seizure? That's rough.
Feels like theres only 1 way you know there should be a warning here :(
Some people get seizures out of nowhere or from very small things @@Gr00t
Finally someone else that understands my deep and unending hatred of glicko 2
I think something is lost when you only play against players of the same skill level because learning comes from watching habits of a better player
This is an insanely well researched written and produced TH-cam video good shit man
awesome video man. really enjoyed it
The ranking system on premier is actually not elo based system. If you stack with higher ranks, you'll be getting like +365/-105 on average and you can climb with a ridiculously low winrate like 33% (when average enemy's team rating vs your average team's rating are the same). On premier it's actually way more important the ranks you stack with, for your rank, than how good you play/how much value over time you give to the team
Great work. Thanks for the effort you put into this video.
I know that I'm only at 10:10 , but I feel like I would analyze this problem with a Montecarlo simulation that you can do in python, with a simple RNN to create the function model to produce the desired ranking function.
A rating for personal perfomance would be great because we have so many parameters in cs to show your performance in a match: kdr, adr, nate damage, entry kill, trading and many more. the worst part about only counting the team performance is, you give trolls and afks so much power about your elo rating. no matter how good you play, if one guy decides to be afk or to troll, the match is almost lost already, and you have nothing to fight for because your performance is not honored. This current elo system is so teriible, i stoped caring and just play
great video, very interesting. Sad it ended
I agree with everything except the point system.
I buy SMGs exclusively because it can be so disruptive to the enemy economy when they are on CT side and it can be a big boon to us when we are on CT side.
Dictating the wider teamplay-style to some degree is okay but trying to enforce individual play styles by punishing players who do interesting things with the game's meta is toxic.
Your system rewards me for out-performing the enemy with inferior weapons but punishes me before I even leave spawn, I understand that you acknowledge that it might have flaws but to even suggest this means the system is too married to the current meta.
Having 1 Mac-10 player on T side means CTs spend 350 extra each to not get clowned on by me, having an MP9 on CT side means I can buy a gun and nade every round regardless and farm money, when we win on CT, I get to drop a free gun every 3rd round (an still buy, assuming an average game).
I been saying this for like ever, win loss ranking systems dont work. great vid
Great video, high quality!
1400 faceit games, 3000 mm and 7400 hours and this video spoke to me. Have been consistently level 8 for the last 3000 hours and I swear it’s not me 😂
The true cause of ELO hell is that ELO exists in the first place. It's a tool designed to create false engagement with a game, an interest in the status of number go up, not of actually becoming skilled at the game.
awesome video. This will definitely take off.
Now I feel bad I forgot to turn off adblock. Commenting helps though right????
Elo rating works well for chess. It also works decently well for fighting games.
But for team games, no matter which, there isn't just your rating at play, but also the rating of your teammates. And how well you act as a team, something completely independent from your individual ratings.
And a group of worse players with good coordination and teamplay can easily win against a group of better players that all play alone. Their individual rating might be worse, but their team rating is higher.
Brilliant video for a channel of this side
this video deserves a lot more in views and attention
There is no fix for this problem. Teamshooters cant be rated with a elosystem because mathematically you are responsible for 20% of the Teams performance, if you consider 5v5. So you could say "Lets rate the Player upon his own performance like K/D or ADR for example". This would result in people baiting 24/7, wich isnt a good idea. So yeah... there could be a fix in making a personal ID mandatory for playing FaceIt for example, wich makes every Player personally attached to his Ingame-Persona. Meaning that behaviour and performance are important, wich would result in way better Communications and Teamplay. Dont forget were talking Teamshooters here, individual performance is important but communications and teamplay are crucial to climb elo, thats why mostly 2-3 Player lobbys grind effectively in Faceit. The key is locking ID to a Players account so that Cheaters cant continue after getting banned and vocally violent and crazy people would face serious repercussions too. Thats the only logical and doable solution if youre not playing a 1on1 Game like in the old days. If 1on1 games would be a thing we could implement the strict elo win/loose rules again.
But your rating will also be heavily influenced by flash assists, trades and potentially successful entries. The whole point of the video is that the current elo system doesn't give any weight to individual performance in any way, which results in huge disparities of deserved ELO and actual ELO.
No system is perfect, but a system that rewards different aspects of individual performance is more accurate, as per video's conclusion
@@Dapplication Yeah, totally agree with you. I was just venturing off in a deeper problem with anonymous Players in a team-based game. Im pretty sure that we should work on the root of problems before considering tweaking individual elo, as a lot could be instantly fixed by making people accountable due to permanent ID lock on accounts. Especially in a Teamgame.
Video literally addressed your baiting point @45:15
@@InigoPhentoya Guess he did not bother to watch it before writing his opinion
@@InigoPhentoyana its fine i just have another opinion. My point is the ID lock, everything else is tinkering on symptoms
What a Banger M0si
Liked and subscribed. Great video man.
I dont even think valve, riot, and literally every comp game uses ELO system.
The system they are using was made to keep you playing in hopes of winning and climbing.
There is no fing way they are using ELO system, it was not created for comp games.
So basically you need to be insanely good at the game to be able to climb or have duo or trio to win more games.
Or grind ur life away to reach high-elo
Faceit uses ELO system, Valve uses some abomination that gives out +300/-100 games and then starts giving +100/-300 games for no logical reason
@@fica1137 Same for Valorant and league of legends. Am talking about big comp games, They all do the same thing. Nothing is fair about their rank system.