Bad ratings for imbalanced games
alster
Join Date: 2003-08-06 Member: 19124Members
Join Date: 2003-08-06 Member: 19124Members
If you have played games where your team gets dominated by the other team then you're most likely to rate the match played with few stars. The reason for this is imbalanced teams. Even with balance mods it's hard to make teams balanced. Newer players may quit playing after a few stacked matches. I know I wanted to.
A good way to fix this:
Server side setting for handicap on.
Every 5 minutes check if a team has double the pres and tres of the other team.
If so then award everyone on the losing team +50 pres and tres.
The losing team will probably still lose, but they will have much more fun playing and then they will rate the game much higher.
Tdlr - "TEAM HANDICAP AWARD +50 RES"
A good way to fix this:
Server side setting for handicap on.
Every 5 minutes check if a team has double the pres and tres of the other team.
If so then award everyone on the losing team +50 pres and tres.
The losing team will probably still lose, but they will have much more fun playing and then they will rate the game much higher.
Tdlr - "TEAM HANDICAP AWARD +50 RES"
Comments
Second of all, if I were to do a handicap the one you proposed is really silly and random.
Actually, a "comeback mechanic" has actually been implemented a few weeks ago, where teams get more pres if they have less RTs than the enemy team (...or something like that).
Although apparently that didn't have a huge impact on this, does anyone have numbers? (@Nordic ?)
That did not go in according to Remi. According to the testing on Tactical Freedom, and it made losing more fun likethe OP wants. http://forums.unknownworlds.com/discussion/comment/2284777/#Comment_2284777
Also worth noting that it did not impact round length or balance that much.. it just felt better. Far less demotivating rounds or reason to concede.
And it was still flawed / biased towards Marines imho. (Tres is not equally powerful between the teams)
I'd like to see a Pres (alien) / Tres (marine) version of it.
One thing is for sure.. It opened my eyes to the unnecessary pain and frustration that comes with restricted options in a team based competitive game.
There is so much depth still with strategy and positioning that not having the funds to execute them seems unnecessarily restrictive, and a left over from RTS fitting into FPS.
We're considering a mechanic where 5 minutes into the game, it just randomly declares a team the winner with a 50-50 chance. That should make everything perfectly 50% balanced.
( Obligatory: Yes this is sarcasm... because some people just can't detect it very well...
Currently Steam doesn't even have this game as a high priority download - it's the end of times, doom, doom doooooom!
That would make aliens overpowered with their 1 min into the game game-ending baserushes!
I suggest making the timeperiod shorter, maybe 20 seconds or so.
Maybe we will see it one day in NS3!
Second: To avoid stacking, the balance-system needs a remake.
beside the classic "shuffle" function the normal joining-phase should count the skilllevel with every player entered a team.
So if marines have 500 hive and alien 420 a stronger player can only join aliens. Then with marines 500 hive and aliens 560 a stronger player can now only join rines.
So stacking with 1800 against 1200 or something can be avoided with this.
some examples of why doing a Forced team balance for "vets" is a bit unfair:
- Do you like playing 5 games in a roll on the same side(aliens/marine)?
- Everyone on your team is "new" due to your "high" hive score, you pretty much stuck in situation where you have to tell everyone what to do. As a new player if you see 1 guy trying to micromanage the whole team, he looks like a douche apparently(in other words; someone who thinks he knows everything). Quite a frustrating experience.
I'm all okay with people learning and stuff. I mean everyone starts off as a rookie, even I still make stupid mistakes now. But the point is, the rookies need to able to listen and hear experienced players out instead of just ignoring and asking them to stfu. Sometimes it isn't their fault due to language barrier(they don't speak English).
I'm not sure how to fix this tbh, it's just the current situation it is now.
At least a few weeks/months ago, there were a lot of "vets"(in my community) around and you can notice those vets giving some "tips" to the rookies(at the time we can see their nick highlighted green) like "as a lerk you need to keep moving"....etc. Now it just like 5? i'm not sure atm. They are outnumbered now
Anyway about the hive score; I reckon to be more "accurate" instead of just win/lose ratio, I think player dmg and structure dmg should be taken in account for field players. Commanders should get hive score based on win/lose ratio only.
Can confirm, have been sitting around 2800 -3200ish range for elo, alien = carry lerk or carry fade or else my team gets rolled, same on marines. If i command my team cant even hold naturals 90% of the time.
Gave up ordering players around when players either ignore me or insult me for attempting to.
Stopped bothering to try to teach players how to play when they 1 don't listen or 2 insist they know how to play ns2 better than i do and that i'm an idiot.
If i happen to bother to get on and play, its purely listening to music killing time before work or something, don't care if its against greens or vets, can't even tell who is who.
If i want to play with friends i never can, we get shuffled apart 99% of the time, can never pick which side i want to play either.
I get players want good games and balance but in reality ns2 pubs are really similar to csgo casuals, except i can never play with friends, and i get forced on a team i don't want to play quite often, if i ever want to play a game with friends now a days, ns2 is not even on the list of games to play but i guess player retention has increased so why would uwe care
That does not mean hive doesn't have problems.
The biggest reason that makes balance hard to get in ns2 is our small playerbase combined with a high skill ceiling. There are not enough players for low skill and high skill players to have servers of their own. We all have to play together. When you have the highest skilled players playing with the lowest skilled players, it does not make for a good experience. The only way to fix this is to make the game more popular, which is a chicken and egg problem.
The shuffle system is pretty good. The problem is the small playerbase. Take a look at this thread and see if you can hand balance a hypothetical team better than shuffle.
http://forums.unknownworlds.com/discussion/140614/forum-game-who-can-hand-balance-better-than-shuffle/p1
This already exists. It is a matter of if server operators enable it or not.
The real issue is the time it takes to be accurate... which is covered in Nordic's second bullet point, and is why I brought that issue up in that thread.
Requiring hundreds and hundreds of hours to be accurate is too tall of an order to ensure rookies get a chance to play balanced games, which can directly contribute to a smaller playercount and thus the negative loop continues.
What in game stats do you think are more representative of skill? Do you think a players KDR, or kill rate is more representative of skill? Do you think that a players Score/minute is more representative of a players skill? What in game metrics do you think matter?
Hive level. Ie: Cumulative score.
Anyway, just what are these stars going to do? Are you going to watch each rated game and see why a particular res node died and one team lost? Even with a massively Google sized algorithm you are not actually going to get any relevant data from stars vs games that you didn't already have available from 100,000 forum posts.
There are so few players that any 'statistics' won't even meet the 1200 sample size required to make inferences about your population.
Also it makes it additionally difficult that we can't segregate players into a skill-range group (apart from rookie only).
Player-Score is a good indicator. At the moment the hive rewards the team. This includes the topscorer but also the one guy with 2% acc and a kdr of 1:27.
Imo it is not fair that the topscorer of the losing side is losing hive skill while the bad players on the winning side are getting it.
With a high hiveskill you always have to carry your teammates, but it is not so easy to command rookies to a win. It is very painful to play a fantastatic round and be awarded with a big hive drop because the rest of the team does not know how to carry a weapon......
And kdr ist a good indicator too. Only with kdr it is possible to reveal smurfs.
Thanks for correcting me, I was not aware of this.
Also I swear I already made this post.
Just think about one extreme scenario to understand my point:
If you have a skill score over 9000 and you lose a round with a kdr of 30:3 against completely new players, shouldn't you lose skill points? You were clearly better than the rest of both teams players, but you clearly don't deserve a skill of 9000... why should you win more points? If you keep winning points just because you are better than the rest of players in your server you could reach infinite skill points...
On the other hand, the worst player in the opposite team with 0 skill score and a kdr of 1:50 deserve to win skill points, because it is his first game and he had a skill score of 0, and he is clearly better than a 0
If player score is all that matters... why should I ever lane block when my whole team is mindlessly pushing to nanogrid in Veil? I will win the game to my team by just killing 3 skulks trying a base rush and yet I will not be one of the top score players of my team
If kdr is all that matters... why should I ever resbite? I will go fade even if my team has only fades dancing around and achieving nothing
And kdr sometimes reveal smurfs, but sometimes (in some rare occasions) just reveal marines with good aim playing against mediocre aliens. And more often reveal terrible aliens doing nothing in base until they can go onos with upgrades
Well, I think Hiveskill isn't that bad, except for the issue that it takes far to long to adjust to the point when it will be a representative number.. which will mean there will be literally hundreds of games with unfair teams, especially for people with past experience which will help them improving very, fast, for instance.. due to ns1.
And the thing I think you dont understand is that it needs more to win than just killing everything in your way.. I have currently only 100hours on my record but I've seen several games with people with over 3000 Hiveskill but loosing because they only care about accuracy and kdr but never build anything even if it's crucial at that time.
So, If you give these arrogant "carries" who dont look where they are really needed a higher hivescore then you'll actually loose more often when playing with them.. it's still a team game.. but lot's of ppl seem to forget that part.
They see it giving and taking points, draw conclusions about how it works, how it should work, and then you get endless, endless threads of people posting nonsense and drowning out the discussion that REALLY contributes to improving the hive system.
If I ran these forums as a dictatorship I'd ban all talk of hive score except from like, 3-4 people who dont shitpost the same old crap every time the subject comes up.
I have had very little sleep so the fact that people are STILL posting "kdr should affect hive score" is fucking baffling to me right now when there has been literally hundreds of forum pages devoted to debunking stuff like this and providing actual productive discussion on the matter.
I have no idea how to accurately measures a NS2 players skill. But I know such a system would need an extremely complex algorithm. NS2 got 2 reasons for an ELO system to never work: 1. It is a team game, but not only that, it is also asymmetrical and and top of that with a commander on every team. You would need an own skill rank for every side and for comm / not comm, also you would need a way to see a players impact on a game in relation to the commanders abilites and how the commanders skill affected both sides players and the outcome of the game. 2. NS2 as a game is way too complex for ELO.
I mean, of course, in the end ELO can be very accurate. But only if every single player in NS2 played 10000 hours minimum.
You are saying that a players Score per minute and KDR are good indicators of skill. That makes me wonder, how well do they correlate to hive skill?
If you don't know what a correlation coefficient is, read this web page. A correlation coefficients measure the strength of association between two variables. Basically it is a measure of how related two variables are. Anything between 0 and 0.5 is considered a weak positive correlation, and anything over .75 is considered a strong positive correlation.
In this graph I am showing how well hive skill correlates to a players win rate, kill rate, and score/minute at different playtimes. Do note that hours recorded in hive do differ from steam hours because 1) Hive wasn't around at the beginning of NS2, and 2) Hive only records from match start to match end while steam even tracks from exe open to exe close.
You will notice that players with less than 100 hours recorded in steam do not have hive skills that correlate very well with the stated skill metrics. Hive is kind of useless at this stage. By the time a player is between 100-200 hours recorded, the correlations are all at least .5 meaning that they are related at least. Hive is getting better at showing a players true skill value. For players with 200 hours or more, hive relates to the stated skill metrics with a correlation of no less than .6.
All this really shows us is that if you think KDR and Score/Minute are good indicators of skill, as you do, then hive skill is related to those metrics. It may not ever be a strong correlation, but it is far from being a weak correlation. This alone proves that hive as a measure of skill is at least not useless. Hive does measure skill similarly to KDR and Score/minute. It just takes awhile to do it, but I already stated there is a proposed fix for that.
What you do not understand is that a players KDR and Score/minute effect how often a player wins. By measuring win/loss you are also measuring KDR and Score/minute indirectly.
If the above graph was not enough to prove that hive is not useless and is at least a somewhat decent measure of skill, check out this table in the spoiler. In that table I show you the same thing I showed you above in a visual manner. I directly compare how well hive ranks players compared to win/loss, KDR, and Score/minute. I took 205 randomly selection players from hive. I ranked them from 1 to 205 in win/loss, KDR, and Score/minute. I then color coded all the ranks from 1 to 205 so that you can visually see when they differ and by how much. I also averaged the ranks of win/loss, KDR, and Score/minute to form an average rank that reflects win/loss, KDR, and Score/minute.
Without looking at the numbers in detail, you can see that hive does a pretty good job ranking players 1 to 205 in terms of skill compared to win/loss, KDR, and Score/minute. We all know there are players that have hive skills much higher or much lower than they should have, and you can visually see them in this table. You can see how common it is that a players hive skill does not seem to match how well they perform in win/loss, KDR, and Score/minute.
If you think 205 players is not enough to compare with, here is a table with 1025 players. In this table it is even more visually apparent how often hive skill does not seem to match how well they perform in win/loss, KDR, and Score/minute.
Are you starting to notice that hive is actually a fairly good measure of a players skill? I have now shown you that it is both mathematically and visually. Hive is far from useless and is does measure a players skill fairly well.
But I am not done yet showing this to you. Ironhorse touched on it slightly. Microsoft created a skill system called TrueSkill. It is heralded as being the industry best. You can read a little bit about it here.
The TrueSkill system adjusts skill individually for each player while hive adjust the whole team by the same amount. TrueSkill adjusts based on both skill and certainty while hive adjusts based only on skill. In theory, it should find your intrinsic skill value faster than hive. Doesn't that sound better to you @Rammler and @The_Welsh_Wizard? Rammler did mention that it is not fair that hive rewards the team and not the player.
You can see the TrueSkill MMR data publicly here.
The server Tactical Freedom has been experimenting with TrueSkill. This got me curious. Which is more accurate? Hive has a lot more data recorded but TrueSkill is statistically significant after just 90 games. The best way I can think of to compare hive and MMR for accuracy is based on how those players compare on the in game skill metrics. Tactical freedom has only 23 players so far who have over 90 games. Their MMR is statistically significant.
I ranked these players 1 to 23 in all of the following metrics: TrueSkill MMR, Hive skill, Winrate from hive data, Killrate from hive data, Score/Minute from hive data, KDR from hive data, KDR from Tactical Freedom data, Takedown ratio from Tactical Freedom data, winrate from Tactical Freedom data, and average marine accuracy from Tactical Freedom data.
The idea was to compare how well TrueSkill MMR and Hive skill ranked players in comparison to those gameplay metrics. In theory, a players rank from 1 to 23 should remain similar among all metrics. This gives a good impression of how the players compare to one another, but it does not measure skill.
I then decided to average the rankings to find a gameplay metric rank. I did this in 3 ways. I made an average rank from all the gameplay metrics, one average rank from just the hive data, and another average rank from just the Tactical Freedom data. I did this because of the following hypothesis. Hive should be more reflective to hive data and TrueSkill MMR should be more reflective of Tactical Freedom data because TrueSkill MMR is based on data only collected at Tactical Freedom.
I took all ranks 1 to 23 and color coded them so that it would be easy to visually see how well the different rankings. This gives a good impression of how the players compare to one another, but it does not measure skill. When you compare those to the hive and MMR ranks you can get an idea of how accurate each one is based on how well it compares players together. This is not a perfect measure because we don't know what environments these players played in. A pub is different than a gather. Comp players would have different metrics than pub players.
The result was this. http://i.imgur.com/alR2A6y.png
Edit: I am told I did not explain some of the terms in that above table. All metric rank is the average rank of all the individual metrics. H metric rank is the average rank of all the metrics I got from hive. TF metric rank is the average rank of all the metrics I got from TF. Individual metrics are hive KDR, TF KDR, TF accuracy, TF win/loss, Hive Win/loss. Tf metrics are from Tactical Freedom. H metrics are from Hive.
I thought this was really interesting that all the metrics in some way show skill, and for the most part line up fairly well. Comparing hive skill to the TrueSkill MMR was a bit challenging. To me it looked like hive a more accurate representation of skill, but it is really hard to tell. I wanted hard numbers.
The only way I think I could quantify the difference was my finding the cumulative difference between ranks. So I took the sum of the absolute differences between the rankings. A hive rank of 1 was 5 different than a all metrics rank of 6.
This is far from a perfect analysis but it gives us an idea how well hive measures skill compared to TrueSkill. In the end, I have concluded that hive is a more accurate representation of skill than the TrueSkill MMR. This could be because hive has a lot more data to work with because it has been collecting for years now over many servers, while Tactical Freedom has only been collecting since April 1st on a single server.
You are not supposed to care what your hive skill number is. Who cares if it drops. It is not how much money you have in your bank account. Hive is not awarding you anything ever. This is exactly why I wish that players hive skills were hidden.
Hive is a long term measure of your skill. Don't worry about it on a match-to-match basis. Hive is a system where only your wins and defeats matter in the context of whom you played against. If you improve in the game, you are more valuable to your team and you will start winning more games on average, and the score will adjust upwards. If it settles around a number, that is most probably your intrinsic skill level. Hive might deviate from your "true" skill value due to bad/good streaks but it will rebound in the long run and keep oscillating around that level.
There is not fairness about it. It does not matter if you drop a little bit or a lot in the short term.
This is a legitimate problem with hive. I agree. Not much can be done about this as far as I know. Having a larger playerbase would help a lot though.
I agree with nordic that Hive score should be hidden maybe even from Hive profile as well.