This seems to be a pretty common complaint, but I'm not sure what the right fix is. The root problem seems to be that there's one player on the server who is just way way better than everyone else. What should balanced teams with that player look like? If there isn't a counterbalancing player who is equivalently skilled on the other team, then it's really hard to compose teams that will make for a satisfying game. Expecting the amazing player to carry a poor team seems reasonable to me, but maybe there's an alternative?
Lets say your server is as follows.
skill 3000 : 1 player
skill 1000: 8 players
skill 0 : 5 players
How would you allocate these players?
If doing it by eye, as in a captin's type pick, it would go:
Team 1:
skill 3000, skill 1000 x 3, skill 0 x 3 = total team skill 6000/7 ~857
Team 2:
skill 1000 x 5, skill 0 x 2 = total skill 5000 / 7 ~714
These numbers won't change as we can move a skill 1000 to either Team 1 or Team 2. So it looks skewered on either side. But still your best chance to have a better game. I have had reasonable games when we do Captain picks and just go by the K/D of the last game. May be you can shuffle by score of last game? Or if that isn't possible. go for shuffle by Hive score, but instead of trying to get an fairly similar skill rating on both sides, rank the players from highest to lowest, then randomly select the first player (the top player say, then make the next player on the list to the other team). That way, you at least split up the top players, and don't have Rookies to skewer the vote. So if the shuffle didn't work (stomp game- for what ever reason - eg. the top player is only a good marine), you can elect to play the match with same teams but switch sides (not sure if this is in Captain's mod or such already).
The point is you need to be able to shuffle vote easily, and may be choose from a few options:
1. shuffle by last game K/D
2. shuffle by hive rank split pick
3. shuffle by hive rank team average
4. random shuffle
Hope that would get some ideas together to work a better shuffle system.
@moultano what you said is right, if i get a good comm and only on alien side i can carry the team till mid game. When some good fade should take over (usually when shootguns shows up, lerk cannot be a carrying unit and agressive). That's why all my alien games don't last more then 15 minutes. But as marine my games are turtling for 25 min or i just get stomped by alien team. Like @Cannon_FodderAUS said aliens will just avoide me unless i start to kill harvesters then im getting raped by 4-5 aliens trying to save the RT. Still i think that skill level should be distributing players on weighted average. So even with one player with skill 2000-3000 will not have games with only rookies, its just not fun.
Just my op, one of the reasons Hive works as it does is to get around this perspective. So much more contributes to a side winning than K/D. Building, healing, welding, teamwork, comm strats etc. But as you said, it's a choice, though servers using this may get worse games. And on that, I think one Force Even method is best, as it remains consistent no matter what server you join.
So if the shuffle didn't work (stomp game- for what ever reason - eg. the top player is only a good marine)
This I think is the crux of the problem, thanks for raising it.
A good marine player is not the same as a good alien player and vice versa. We might all love playing both sides, but usually are stronger at one over the other. There's only one Hive score right now for players, when there should be two -- one for marines, one for aliens. If this is added, more balanced teams can be achieved: the best player for a side is added to that side as each team is filled. There are just two caveats to implement with this:
Which side gets filled first each game is decided randomly by FE -- otherwise you'll always end up with the same highest marine player on marine etc (I'm pretty FE does this atm anyway).
To ensure players don't get allocated to their 'best' team all the time -- we all love variety after-all -- either join a team first (as it works now) or perhaps every now and then start filling with the 2nd or 3rd best on a server and then go back to 1st to top down (similar to the above, this switches the order in which teams are filled, and in turn which 'score', marine or alien, is used to allocate you)
All of the above may also be complete ass. It's late and I'm tired.
Just my op, one of the reasons Hive works as it does is to get around this perspective. So much more contributes to a side winning than K/D. Building, healing, welding, teamwork, comm strats etc. But as you said, it's a choice, though servers using this may get worse games. And on that, I think one Force Even method is best, as it remains consistent no matter what server you join.
@Martigen Good point about K/D, how about score from last game? The reason why I think this can work, is because it is looking at how you performed against the current cohort of players you have just played against. Hive is very good for tracking global stats (ie. how you are as a player against the rest of the population (over time, you will play against many many opponents - that is the assumption). Where as in the case of a shuffle based on last game, it will be more accurate for now (ie. the next game), but could be inconsistent if you take a global approach.
Yes, that doesn't take in account of if you were alien or marines last round, nor your skill playing on either side. However, bear in mind NS2 is a very complex game (as you have mentioned), and it depends on so many factors, with FE just trying, TRYING to mitigate one of these factors.
So if the shuffle didn't work (stomp game- for what ever reason - eg. the top player is only a good marine)
This I think is the crux of the problem, thanks for raising it.
A good marine player is not the same as a good alien player and vice versa. We might all love playing both sides, but usually are stronger at one over the other. There's only one Hive score right now for players, when there should be two -- one for marines, one for aliens. If this is added, more balanced teams can be achieved: the best player for a side is added to that side as each team is filled. There are just two caveats to implement with this:
Which side gets filled first each game is decided randomly by FE -- otherwise you'll always end up with the same highest marine player on marine etc (I'm pretty FE does this atm anyway).
To ensure players don't get allocated to their 'best' team all the time -- we all love variety after-all -- either join a team first (as it works now) or perhaps every now and then start filling with the 2nd or 3rd best on a server and then go back to 1st to top down (similar to the above, this switches the order in which teams are filled, and in turn which 'score', marine or alien, is used to allocate you)
All of the above may also be complete ass. It's late and I'm tired.
See my switch sides argument on this, so you get to play both sides as a team. Though I hate to restrict ppl on how they want to play the game, eg. some people just plain hate going marines as they prefer the alien life style. There in lays another problem of people F4 to go to prefer side.
It is getting late, and I don't think there is a perfect solution, if I could work out an algorithm that can predict a fair game in NS2, I would be rich predicting sports games. Not gonna happen.
I completely agree that the same skill one player have for marines, they wont have for aliens and vice versa. I am much better at aliens than marines. There are people I play who are deadly marines but can't skulk. It just is what it is in that way.
There should be an option to choose your preferred side and FE could consider this when making teams. Last time I played public it put me to gayliens about 8-9/10 games.
edit: Also, what does the skill number say? What is the average skill? How does one know if he's considered being good or bad? Hive doesn't say anything except showing some "top" players.
There should be an option to choose your preferred side and FE could consider this when making teams. Last time I played public it put me to gayliens about 8-9/10 games.
edit: Also, what does the skill number say? What is the average skill? How does one know if he's considered being good or bad? Hive doesn't say anything except showing some "top" players.
There already is an ad-hoc option to choose your preferred team; just join a team before the vote passes. A minimum amount of switches will be made so your chances of staying on that team are good, especially if there are still players in the ready room.
And the skill number is only relevant with reference to other players, there is no absolute level of 'good' or 'bad' in a vacuum.
@moultano
Some thoughts on the current implementation:
* I've noticed that FET moves players to make the teams having the exact same skill rating. If you allow for a small difference in skill between teams more players should be able to stay on their chosen team.
* When there is an uneven amount of players on the server the team with one more player gets a lower average skill rating to compensate. The FET assumes that the round will be played with uneven teams but that is rarely the case since servers with one free player slot usually fills up in no time.
* It takes too long for players that start with a skill rating of 0 get their rating up.
In general I think FET works really well and people who say it sucks probably haven't thought about the how complex it is to balance teams in NS2. We have a lot less bad rounds since FET was implemented.
@moultano
Some thoughts on the current implementation:
* I've noticed that FET moves players to make the teams having the exact same skill rating. If you allow for a small difference in skill between teams more players should be able to stay on their chosen team.
Interesting idea!
If viable, variance should be limited to at maximum 50-70 pts. It seems to me that an average skill difference of anything over 100 is pretty significant.
dePARAJoin Date: 2011-04-29Member: 96321Members, Squad Five Blue
edited February 2015
With NS2+ we have a good instrument to track the accuracy.
Did someone ever tried to balance based of average accuarcy of the last 3 rounds?
The players with the highest skill-score are normally the ones with the highest accuracy anyway.
If you compare 2 players with 2500 skill score there can be a HUGE difference in skill cause the system dont care from where these points come from (2500 earned on wooza are not the same like 2500 earned in 6v6 gathers)
Sometimes the skill-system is "thinking":
"Wow, thats an amazing player"
But the truth can be:
"Looks like this is one of these days where im hitting nothing"
Maybe a combination of "hive skill" and accuracy represents the real skill more cause it includes the history of the player (hive) and the current ability to hit things.
Something like:
"hive-skill" (base score) * accuracy/10 = shuffle score
Btw. you dont need skill to build structures. Its important for sure but dealing damage is a combination of aim and positioning and thats what i call skill.
There are people with 20% accuracy who kill more than anybody else on the server. Accuracy is not a good indicator. Aliens have higher accuracy as marines, if you shoot to suppress the enemy or to search for cloaked aliens etc. you have a lower acc., but you help the team as well. I think it is an intersting thing to know, but base teams on it....dunno, I would not integrate that factor a lot in the calculation.
There are people with 20% accuracy who kill more than anybody else on the server. Accuracy is not a good indicator. Aliens have higher accuracy as marines, if you shoot to suppress the enemy or to search for cloaked aliens etc. you have a lower acc., but you help the team as well. I think it is an intersting thing to know, but base teams on it....dunno, I would not integrate that factor a lot in the calculation.
If it were done that way you would have to differentiate marine vs alien skills somehow as previously suggested. I like that idea simply because I am a far worse marine than I am an alien.
Your example of people checking for cloaked skulks I don't fine worthwhile because every marine should be doing stuff like that. I think that would balance out.
I do see potential for abuse though with someone artificially lowering their accuracy by shooting a wall for a long period of time to stack, but I find it unlikely that people would be so low to do that.
dePARAJoin Date: 2011-04-29Member: 96321Members, Squad Five Blue
The base idea behind my previous post was: combine different systems for an better result.
We have:
NS2stats
Hive stats
Accuracy and damage stats @ roundend
Different shuffle modes from shine (pure random, elo, kdr, score)
Take the shine elo vote for example.
It is based on my ideas and i modded it a bit on the HBZ server.
How does it work?
I splits the 8 players with the highest ns2stats elo score (4 on the hbz server) and it shuffles the other players based on the average kdr of the last 3 rounds.
Why elo and kdr?
You have 2 carrys per team in most games, thats why the elo split is so important and players who are able to kill the enemy have an bigger impact on the game.
It worked pretty well, better than hive stats in my opinion.
Lets find a good algorithm to combine some of these systems/data for an better result.
This seems to be a pretty common complaint, but I'm not sure what the right fix is. The root problem seems to be that there's one player on the server who is just way way better than everyone else. What should balanced teams with that player look like? If there isn't a counterbalancing player who is equivalently skilled on the other team, then it's really hard to compose teams that will make for a satisfying game. Expecting the amazing player to carry a poor team seems reasonable to me, but maybe there's an alternative?
Lets say your server is as follows.
skill 3000 : 1 player
skill 1000: 8 players
skill 0 : 5 players
How would you allocate these players?
If doing it by eye, as in a captin's type pick, it would go:
Team 1:
skill 3000, skill 1000 x 3, skill 0 x 3 = total team skill 6000/7 ~857
Team 2:
skill 1000 x 5, skill 0 x 2 = total skill 5000 / 7 ~714
These numbers won't change as we can move a skill 1000 to either Team 1 or Team 2. So it looks skewered on either side. But still your best chance to have a better game. I have had reasonable games when we do Captain picks and just go by the K/D of the last game. May be you can shuffle by score of last game? Or if that isn't possible. go for shuffle by Hive score, but instead of trying to get an fairly similar skill rating on both sides, rank the players from highest to lowest, then randomly select the first player (the top player say, then make the next player on the list to the other team). That way, you at least split up the top players, and don't have Rookies to skewer the vote. So if the shuffle didn't work (stomp game- for what ever reason - eg. the top player is only a good marine), you can elect to play the match with same teams but switch sides (not sure if this is in Captain's mod or such already).
The point is you need to be able to shuffle vote easily, and may be choose from a few options:
1. shuffle by last game K/D
2. shuffle by hive rank split pick
3. shuffle by hive rank team average
4. random shuffle
Hope that would get some ideas together to work a better shuffle system.
Given that breakdown, I would go for something like
3000x1
1000x2
0x5
vs
1000x7
I still think that this would be in favor of the team with the 3000 skill player but the 1000s team might have a small chance. Replace one of the 1000s on the 3000s' team with a 0 and it should be a close game.
I feel that the skill raating is more exponential than linear. A player with double my skill rating has WAY more than double my performance in game. You are basically looking at div 1 pros vs a pubber scrub, 12 of me wouldn't have a chance.
Lets say your server is as follows.
skill 3000 : 1 player
skill 1000: 8 players
skill 0 : 5 players
That is 7 players per team. 1 highly skilled player, 8 moderately skilled players, and 5 rookies. Rookies at best follow orders but struggle to lose most engagements, at worst a rookie can be so bad they might as well not even be on your team.
Option 1:
Team A: 3000, 1000, 0, 0, 0, 0, 0 = 4000 total team score
Team B: 1000, 1000, 1000, 1000,1000, 1000, 1000 = 6000 total team score
This option would be bad. There are too many rookies on team A vs the moderately skilled players of team B. This would create a bad experience. The current hive system probably would not balance this way because the total team scores could be more even. On multiple levels it is obvious that team B would win here.
Option 2:
Team A: 3000, 1000, 1000, 0, 0, 0, 0 = 5000 total team score
Team B: 1000, 1000, 1000, 1000,1000, 1000, 0 = 6000 total team score
This is how I feel hive would balance the teams now. Without math I would predict a loss for team A because they have more rookies. Mathematically this is one option to get the most even possible total team scores. It fails though because rookies at worst might as well not even be on your team.
Option 3:
Team A: 3000, 1000, 1000, 1000, 0, 0, 0 = 6000 total team score
Team B: 1000, 1000, 1000, 1000,1000, 0, 0 = 5000 total team score
Although this option is as mathematically balanced as option 2 I think it is more balanced in reality. The rookies are better spread out making for a more positive game. At worst, rookies can be so bad they as well not even be on your team. This essentially makes a 4v5 game with one highly skilled player. Then the question becomes, can the highly skilled player make up for having less members on your team? Saying that I think the game could go either way, but team A would have the advantage.
Rookies at best follow orders but struggle to lose most engagements, at worst a rookie can be so bad they might as well not even be on your team.
Nordic makes a good point here and I think -- any other long-term changes aside, which may take more time to develop -- @moultano could improve the algorithm quickly and easily by breaking down the Force Even allocation into two rounds. First define a rookie, which I'd say we should base on hive score but since we have a mix of rookies that started at 0 and started at 1000, may be better to base on hive hours.
So, for eg, rookie= 35 hrs or less.
Force Even balances team excluding rookies
Force Even then allocates the pool of rookies to each team*
*maybe ranked by hive score, putting one player on one team, the next best player on the other team etc, but not using the score itself. Since, at the moment, it seems FE will use the lower-scored players to 'fill in the gaps' to make the total average scores balance, and this is why we sometimes see the lion's share of rookies being lumped onto one team (at least, I presume that's why this happens).
TL;DR -- don't include rookie scores as part of the balancing, and instead allocate rookies as fairly as possible after regular players are balanced.
Cannon_FodderAUSBrisbane, AUJoin Date: 2013-06-23Member: 185664Members, Squad Five Blue, Squad Five Silver, Reinforced - Shadow
edited February 2015
@Martigen I think excluding rookies are a good idea. I am friendly to them and try to help them learn, but lets face it, until rookies understand the game better, they are about as useful as a potato on the field. So balance method you suggest would be (sorry the way you wrote it, I had to read it twice before I think I got it):
Round 1: rookie filter
Hive hours of 35 and less are taken out of the equation.
Round 2:Allocate remaining players to a team based on hive score (be it weighted average or ranked select)
Round 3: Rookie allocation
Could I just say, the point of the FE is to try (yes TRY) to get some semblance of equally skilled teams so it allows everyone to have fun playing the game, not lead to one side feeling bitter because they had no fighting chance what so ever. Remember the game is way more complex than any algorithm can predict as the human factor and team work in this game is insane compared to other games on the market (ie. too many parameters and variables to model), so next time you have a crappy FE game, before you let loose, just bear in mind that having ONE vocal field commander might have made all the difference because he/she rallied the troops and made the crucial play when needed!
...so next time you have a crappy FE game, before you let loose, just bear in mind that having ONE vocal field commander might have made all the difference because he/she rallied the troops and made the crucial play when needed!
I have been a vocal field commander many a time when faced with a team of rookies, SilentComm(TM), or a team of regulars who are derping for whatever reason that day/week/month. It gets tiring and I don't do it anymore, except at certain key points such as asking why we still don't have shotguns even though fades have been up for 5 minutes.
...so next time you have a crappy FE game, before you let loose, just bear in mind that having ONE vocal field commander might have made all the difference because he/she rallied the troops and made the crucial play when needed!
I have been a vocal field commander many a time when faced with a team of rookies, SilentComm(TM), or a team of regulars who are derping for whatever reason that day/week/month. It gets tiring and I don't do it anymore, except at certain key points such as asking why we still don't have shotguns even though fades have been up for 5 minutes.
I definitely hear you, I go through bouts of NS2 paternalism. I will try to get really really helpful for a week, then just play the silent player for a while after I have had a gut full talking to the silent rookies. Mix it up, don't let it get to you @SupaDupaNoodle . Keep up the good work.
@moultano, I think we have come to some good ideas here recently. They are discussed in greater detail above, but I will summarize here.
1) Have separate hive scores for individuals based on marine skill and alien skill.
2) Treat rookies differently during a force even vote.
a) Define rookies so that they can be separated. One suggested definition was 35 hours or less playtime recorded in hive.
b)Balance teams based on hive scores for everyone excluding rookies.
c)Distribute rookies as evenly as possible among marines and aliens as if they had no skill.
3) Start rookies at a higher hive skill than 0. Possibly something 1 standard deviation below the mean hive skill.
If you guys feel that my summery is incomplete or needs corrected let me know. Please feel free to keep discussing. This thread is on a role here.
@benson, I think a positive thread title leads to a more productive discussion.
The only vote I miss is being able to play the with teams based on the previous game, even after a map change. If you had a great, balanced game, vote for another game with the same teams and it could try to make it as similar as possible by filling in the blanks due to players leaving and joining using FET.
For an extra gold star put the players as the same race as last game but automatically start another vote after allocation asking if the teams should be swapped, before the game can be started.
@moultano, I think we have come to some good ideas here recently. They are discussed in greater detail above, but I will summarize here.
1) Have separate hive scores for individuals based on marine skill and alien skill.
This is something I would really like to do. The CDT is at least considering a hive rewrite and moving to different infrastructure which would be a good opportunity to do this. In the meantime, we could probably simulate it. If you assume your chosen race is independent of the skill level of your opponents, then I think the approximate values for your race specific skill would be
2) Treat rookies differently during a force even vote.
a) Define rookies so that they can be separated. One suggested definition was 35 hours or less playtime recorded in hive.
b)Balance teams based on hive scores for everyone excluding rookies.
c)Distribute rookies as evenly as possible among marines and aliens as if they had no skill.
3) Start rookies at a higher hive skill than 0. Possibly something 1 standard deviation below the mean hive skill.
I keep getting conflicting information about how rookies should be handled. People say the skill value of 0 for rookies is too low, but they also say that the team with more rookies almost always loses. These things can't be simultaneously true! I'm inclined to not try to special case rookies in any part of it, but just to handle them as players of low skill. The skill values of other players should gradually rise and fall against whatever value we pick for rookies since there's a continuous stream of them coming in, so I suspect there isn't too much improvement to be made there.
@moultano
Some thoughts on the current implementation:
* I've noticed that FET moves players to make the teams having the exact same skill rating. If you allow for a small difference in skill between teams more players should be able to stay on their chosen team.
This seems reasonable, what would you choose as an acceptable bias? What should the probability of victory be before we say it's good enough?
* When there is an uneven amount of players on the server the team with one more player gets a lower average skill rating to compensate. The FET assumes that the round will be played with uneven teams but that is rarely the case since servers with one free player slot usually fills up in no time.
You're right, and this seems like something we should just fix. It complicates the algorithm a little bit, but not too much.
* It takes too long for players that start with a skill rating of 0 get their rating up.
Yeah it does. I didn't include anything like this initially since there weren't a huge number of rookies in the game at the time, and I wanted to make this as simple as possible to implement, but this would be a definite improvement. The current update rate is uniformly 8, and I think you could change that to something like 8 * (1 + (5 / (5 + games_played))) and it would work better.
2) Treat rookies differently during a force even vote.
a) Define rookies so that they can be separated. One suggested definition was 35 hours or less playtime recorded in hive.
b)Balance teams based on hive scores for everyone excluding rookies.
c)Distribute rookies as evenly as possible among marines and aliens as if they had no skill.
3) Start rookies at a higher hive skill than 0. Possibly something 1 standard deviation below the mean hive skill.
I keep getting conflicting information about how rookies should be handled. People say the skill value of 0 for rookies is too low, but they also say that the team with more rookies almost always loses. These things can't be simultaneously true! I'm inclined to not try to special case rookies in any part of it, but just to handle them as players of low skill. The skill values of other players should gradually rise and fall against whatever value we pick for rookies since there's a continuous stream of them coming in, so I suspect there isn't too much improvement to be made there.
There does seem to be a conflict there. #2 and #3 are trying to solve different problems though.
#2 I feel is more important because rookies at best follow orders but struggle to lose most engagements, at worst a rookie can be so bad they might as well not even be on your team. That seems pretty agreed upon in this thread.
#3 involves math, in which I personally am mostly relying on a conceptual understanding of the hive system and feedback in this thread. None of us know the mean skill rating, and rookies should be less. I feel someone else could explain it better, but 0 seems too low to start rookies out on. In fact #3 might not even be necessary if #2 is implemented.
Marine rookies stick together and kill stuff at least in the early game, but alien rookies on the other hand are not very helpful even during the beginning of a round. If you messure rines and aliens apart, you could as well let the rookies have more starting points on the rine side and less on the alien side.
Marine rookies stick together and kill stuff at least in the early game, but alien rookies on the other hand are not very helpful even during the beginning of a round. If you messure rines and aliens apart, you could as well let the rookies have more starting points on the rine side and less on the alien side.
And (just to add more to consider for @moultano's changes) -- rookies will tend to choose marines first consistently for a while, and so they should. Aliens are completely alien to someone who hasn't played NS before. To some degree if a rookie is assigned aliens too often when all they want to do is get a feel of the game first as marine -- because it's similar to other FPSes and they can get a sense of accomplishment playing -- then putting them on aliens too frequently may work against retaining them.
Yeah, no idea how we can consider that in a FE vote whilst also trying to distribute the rookies evenly across teams... unless you simply allow for +1 to the marines. If there's odd number of rookies, have that extra one on the rines.
You might think he is playing against rookies, and he could be. In higher level pubs I still have some gorges do this to me.
Tufts is amazing as a gorge, but he's still more effective as nearly anything else. When I Nerf myself its always going gorge. Or pistol. Spit efficacy falls off drastically with armor upgrades until bile bomb, then some more with wep ups.
I've gotten fantastic kds with gorge and decent kill totals. But I would hav gotten far Whittier kds but double the kills as a skulk.
Gorges just can't kill things fast enough. That's not their strength... You want to see lopsided though?/ try getting a base up in their natural. If they don't push it hard enough, it turns into a free win.
Comments
If doing it by eye, as in a captin's type pick, it would go:
Team 1:
skill 3000, skill 1000 x 3, skill 0 x 3 = total team skill 6000/7 ~857
Team 2:
skill 1000 x 5, skill 0 x 2 = total skill 5000 / 7 ~714
These numbers won't change as we can move a skill 1000 to either Team 1 or Team 2. So it looks skewered on either side. But still your best chance to have a better game. I have had reasonable games when we do Captain picks and just go by the K/D of the last game. May be you can shuffle by score of last game? Or if that isn't possible. go for shuffle by Hive score, but instead of trying to get an fairly similar skill rating on both sides, rank the players from highest to lowest, then randomly select the first player (the top player say, then make the next player on the list to the other team). That way, you at least split up the top players, and don't have Rookies to skewer the vote. So if the shuffle didn't work (stomp game- for what ever reason - eg. the top player is only a good marine), you can elect to play the match with same teams but switch sides (not sure if this is in Captain's mod or such already).
The point is you need to be able to shuffle vote easily, and may be choose from a few options:
1. shuffle by last game K/D
2. shuffle by hive rank split pick
3. shuffle by hive rank team average
4. random shuffle
Hope that would get some ideas together to work a better shuffle system.
This I think is the crux of the problem, thanks for raising it.
A good marine player is not the same as a good alien player and vice versa. We might all love playing both sides, but usually are stronger at one over the other. There's only one Hive score right now for players, when there should be two -- one for marines, one for aliens. If this is added, more balanced teams can be achieved: the best player for a side is added to that side as each team is filled. There are just two caveats to implement with this:
- Which side gets filled first each game is decided randomly by FE -- otherwise you'll always end up with the same highest marine player on marine etc (I'm pretty FE does this atm anyway).
- To ensure players don't get allocated to their 'best' team all the time -- we all love variety after-all -- either join a team first (as it works now) or perhaps every now and then start filling with the 2nd or 3rd best on a server and then go back to 1st to top down (similar to the above, this switches the order in which teams are filled, and in turn which 'score', marine or alien, is used to allocate you)
All of the above may also be complete ass. It's late and I'm tired.See my switch sides argument on this, so you get to play both sides as a team. Though I hate to restrict ppl on how they want to play the game, eg. some people just plain hate going marines as they prefer the alien life style. There in lays another problem of people F4 to go to prefer side.
It is getting late, and I don't think there is a perfect solution, if I could work out an algorithm that can predict a fair game in NS2, I would be rich predicting sports games. Not gonna happen.
edit: Also, what does the skill number say? What is the average skill? How does one know if he's considered being good or bad? Hive doesn't say anything except showing some "top" players.
There already is an ad-hoc option to choose your preferred team; just join a team before the vote passes. A minimum amount of switches will be made so your chances of staying on that team are good, especially if there are still players in the ready room.
And the skill number is only relevant with reference to other players, there is no absolute level of 'good' or 'bad' in a vacuum.
Some thoughts on the current implementation:
* I've noticed that FET moves players to make the teams having the exact same skill rating. If you allow for a small difference in skill between teams more players should be able to stay on their chosen team.
* When there is an uneven amount of players on the server the team with one more player gets a lower average skill rating to compensate. The FET assumes that the round will be played with uneven teams but that is rarely the case since servers with one free player slot usually fills up in no time.
* It takes too long for players that start with a skill rating of 0 get their rating up.
In general I think FET works really well and people who say it sucks probably haven't thought about the how complex it is to balance teams in NS2. We have a lot less bad rounds since FET was implemented.
If viable, variance should be limited to at maximum 50-70 pts. It seems to me that an average skill difference of anything over 100 is pretty significant.
Did someone ever tried to balance based of average accuarcy of the last 3 rounds?
The players with the highest skill-score are normally the ones with the highest accuracy anyway.
If you compare 2 players with 2500 skill score there can be a HUGE difference in skill cause the system dont care from where these points come from (2500 earned on wooza are not the same like 2500 earned in 6v6 gathers)
Sometimes the skill-system is "thinking":
"Wow, thats an amazing player"
But the truth can be:
"Looks like this is one of these days where im hitting nothing"
Maybe a combination of "hive skill" and accuracy represents the real skill more cause it includes the history of the player (hive) and the current ability to hit things.
Something like:
"hive-skill" (base score) * accuracy/10 = shuffle score
Btw. you dont need skill to build structures. Its important for sure but dealing damage is a combination of aim and positioning and thats what i call skill.
If it were done that way you would have to differentiate marine vs alien skills somehow as previously suggested. I like that idea simply because I am a far worse marine than I am an alien.
Your example of people checking for cloaked skulks I don't fine worthwhile because every marine should be doing stuff like that. I think that would balance out.
I do see potential for abuse though with someone artificially lowering their accuracy by shooting a wall for a long period of time to stack, but I find it unlikely that people would be so low to do that.
We have:
NS2stats
Hive stats
Accuracy and damage stats @ roundend
Different shuffle modes from shine (pure random, elo, kdr, score)
Take the shine elo vote for example.
It is based on my ideas and i modded it a bit on the HBZ server.
How does it work?
I splits the 8 players with the highest ns2stats elo score (4 on the hbz server) and it shuffles the other players based on the average kdr of the last 3 rounds.
Why elo and kdr?
You have 2 carrys per team in most games, thats why the elo split is so important and players who are able to kill the enemy have an bigger impact on the game.
It worked pretty well, better than hive stats in my opinion.
Lets find a good algorithm to combine some of these systems/data for an better result.
Given that breakdown, I would go for something like
3000x1
1000x2
0x5
vs
1000x7
I still think that this would be in favor of the team with the 3000 skill player but the 1000s team might have a small chance. Replace one of the 1000s on the 3000s' team with a 0 and it should be a close game.
I feel that the skill raating is more exponential than linear. A player with double my skill rating has WAY more than double my performance in game. You are basically looking at div 1 pros vs a pubber scrub, 12 of me wouldn't have a chance.
skill 3000 : 1 player
skill 1000: 8 players
skill 0 : 5 players
That is 7 players per team. 1 highly skilled player, 8 moderately skilled players, and 5 rookies. Rookies at best follow orders but struggle to lose most engagements, at worst a rookie can be so bad they might as well not even be on your team.
Option 1:
Team A: 3000, 1000, 0, 0, 0, 0, 0 = 4000 total team score
Team B: 1000, 1000, 1000, 1000,1000, 1000, 1000 = 6000 total team score
This option would be bad. There are too many rookies on team A vs the moderately skilled players of team B. This would create a bad experience. The current hive system probably would not balance this way because the total team scores could be more even. On multiple levels it is obvious that team B would win here.
Option 2:
Team A: 3000, 1000, 1000, 0, 0, 0, 0 = 5000 total team score
Team B: 1000, 1000, 1000, 1000,1000, 1000, 0 = 6000 total team score
This is how I feel hive would balance the teams now. Without math I would predict a loss for team A because they have more rookies. Mathematically this is one option to get the most even possible total team scores. It fails though because rookies at worst might as well not even be on your team.
Option 3:
Team A: 3000, 1000, 1000, 1000, 0, 0, 0 = 6000 total team score
Team B: 1000, 1000, 1000, 1000,1000, 0, 0 = 5000 total team score
Although this option is as mathematically balanced as option 2 I think it is more balanced in reality. The rookies are better spread out making for a more positive game. At worst, rookies can be so bad they as well not even be on your team. This essentially makes a 4v5 game with one highly skilled player. Then the question becomes, can the highly skilled player make up for having less members on your team? Saying that I think the game could go either way, but team A would have the advantage.
So, for eg, rookie= 35 hrs or less.
Force Even balances team excluding rookies
Force Even then allocates the pool of rookies to each team*
*maybe ranked by hive score, putting one player on one team, the next best player on the other team etc, but not using the score itself. Since, at the moment, it seems FE will use the lower-scored players to 'fill in the gaps' to make the total average scores balance, and this is why we sometimes see the lion's share of rookies being lumped onto one team (at least, I presume that's why this happens).
TL;DR -- don't include rookie scores as part of the balancing, and instead allocate rookies as fairly as possible after regular players are balanced.
Round 1: rookie filter
Hive hours of 35 and less are taken out of the equation.
Round 2:Allocate remaining players to a team based on hive score (be it weighted average or ranked select)
Round 3: Rookie allocation
Could I just say, the point of the FE is to try (yes TRY) to get some semblance of equally skilled teams so it allows everyone to have fun playing the game, not lead to one side feeling bitter because they had no fighting chance what so ever. Remember the game is way more complex than any algorithm can predict as the human factor and team work in this game is insane compared to other games on the market (ie. too many parameters and variables to model), so next time you have a crappy FE game, before you let loose, just bear in mind that having ONE vocal field commander might have made all the difference because he/she rallied the troops and made the crucial play when needed!
I have been a vocal field commander many a time when faced with a team of rookies, SilentComm(TM), or a team of regulars who are derping for whatever reason that day/week/month. It gets tiring and I don't do it anymore, except at certain key points such as asking why we still don't have shotguns even though fades have been up for 5 minutes.
I definitely hear you, I go through bouts of NS2 paternalism. I will try to get really really helpful for a week, then just play the silent player for a while after I have had a gut full talking to the silent rookies. Mix it up, don't let it get to you @SupaDupaNoodle . Keep up the good work.
1) Have separate hive scores for individuals based on marine skill and alien skill.
2) Treat rookies differently during a force even vote.
a) Define rookies so that they can be separated. One suggested definition was 35 hours or less playtime recorded in hive.
b)Balance teams based on hive scores for everyone excluding rookies.
c)Distribute rookies as evenly as possible among marines and aliens as if they had no skill.
3) Start rookies at a higher hive skill than 0. Possibly something 1 standard deviation below the mean hive skill.
If you guys feel that my summery is incomplete or needs corrected let me know. Please feel free to keep discussing. This thread is on a role here.
@benson, I think a positive thread title leads to a more productive discussion.
For an extra gold star put the players as the same race as last game but automatically start another vote after allocation asking if the teams should be swapped, before the game can be started.
This is something I would really like to do. The CDT is at least considering a hive rewrite and moving to different infrastructure which would be a good opportunity to do this. In the meantime, we could probably simulate it. If you assume your chosen race is independent of the skill level of your opponents, then I think the approximate values for your race specific skill would be
I keep getting conflicting information about how rookies should be handled. People say the skill value of 0 for rookies is too low, but they also say that the team with more rookies almost always loses. These things can't be simultaneously true!
This seems reasonable, what would you choose as an acceptable bias? What should the probability of victory be before we say it's good enough? You're right, and this seems like something we should just fix. It complicates the algorithm a little bit, but not too much. Yeah it does. I didn't include anything like this initially since there weren't a huge number of rookies in the game at the time, and I wanted to make this as simple as possible to implement, but this would be a definite improvement. The current update rate is uniformly 8, and I think you could change that to something like 8 * (1 + (5 / (5 + games_played))) and it would work better.
There does seem to be a conflict there. #2 and #3 are trying to solve different problems though.
#2 I feel is more important because rookies at best follow orders but struggle to lose most engagements, at worst a rookie can be so bad they might as well not even be on your team. That seems pretty agreed upon in this thread.
#3 involves math, in which I personally am mostly relying on a conceptual understanding of the hive system and feedback in this thread. None of us know the mean skill rating, and rookies should be less. I feel someone else could explain it better, but 0 seems too low to start rookies out on. In fact #3 might not even be necessary if #2 is implemented.
Definitely seems like a positive title sets people's mindset to a more positive level, and, generally, produces a more productive discussion.
Yay science!
Yeah, no idea how we can consider that in a FE vote whilst also trying to distribute the rookies evenly across teams... unless you simply allow for +1 to the marines. If there's odd number of rookies, have that extra one on the rines.
What if 2000skill alien play Gorge ?
You might think he is playing against rookies, and he could be. In higher level pubs I still have some gorges do this to me.
Tufts is amazing as a gorge, but he's still more effective as nearly anything else. When I Nerf myself its always going gorge. Or pistol. Spit efficacy falls off drastically with armor upgrades until bile bomb, then some more with wep ups.
I've gotten fantastic kds with gorge and decent kill totals. But I would hav gotten far Whittier kds but double the kills as a skulk.
Gorges just can't kill things fast enough. That's not their strength... You want to see lopsided though?/ try getting a base up in their natural. If they don't push it hard enough, it turns into a free win.