Ok, I think something needs to be tweaked in how the skill system handles rookie players.
I already posted one example above where many rookies were present, and the skill change was quite excessive.
Now I've got another case where I went "teaching commander" for a team with 6 rookies out of the 8 players. As the other team had 4 experienced players, we lost - and then I saw that I lost 50 points from that round. :-S Now I don't care too much about my skill ranking in and by itself, but it shows that something is off. Also, some people *do* care about their skill, and this may discourage them from joining rookie teams, which reinforces stacked teams... one of the things the skill system is explicitly trying to prevent.
When I play on servers with mostly veteran players, the skill changes seem reasonable though, which makes me think rookie level handling may be the issue here.
I agree , and I'm trying to come up with a good solution. (My wife and I just had a baby though, so don't expect anything soon. ) The central issue is that there weren't a large enough pool of rookies in the game since the skill system was implemented to bias the ratings higher than 1000 for most active players. Everyone started at 1000, and because the score contributions are mostly symmetric for the winning and losing teams, the skills are going to stay centered at 1000. A continuous stream of new players however will bias the ratings of active players to be higher, as everyone will have a few recent games where they beat the rookies (or beat the people who beat the rookies.) A bunch of rookies will drop out, having contributed some fraction of their 1000 points to the remaining pool of active players.
Essentially, if a new player is defined to have skill of 1000, the skill system can't calibrate for that unless there are a lot of new players feeding the system.
I'd kinda like to do something like starting everyone at a skill of 0, and still making the minimum skill value 0. This is nice psychologically, because as you start playing games your skill mostly just goes up, and it handles rookies appropriately. I haven't thought through all the other consequences it would have though, and updating the current skill values to accommodate this will take some care.
I do have a nice fix in mind for the the "seeding the server" issue. I'll work with @AceDude to get it implemented.
Yesterday i saw a successful vote ending in this :
-4 new players on marine side.
-all experienced player on alien / 0 new player.
A clear stack. Even if you remove the new player. We played gorge and sacrifice from time to time. It just ended in the marine loosing the game. Obviously.
MeatMachine dropped by on another game. Same kind of randomness. Aliens finally decided to end the game (obliterate the base) at some point to start another one. Weird feeling about 'random' confirmed.
This is not handling rookie that is the issue IMO.
Suggested simple hack to deal with rookie skills being the average veteran skill or close to it, since the system was not seeded by total playing time from the start.
# static, can never change without recalculating all players scores from round history
# after this many hours of playing...
bonus_hours = 20
# ...the player has gotten back this many points, which we will steal a few lines down...
bonus_points = 200
# ...at a rate of 10 points/h.
bonus_points_per_hour = bonus_points / bonus_hours
# the original starting score
looks_like_a_nice_number_for_a_starting_score = 1000
# a new player starts with a lower amount of points, these stolen points will then be awarded over time
starting_score = looks_like_a_nice_number_for_a_starting_score - bonus_points
# updated for each player every round
# in this example the player has a total of 108 minutes played *before* this round has been applied
total_hours = 1.8
# in this example the player played for 18 minutes in this round
hours_played_this_round = 0.3
# bonus hours rewarded from one round can not be higher than what is left to award and not lower than 0
bonus_hours_to_award = max(min(hours_played_this_round, (bonus_hours - total_hours)), 0)
bonus_time_points = bonus_hours_to_award * bonus_points_per_hour
# if this was the real code we would have added this many points to the
# skill points to be awarded (which could be negative) for this round
# in this example it's 3 points added to the skill
print bonus_time_points
Could be applied retroactively reducing the score of all players in the database with less than X hours and give them the time bonus, which is necessary since there was a sale recently.
I'd kinda like to do something like starting everyone at a skill of 0, and still making the minimum skill value 0. This is nice psychologically, because as you start playing games your skill mostly just goes up, and it handles rookies appropriately. I haven't thought through all the other consequences it would have though, and updating the current skill values to accommodate this will take some care..
Um, I looked up some players on the Hive I was playing with and it seems the Hive already does since days back?
The skill system thinks there's a bigger difference between those two rookies than between me and the 2nd rookie.
Since the skill of all players still has not been recalculated from history since the 2 (known) huge changes means "even teams" and stats calculation must be entirely broken now if it wasn't already.
Well, I know there will always be anomalies, but this one is pretty severe and he/she wacks out force even or any sort of balance that uses hive skill ratings. At time of writing, this player wins 2.42 of games. Has a K:D of 7.01 (amazingly playing with 200+ ping on servers I've seen them on... and amongst players that are of decent if not really good skill). And only has a skill rating of only 429. Needless to say, this person definitely influences game outcomes... not many players can do this.
GhoulofGSG9Join Date: 2013-03-31Member: 184566Members, Super Administrators, Forum Admins, Forum Moderators, NS2 Developer, NS2 Playtester, Squad Five Blue, Squad Five Silver, Reinforced - Supporter, WC 2013 - Supporter, Pistachionauts
edited September 2014
Currently we have the issue that rookies start with 0 while vets did start with 1000. This means there is a gap of 1000 skill points between rookies and vets.
Due to this you can't really compare the skill ratings of rookies and veterans atm. This will fix itself if those rookies stay playing ns2. The skill ratings of the highest rated vets will drop slightly while the ratings of the rookies will come closer to the ones of the vets.
Well, I know there will always be anomalies, but this one is pretty severe and he/she wacks out force even or any sort of balance that uses hive skill ratings. At time of writing, this player wins 2.42 of games. Has a K:D of 7.01 (amazingly playing with 200+ ping on servers I've seen them on... and amongst players that are of decent if not really good skill). And only has a skill rating of only 429. Needless to say, this person definitely influences game outcomes... not many players can do this.
He's cracked 1000 after 35 hours, so he'll be well on his way to an accurate value soon. Remi and I are talking about ways to make it converge faster, but we'll have to get more access to the database to implement it, so stay tuned.
Currently we have the issue that rookies start with 0 while vets did start with 1000. This means there is a gap of 1000 skill points between rookies and vets.
Due to this you can't really compare the skill ratings of rookies and veterans atm. This will fix itself if those rookies stay playing ns2. The skill ratings of the highest rated vets will drop slightly while the ratings of the rookies will come closer to the ones of the vets.
Highest rated vets drop slightly is a misnomer, a team with mostly vets and one or two newplayers wins the round and ELO drops ~50 regularly. The current system seems to severely punish any losses, and even if you lose 5 games then win 5 games with similar teams (of course individual changes makes it hard to track) the net ELO change is significantly down. In my experience the losses far outweigh the wins in terms of ELO change, particularly losses against teams that even have one rookie even though in theory the win or the loss is what matters in the new system not the individual skill of any one player. Perhaps rookie ELO values should not be factored into the average ELO rank of a team for calculating win/loss changes.
Can't we just do hive stat resets every season, like the ladder feature on Diablo & Path of Exile, that kind of thing? Would give everyone an even playing field, you could even get competitive about it lol
the skill system is completely out of whack, i have lost 9 games in a row and yet im a good player. infact i have dropped 200 skill points in 2 days and yet im often the heighest scoring player on the team. My skill level does not reflect my actual skill level and there are players with 2k skill points who in reality are more like 1200 and players with low skill who are 2k + simply because there team loses.
This is especially true on high player cap servers where the actions of 1 or 2 arent enough to sway the entire battle.
Infact if you play marines exclusively on high pop servers u will lose skill points over time because the win loss ratio is much higher for aliens so everyone stacks aliens
Yup. It is EXTREMELY hard to gain any significant amount of points as long as all these <1000 points players are about. I literally need a >75% win ratio to make any progress, and every time I lose a game, regardless of what actually happened, regardless of my performance, the last 3-4 wins are rendered meaningless.
There has to be a way to sensibly integrate in-game points into the hive score in addition to the actual outcome of a round. If I counter-stack an OP team and carry the losing team for 25 minutes going 43-5 until eventually losing, I want to be rewarded for that at least SLIGHTLY. Also, if I'm in the winning team where we have the upper hand the entire game and then lose to a silly base rush because our comm didn't pay attention, I find it to be INFURIATING to lose a shitload of points. I'd maybe accept this as collateral damage and go "oh well" if I sensed any relatable, logical system behind how points are distributed, but, especially since the "average team skill" display has been introduced, I've definitely gone "WTF!" looking at my received points after a game a lot of times, both for losing and gaining what felt like way too many points.
The way this system is working at this point can't be the intended way. No matter what anyone says or has said in the past - and I don't mean this in any deprecating way towards the people behind the system - it simply isn't working. Maybe it's just a matter of tweaking numbers as opposed to completely changing it once again, but it's hard to tell. Let's hear some more feedback both from other players and the coders.
First this system doesn't consider the roles on alien side.
The misconception i see is:
You can be the best top killer player ever, if you do gorge stuff and still be excellent at it but the game is lost you will loose points.
On the other side you can be an excellent fade but a poor gorge. The ELO can't be accurate for both depending on what you will do.
The ELO based as stated when developed is "general". The thing is it simply cannot be general due to the specific stuff on alien side.
Something everybody did see at some point : If the top alien players evolve to gorge and leave the average players do all the "fighting" jobs... It turns out in many games to be a failure. While, for pedagogy sakes, it can show some interesting things for green players. Understanding the complementarity of alien lifeforms.
If you "Gorge" (or take "Bob the builder Job") it just weakens the team so badly that it looks like rolling the dice is the same all things considered.
Force even team on ELO is just not working in this cases. Problem is : These case are most of the games at hand.
Two potential ways to deal with it:
I still vote for Huddle. You would be able to balance and assign players to specific task. ELO will help undoubtedly.
Or dynamically weight the end scores depending on what the players actually played.
Ex: Gorge shouldn't have the same number of point lost (punishment) compared to fade. The combat potential isn't the same. Also weight the scores based on the time played as this lifeform. A gorge surviving the entire game is quite a achievement.
I think you guys need to be pickier about your servers. I lately have primarily been playing on TGNS, where forced balance has been working wonderfully. There are mostly regulars, but rookies come on sometimes too.
dePARAJoin Date: 2011-04-29Member: 96321Members, Squad Five Blue
Small example why this system didnt work.
Take 2 players.
One is playing on Wooza only.
The other is playing PCWs and gathers only against the best players.
Both end in an skill-score of lets say 2000.
Im sure the Wooza player is far away from the PCW player skillwise.
Now these 2 players joining a 18 slot server and "force even teams" kicks in wich end in these 2 players are on different teams.
Put more players on the server with the same backround and a "force even teams" end in an brutal stack.
Now these 2 players joining a 18 slot server and "force even teams" kicks in wich end in these 2 players are on different teams.
Put more players on the server with the same backround and a "force even teams" end in an brutal stack.
dePARA, tell me in which (parallel)universe this scenario happens
Now these 2 players joining a 18 slot server and "force even teams" kicks in wich end in these 2 players are on different teams.
Put more players on the server with the same backround and a "force even teams" end in an brutal stack.
dePARA, tell me in which (parallel)universe this scenario happens
Now these 2 players joining a 18 slot server and "force even teams" kicks in wich end in these 2 players are on different teams.
Put more players on the server with the same backround and a "force even teams" end in an brutal stack.
dePARA, tell me in which (parallel)universe this scenario happens
It was just an example. The point being players who play mostly on one specific server will have a hive score that is absolutely worthless if they leave their usual territory because different servers/communities have different skill levels. I wouldn't have the slightest idea how to avoid this though.
Now these 2 players joining a 18 slot server and "force even teams" kicks in wich end in these 2 players are on different teams.
Put more players on the server with the same backround and a "force even teams" end in an brutal stack.
dePARA, tell me in which (parallel)universe this scenario happens
It was just an example. The point being players who play mostly on one specific server will have a hive score that is absolutely worthless if they leave their usual territory because different servers/communities have different skill levels. I wouldn't have the slightest idea how to avoid this though.
The only way to avoid is if the skill sytem would force to match people of similiar skill to play against each other on specially selected server. Games on community servers then should not be accounted. People who are very good would soon increase their position in the ladder and could not play against significantly worse players if they dont create another account or lose on purpose to decrease their ranking. This does not fix balance on community servers though. Propably you would have to handpick each player but as team work is important, this would not to 100 % assure you that all games are balanced.
Now these 2 players joining a 18 slot server and "force even teams" kicks in wich end in these 2 players are on different teams.
Put more players on the server with the same backround and a "force even teams" end in an brutal stack.
dePARA, tell me in which (parallel)universe this scenario happens
It was just an example. The point being players who play mostly on one specific server will have a hive score that is absolutely worthless if they leave their usual territory because different servers/communities have different skill levels. I wouldn't have the slightest idea how to avoid this though.
That is my point. Play in "your territory" and the skill system works great. To intermix everyone we just need a small portion of players to move around infecting others.
Overall, I think the system is probably the best we had so far. Not saying it is great, just the best so far (apart from Captains mode which is superior on multiple levels (and works good in conjunction), but there are no fully working mods for it yet as far as I know).
I think giving the rookies 0 skill points is fair. What it will do is disrupt the current scores and pull them down to lower levels. What snb observed ("need a >75% win ratio to make any progress") is fine as long as it applies to everyone. It should only be a temporary state of the scores, and players who play regularily and have a high score right now, will most likely lose a lot of points, because they are drawn to the average scores of the players they play with and against.
I think that maybe there should be some kind of "Natural decay" of skill points, meaning that you lose x points every day. This would slowly work on the problem of having players sit around with an overly high score, who dont play anymore and therefore are not giving the system the chance to readjust their scores. In the long run, this should increase the credibility of the Skill ranks, meaning that the top players in it should be prem div and other comp players.
The reason people are given 1000 points to start is to avoid having people drop down to negative values and screw up the whole algorithm by doing so. Remember when the hive was first launched and people were starting with 0 points instead of the planned 1000.
As for resetting the values from time to time, it would only make the system less valuable, as it's highly dependent on a very large samplesize to correctly portray skill values (or at least portray them as best as possible). The longer it keeps running, the better.
I also don't see the problem of having people sit on a high ranking, being inactive. The moment he starts playing again, it will readjust itself. And as long as he's not playing, there's no problem anyway.
I'm not sure how the algorithm eventually was implemented, but in the original suggestion from Moultano, it would also take into account the difference in winrate between marines and aliens, and the amount of time you played at which phase in the game (so joining a game 20min in, and only playing 3 minutes won't affect much).
The only real sad part of the ranking system is that for competitive players to be ranked somewhat correctly, they actually have to go play pubs and don't hold back.
Now these 2 players joining a 18 slot server and "force even teams" kicks in wich end in these 2 players are on different teams.
Put more players on the server with the same backround and a "force even teams" end in an brutal stack.
dePARA, tell me in which (parallel)universe this scenario happens
It was just an example. The point being players who play mostly on one specific server will have a hive score that is absolutely worthless if they leave their usual territory because different servers/communities have different skill levels. I wouldn't have the slightest idea how to avoid this though.
That is my point. Play in "your territory" and the skill system works great. To intermix everyone we just need a small portion of players to move around infecting others.
Except there isn't a single EU community left that has enough regulars at any given point to fill an entire server or the majority of one. YMMV for NA, but the times I've been on TGNS or KKG or whatever, FET was just as useless as anywhere else.
Now these 2 players joining a 18 slot server and "force even teams" kicks in wich end in these 2 players are on different teams.
Put more players on the server with the same backround and a "force even teams" end in an brutal stack.
dePARA, tell me in which (parallel)universe this scenario happens
It was just an example. The point being players who play mostly on one specific server will have a hive score that is absolutely worthless if they leave their usual territory because different servers/communities have different skill levels. I wouldn't have the slightest idea how to avoid this though.
I made a post about this a while ago, the solution is for players to try and intermix more by playing different servers and not just regging a select couple.
whats the cap on the skill system? cause a guy with 4k skill points isnt going to be 2x as good as a guy with 2k it will mess itself up when you have huge accumulated skill differences from players stacking
so the balancing will all go to whack, 2 2k players will be better than 1 4k and 1 1k etc
whats the cap on the skill system? cause a guy with 4k skill points isnt going to be 2x as good as a guy with 2k it will mess itself up when you have huge accumulated skill differences from players stacking
so the balancing will all go to whack, 2 2k players will be better than 1 4k and 1 1k etc
A skill cap makes zero sense. What you want is diminishing returns.
I've won the majority off my past 20 games, I'm like 30 points back from where I started despite like 9 consecutive wins.
Each win i gain like 6 pts, each loss I lose like 30. What's most annoying is say the teams are forced even, my team wins, I get like nothing. I'd expect ti at least get more than 6 pts when forced even. Same scenario, force even teams and I lose, I'm back to where I was 6 games ago.
Comments
I agree , and I'm trying to come up with a good solution. (My wife and I just had a baby though, so don't expect anything soon.
Essentially, if a new player is defined to have skill of 1000, the skill system can't calibrate for that unless there are a lot of new players feeding the system.
I'd kinda like to do something like starting everyone at a skill of 0, and still making the minimum skill value 0. This is nice psychologically, because as you start playing games your skill mostly just goes up, and it handles rookies appropriately. I haven't thought through all the other consequences it would have though, and updating the current skill values to accommodate this will take some care.
I do have a nice fix in mind for the the "seeding the server" issue. I'll work with @AceDude to get it implemented.
Yesterday i saw a successful vote ending in this :
-4 new players on marine side.
-all experienced player on alien / 0 new player.
A clear stack. Even if you remove the new player. We played gorge and sacrifice from time to time. It just ended in the marine loosing the game. Obviously.
MeatMachine dropped by on another game. Same kind of randomness. Aliens finally decided to end the game (obliterate the base) at some point to start another one. Weird feeling about 'random' confirmed.
This is not handling rookie that is the issue IMO.
Could be applied retroactively reducing the score of all players in the database with less than X hours and give them the time bonus, which is necessary since there was a sale recently.
If you'd like to play with it: http://repl.it/YuL
Um, I looked up some players on the Hive I was playing with and it seems the Hive already does since days back?
Example:
http://hive.naturalselection2.com/profile/137170216 Rookie that started playing recently
http://hive.naturalselection2.com/profile/28683392 Rookie that played once before, months ago.
The skill system thinks there's a bigger difference between those two rookies than between me and the 2nd rookie.
Since the skill of all players still has not been recalculated from history since the 2 (known) huge changes means "even teams" and stats calculation must be entirely broken now if it wasn't already.
http://hive.naturalselection2.com/profile/194239631
Due to this you can't really compare the skill ratings of rookies and veterans atm. This will fix itself if those rookies stay playing ns2. The skill ratings of the highest rated vets will drop slightly while the ratings of the rookies will come closer to the ones of the vets.
It seems that Mighty Jeebus is back!!!
and he changed his name to Wroth!
http://hive.naturalselection2.com/profile/0
What does this mean???
Is this good or bad for NS2 ???
I'm confused...
I guess that this indicates the start of a new era for NS2
He's cracked 1000 after 35 hours, so he'll be well on his way to an accurate value soon. Remi and I are talking about ways to make it converge faster, but we'll have to get more access to the database to implement it, so stay tuned.
Highest rated vets drop slightly is a misnomer, a team with mostly vets and one or two newplayers wins the round and ELO drops ~50 regularly. The current system seems to severely punish any losses, and even if you lose 5 games then win 5 games with similar teams (of course individual changes makes it hard to track) the net ELO change is significantly down. In my experience the losses far outweigh the wins in terms of ELO change, particularly losses against teams that even have one rookie even though in theory the win or the loss is what matters in the new system not the individual skill of any one player. Perhaps rookie ELO values should not be factored into the average ELO rank of a team for calculating win/loss changes.
This is especially true on high player cap servers where the actions of 1 or 2 arent enough to sway the entire battle.
Infact if you play marines exclusively on high pop servers u will lose skill points over time because the win loss ratio is much higher for aliens so everyone stacks aliens
There has to be a way to sensibly integrate in-game points into the hive score in addition to the actual outcome of a round. If I counter-stack an OP team and carry the losing team for 25 minutes going 43-5 until eventually losing, I want to be rewarded for that at least SLIGHTLY. Also, if I'm in the winning team where we have the upper hand the entire game and then lose to a silly base rush because our comm didn't pay attention, I find it to be INFURIATING to lose a shitload of points. I'd maybe accept this as collateral damage and go "oh well" if I sensed any relatable, logical system behind how points are distributed, but, especially since the "average team skill" display has been introduced, I've definitely gone "WTF!" looking at my received points after a game a lot of times, both for losing and gaining what felt like way too many points.
The way this system is working at this point can't be the intended way. No matter what anyone says or has said in the past - and I don't mean this in any deprecating way towards the people behind the system - it simply isn't working. Maybe it's just a matter of tweaking numbers as opposed to completely changing it once again, but it's hard to tell. Let's hear some more feedback both from other players and the coders.
The misconception i see is:
You can be the best top killer player ever, if you do gorge stuff and still be excellent at it but the game is lost you will loose points.
On the other side you can be an excellent fade but a poor gorge. The ELO can't be accurate for both depending on what you will do.
The ELO based as stated when developed is "general". The thing is it simply cannot be general due to the specific stuff on alien side.
Something everybody did see at some point : If the top alien players evolve to gorge and leave the average players do all the "fighting" jobs... It turns out in many games to be a failure. While, for pedagogy sakes, it can show some interesting things for green players. Understanding the complementarity of alien lifeforms.
If you "Gorge" (or take "Bob the builder Job") it just weakens the team so badly that it looks like rolling the dice is the same all things considered.
Force even team on ELO is just not working in this cases. Problem is : These case are most of the games at hand.
Two potential ways to deal with it:
I still vote for Huddle. You would be able to balance and assign players to specific task. ELO will help undoubtedly.
Or dynamically weight the end scores depending on what the players actually played.
Ex: Gorge shouldn't have the same number of point lost (punishment) compared to fade. The combat potential isn't the same. Also weight the scores based on the time played as this lifeform. A gorge surviving the entire game is quite a achievement.
Take 2 players.
One is playing on Wooza only.
The other is playing PCWs and gathers only against the best players.
Both end in an skill-score of lets say 2000.
Im sure the Wooza player is far away from the PCW player skillwise.
Now these 2 players joining a 18 slot server and "force even teams" kicks in wich end in these 2 players are on different teams.
Put more players on the server with the same backround and a "force even teams" end in an brutal stack.
3-13 in those 16 games good going skill system. You said it was balanced i dropped 300 skill points
dePARA, tell me in which (parallel)universe this scenario happens
Alpha Omega X2307
The only way to avoid is if the skill sytem would force to match people of similiar skill to play against each other on specially selected server. Games on community servers then should not be accounted. People who are very good would soon increase their position in the ladder and could not play against significantly worse players if they dont create another account or lose on purpose to decrease their ranking. This does not fix balance on community servers though. Propably you would have to handpick each player but as team work is important, this would not to 100 % assure you that all games are balanced.
That is my point. Play in "your territory" and the skill system works great. To intermix everyone we just need a small portion of players to move around infecting others.
I think giving the rookies 0 skill points is fair. What it will do is disrupt the current scores and pull them down to lower levels. What snb observed ("need a >75% win ratio to make any progress") is fine as long as it applies to everyone. It should only be a temporary state of the scores, and players who play regularily and have a high score right now, will most likely lose a lot of points, because they are drawn to the average scores of the players they play with and against.
I think that maybe there should be some kind of "Natural decay" of skill points, meaning that you lose x points every day. This would slowly work on the problem of having players sit around with an overly high score, who dont play anymore and therefore are not giving the system the chance to readjust their scores. In the long run, this should increase the credibility of the Skill ranks, meaning that the top players in it should be prem div and other comp players.
As for resetting the values from time to time, it would only make the system less valuable, as it's highly dependent on a very large samplesize to correctly portray skill values (or at least portray them as best as possible). The longer it keeps running, the better.
I also don't see the problem of having people sit on a high ranking, being inactive. The moment he starts playing again, it will readjust itself. And as long as he's not playing, there's no problem anyway.
I'm not sure how the algorithm eventually was implemented, but in the original suggestion from Moultano, it would also take into account the difference in winrate between marines and aliens, and the amount of time you played at which phase in the game (so joining a game 20min in, and only playing 3 minutes won't affect much).
The only real sad part of the ranking system is that for competitive players to be ranked somewhat correctly, they actually have to go play pubs and don't hold back.
I made a post about this a while ago, the solution is for players to try and intermix more by playing different servers and not just regging a select couple.
so the balancing will all go to whack, 2 2k players will be better than 1 4k and 1 1k etc
A skill cap makes zero sense. What you want is diminishing returns.
Each win i gain like 6 pts, each loss I lose like 30. What's most annoying is say the teams are forced even, my team wins, I get like nothing. I'd expect ti at least get more than 6 pts when forced even. Same scenario, force even teams and I lose, I'm back to where I was 6 games ago.