Concern: Skill ranking for commanded games is always zero
Adambean
Cardiff, South Wales Join Date: 2005-06-03 Member: 53038Members, Constellation, Reinforced - Shadow
Cardiff, South Wales Join Date: 2005-06-03 Member: 53038Members, Constellation, Reinforced - Shadow
Discussed already? Don't know, this was a hard one to search for... what with "hive" and "commander" being used for alien commanders in general.
I have a concern about the Hive rankings for games played as commander. If you command a game whether you win or lose, you will be given a skill ranking of zero for these games. This will quickly cause your skill rating to plummet if you command often, giving the appearance of an inexperienced or bad player when in fact you may be experienced / good. (in my case just experienced -- certainly not good)
I guess the skill for each round is calculated by your score, kill count, death count, and kill:death ratio. Maybe not all of those. Obviously it would be ridiculous to make some sort of prediction of what skill rating a commander should get, or take some form of average from the players in the team. It doesn't help that commanders get zero score throughout all games (apart from sudden base defence kills and sentry/whip kills), not even a score for structures placed (then completed building) or ARC kill points. What may be better is to exclude all commanded games from a player's overall skill rating all together. Just display a hyphen '-' against those rounds instead of zero.
There is also the issue of how do we determine the commander as you can just switch. NS2 Stats seems to track this fine. It may be fair to assume that if you've been a commander for at least 33% of the round time, you're marked as a commander for that round. Even if there were more than one of you. In most games the commander doesn't leave the chair/hive for 10% of the game time, let alone a third of it.
Why would anyone care about the original point? The problem lies in a couple of places.
1) The server browser: This displays an average skill rating of those players present (after completing connection and being "in the world") -- this display is impaired as good/experienced appear as bad/inexperienced prompting more newer people into a game they think they are suited for.
2) Balance modifications: I've seen a couple of servers now using a skill balance modification using the Hive ranking data. This is in an effort to stop stacking by balancing teams based on accumulative skill rating. Not perfect, but better than just randomizing players. This is again impaired as I appear to have low skill then get "balanced" onto a team with higher skill people, and these plug-ins prevent you from switching teams manually. Build 263 now uses Hive ranking data in the same way for its' new balance vote.
3) Inaccurate representation: The least concerned part. You simply look bad on the Hive rankings, and makes all of the data less meaningful. Stats are rarely perfect, but this is still something that can be considered to make them slightly more respectable.
Also I'm not sure how true this is but I've read somewhere here that the current Hive ranking system is incomplete and going to be wiped clean/restarted once finalized? True or not the original point here should still be considered, especially as the data is tied in with the game on multiple occasions now.
Here come the flames.
I have a concern about the Hive rankings for games played as commander. If you command a game whether you win or lose, you will be given a skill ranking of zero for these games. This will quickly cause your skill rating to plummet if you command often, giving the appearance of an inexperienced or bad player when in fact you may be experienced / good. (in my case just experienced -- certainly not good)
I guess the skill for each round is calculated by your score, kill count, death count, and kill:death ratio. Maybe not all of those. Obviously it would be ridiculous to make some sort of prediction of what skill rating a commander should get, or take some form of average from the players in the team. It doesn't help that commanders get zero score throughout all games (apart from sudden base defence kills and sentry/whip kills), not even a score for structures placed (then completed building) or ARC kill points. What may be better is to exclude all commanded games from a player's overall skill rating all together. Just display a hyphen '-' against those rounds instead of zero.
There is also the issue of how do we determine the commander as you can just switch. NS2 Stats seems to track this fine. It may be fair to assume that if you've been a commander for at least 33% of the round time, you're marked as a commander for that round. Even if there were more than one of you. In most games the commander doesn't leave the chair/hive for 10% of the game time, let alone a third of it.
Why would anyone care about the original point? The problem lies in a couple of places.
1) The server browser: This displays an average skill rating of those players present (after completing connection and being "in the world") -- this display is impaired as good/experienced appear as bad/inexperienced prompting more newer people into a game they think they are suited for.
2) Balance modifications: I've seen a couple of servers now using a skill balance modification using the Hive ranking data. This is in an effort to stop stacking by balancing teams based on accumulative skill rating. Not perfect, but better than just randomizing players. This is again impaired as I appear to have low skill then get "balanced" onto a team with higher skill people, and these plug-ins prevent you from switching teams manually. Build 263 now uses Hive ranking data in the same way for its' new balance vote.
3) Inaccurate representation: The least concerned part. You simply look bad on the Hive rankings, and makes all of the data less meaningful. Stats are rarely perfect, but this is still something that can be considered to make them slightly more respectable.
Also I'm not sure how true this is but I've read somewhere here that the current Hive ranking system is incomplete and going to be wiped clean/restarted once finalized? True or not the original point here should still be considered, especially as the data is tied in with the game on multiple occasions now.
Here come the flames.
Comments
Is the level rating simply an accumulation of score? (nothing to do with kills/deaths directly)
=> Proposal for abuse-proof skill system
I command a ton and I have noticed this.
Or make it an average of what the team received during that round. Since the commander is in charge of their team...
Ok consider this:
Proplayer (win/lose = 7:3) joins game.
Noobplayer (win / lose 3:7) joins game.
->
Shuffle based on win/lose ratio
->
Proplayer's chance of winning the game: 50% (fair teams)
Noobplayer's chance of winning the game: 50% (fair teams)
->
Game ends, data send to Hive.
->
Repeat x 100 games.
->
Proplayer win/lose: 5:5
Noobplayer win/lose: 5:5
So the trouble with this pure win/lose ratio as a base for the system is that the system will fuck up itself (its own data) by making the teams fair.
Or is my logic flawed?
I do respect statistics, but I refuse to believe that your W/L stats will even out to your 'true level' after enough observations simply because the player always gets to choose whether to stack or anti-stack. In effect, the player gets to choose whether to handicap themselves or the opposing team. A good player always anti-stacking would end up with a lower W/L ratio than a player of equal skill who always stacks.
How would you make it work? An Elo-type scoring with your personal score compared to the average of the opposing team, or your team's average score against that of the opponents?
Because the skill gained is weighted based on the teams and the expected outcome. I.e losing against a team with higher average skill and in turn a higher chance of winning results in a lower/negligible skill loss as they were expected to lose. By the same argument winning a game you are expected to win will result in less/negligible skill gain as you were expected to win. So stacking/anti stacking is irrelevant.
You should read Moultano's original thread to answer those points.
Really, the theory works beautifully. So much better than any other implementation I've seen. Now we just need to test it on a live system. It does need a bit of tweaking as regards the commander role, but I think Moultano said he was thinking about that already.
commander.playerlevel = commander.playerlevel + 1
else
if team loses
commander.playerlevel = 0