New Matchmaking System

Everything dealing with the video game developed by Cyanide!
licker34
Posts: 163
Joined: 09 March 2016, 17:40

Re: New Matchmaking System

Postby licker34 » 23 March 2016, 19:31

Did you miss the context of discussing teams with fewer than 5 games played?
I know, and that doesn't change a word of what I said.
Then what you said was completely tangential and irrelevant to the point I was making.
Teams with that few games have not had the time to establish their actual performance level to a satisfactory level.
Which is a reason to INCREASE the 50, not decrease it. Because your team with 3 wins more might actually find playing the sameTVplusWizard team a walkover.
Where did I say anything about changing the zsum multiplier? Set it to whatever you want, it STILL WON'T WORK WELL FOR TEAMS WITH FEW GAMES.

Math, that's what Mike would say, and you don't understand it. And they he'd change your nickname (probably to Dope74, but I don't think you are one) and call you a troll.
I don't even care if that's fair or not, it's dumb is the point.
No, you just don't like it. There's a pretty big difference. If they are evenly matched and coach B wins then the system is doing its job! And you're still talking as if a wizard is an iWin button...
Really? When I specifically said spend your 160k on something other than a wizard? But none of that remotely addresses the point I was making.

For some reason you didn't quote the actual example I gave. You seem to do that a lot when the examples might prove inconvenient to the truth you are trying to sell.

The fact is that no matter what zsum multiplier you want to choose it won't accomplish it's goal for teams with few games played. Mike actually acknowledged this in some thread somewhere before he got all huffy at me for being smarter than him. See I agree that TV++ will do the job it's supposed to do when the environment it's applied to is large and robust (meaning coaches not churning teams at a high rate). Now I don't think that job needs to be done, but that's beside the point of what I'm getting at now.

Which is that at low games played TV++ won't do what it's supposed to do, and it will lead to more 'uneven' pairings because of the variance in performance at low TV (and I really mean only for new teams, but also true for teams with few skills/rrs even if they have played more games). If your numbers from before are true, and most teams are retired around the 5 game mark than it follows that most of the games actually played are being played by teams with few games, and thus inaccurate zsums.

That's not really a problem if you only look at aggregates, but the problem is that those aggregates are created by individuals, and it's the experience had in those individuals which actually matters. If TV++ makes that experience worse at this end, that's a problem, and that's why the suggestion was made to not even apply TV++ until after a certain number of games had been played by the team, and Mike wasn't opposed to that because he recognized this as a potential issue.

Let's say you and Mike both make new Human teams with the same build, and lets say you play each other in the first game for each team. Let's say the game is 1-1 and on turn 15 you have to make a simple dodge with your catcher to score. You fail the 1/9, Mike is able to pick it up and hand it off to his own catcher who then runs away and scores the next turn. As in my example, now if you two are matched again, Mike will be facing ~150k in inducements against him. And for what reason? Because you hit an 11% and he didn't? That sounds reasonable to you?

Because here's the thing. If you (meaning Mike probably) had played your 10s or 100s or 1000s of open ladder matches you'd know how common results like that are between evenly matched coaches. It's the pure variance of the dice which turn wins into losses. As TV grows though, this variance is reduced because you have added more skills (which lower it) and more rerolls, and even have a bench or apo to reduce casualty variance.

At low games played you have none of that, yet you want to handicap these mechanically identical teams anyway. That pushing for an aggregate result of 50/50, but as I've said so many times, it's the individual games which actually matter, and people are not going to be thrilled to face ridiculous inducements after one game.

And that's like my opinion. But one based off of 100s of actual games played in actual ladders. You know what though? It really won't bother me that much, I'm not worried about it affecting my ability to reach whatever zsum I wind up reaching. I'm not even worried about it annoying me in the short term, because I can get over it. I'm not the guy who's recycling his teams every 5 matches.

But for those guys who are prone to do that? Yeah, I think this makes it worse for them. Not that it matters right? We're not getting TV++, we're getting something else which will do something else. Still, if you're serious about making TV++ the best it can be, then you do need to address this low games played issue.

Or, just ignore it because some pool of data says something and guys who've played approximately zero ladder games know what's best for guys who've played 1000s of them, and chatted with their opponents in them. You want to paint me as a power mad fumbbl douchebag (ok, you're more classy than Mike, and I appreciate it) who only cares about my win rate. In fact that's just not true. Well the fumbbl part is and douchebag part might be ;) I'm a lot like you, I want a better environment than what we have, though, as we agree, that environment is really beyond our control or ability to greatly influence.

I still think you and Mike are missing a golden opportunity to really balance out these matches though. You can mine your data pool to see what the actual TV imbalances are for the purely mechanical differences at low TV and just give Chaos some free inducements when they play Dark Elfs since they are clearly an underdog, and the goal is to give every team a 50% chance of winning.

User avatar
VoodooMike
Posts: 1614
Joined: 14 July 2009, 07:44
Location: Calgary, Alberta
Contact:

Re: New Matchmaking System

Postby VoodooMike » 23 March 2016, 19:34

Man, you boys get ugly quick. I'll concede y'all are passionate about this and I hope Cyanide is finding this helpful.
A-yup... but the follow-up hate-sex is something to behold!
Friendly Reminder: Correlation does not equal Causation - tattoo it on the inside of your eyelids if it'll help.

User avatar
dode74
Posts: 7041
Joined: 11 December 2008, 11:18
Location: Nr. Reading, UK
Contact:

Re: New Matchmaking System

Postby dode74 » 23 March 2016, 19:58

Set it to whatever you want, it STILL WON'T WORK WELL FOR TEAMS WITH FEW GAMES.
AND IT WILL STILL WORK BETTER THAN TV WILL!

Math, WindowLicker34 ;) (again, in jest).
The fact is that no matter what zsum multiplier you want to choose it won't accomplish it's goal for teams with few games played.
Not completely, obviously (and I have said as much before). But it STILL does it better than TV does alone. You're complaining it's not perfect even when it is better than what we have. The best is the enemy of the good.
Image

licker34
Posts: 163
Joined: 09 March 2016, 17:40

Re: New Matchmaking System

Postby licker34 » 23 March 2016, 20:06

No, it's likely worse than what we have. Because you don't need to throw around extra inducements at low games played to try and even out win rates.

At the point in a teams life time when their variance is highest you just want to dump on more variance.

That' s a bad design, and it's why TV++ should only kick in after some number of games have been played. I'd say start it at 5, but that number might need to be higher. And or the zsum might need a kicker as plasmoid suggested. Anyway, those are all just tweaks to smooth out big jumps in the handicap adjustments.

I'm still wondering your opinion on using the data to provide inducements for obvious mechanical differences at low games played. For example the blueyonder data shows that Dark Elfs only have a win rate of 40% vs. Norse up to 1100TV. Seems they would deserve some extra inducements for those matches.

twitch/the_sage_bb
Posts: 616
Joined: 17 December 2015, 08:06
Facebook: https://www.facebook.com/theSagebb/
Twitter: https://twitter.com/the_Sage_BB

Re: New Matchmaking System

Postby twitch/the_sage_bb » 23 March 2016, 20:28

Man, you boys get ugly quick. I'll concede y'all are passionate about this and I hope Cyanide is finding this helpful. Now bring on the Norse!
Ewww, another conceder! Burn them all!
Content: Twitch / Youtube ; Updates: Facebook / Twitter
(because people with big banners are just compensating)

User avatar
dode74
Posts: 7041
Joined: 11 December 2008, 11:18
Location: Nr. Reading, UK
Contact:

Re: New Matchmaking System

Postby dode74 » 23 March 2016, 20:29

No, it's not worse. It actually can't be generally worse since there is more information in the system and the system is acting on it. Your "low start" solution simply means it takes longer for the team to get to even close to the right zSum.
At the point in a teams life time when their variance is highest you just want to dump on more variance.
No, I want to get it out of the point where TV, the poorer method of matching which is causing that initial high variance (which it must be if the variance is highest when zSum is zero), has the most influence and into the point where zSum, actual information in the system, has the most.

I've considered things like altering zSum multiplier at varying stages (similar to your "5 games without it" concept) but they can all be gamed. 5 games without it also means you miss at least half of all teams.
blueyonder data shows that Dark Elfs only have a win rate of 40% vs. Norse up to 1100TV
276 games played as of right now. Your 40.4% mean actually fails to properly represent a range from 34.5 to 46.3. Feel free to equate that to some sort of inducement value...
Image

User avatar
VoodooMike
Posts: 1614
Joined: 14 July 2009, 07:44
Location: Calgary, Alberta
Contact:

Re: New Matchmaking System

Postby VoodooMike » 23 March 2016, 20:40

I'm not as much of a masochist as Mike makes out. You can do your circular reasoning as loudly as you like, but you'll be doing it to yourself.
I can hear you muttering "it hurts so good" under your breath again, Dode.
Friendly Reminder: Correlation does not equal Causation - tattoo it on the inside of your eyelids if it'll help.

licker34
Posts: 163
Joined: 09 March 2016, 17:40

Re: New Matchmaking System

Postby licker34 » 23 March 2016, 20:45

No, it's not worse. It actually can't be generally worse since there is more information in the system and the system is acting on it. Your "low start" solution simply means it takes longer for the team to get to even close to the right zSum.
The zsum is still calculated, it's just not used. I don't think accurate zsums will happen within 5 or 10 games anyway, so that part really doesn't matter. And it certainly can be worse. You're applying an additional inducement value on teams which have not established their actual performance rate. You are throwing more variance at them for no practical reason. Why? Because your goal is to produce 50% win rates, you're not doing that in any reasonable manner though as the in game variance at low games played is simply too high for tv++ to smooth out, you just make it worse.
At the point in a teams life time when their variance is highest you just want to dump on more variance.
No, I want to get it out of the point where TV, the poorer method of matching which is causing that initial high variance (which it must be if the variance is highest when zSum is zero), has the most influence and into the point where zSum, actual information in the system, has the most.
But TV doesn't change appreciably for low games teams while TV++ is an instant swing of 100 (that's 5 skills or an extra positional!) for a 1 game team. These teams are nominally mechanically identical since they have had no chance to add anything yet.
I've considered things like altering zSum multiplier at varying stages (similar to your "5 games without it" concept) but they can all be gamed. 5 games without it also means you miss at least half of all teams.
I'm not sure what 'missing half off all teams' means. And you can game the system however it's implemented, that really isn't a consideration in my opinion.
blueyonder data shows that Dark Elfs only have a win rate of 40% vs. Norse up to 1100TV
276 games played as of right now. Your 40.4% mean actually fails to properly represent a range from 34.5 to 46.3. Feel free to equate that to some sort of inducement value...
I equate it to 12 wizards. If that's too much we'll just let the system self correct.

User avatar
dode74
Posts: 7041
Joined: 11 December 2008, 11:18
Location: Nr. Reading, UK
Contact:

Re: New Matchmaking System

Postby dode74 » 23 March 2016, 22:28

I don't think accurate zsums will happen within 5 or 10 games anyway
Because.. you don't think?
You're applying an additional inducement value on teams which have not established their actual performance rate.
On the contrary, they have established their precise performance rate so far. You seem to be operating under the illusion that a team has a particular zSum at which it will sit, whereas we know that performance of teams changes as the team itself changes. Rolled a +ST? Awesome. Lost the AG6 Wardancer? Less so.
But TV doesn't change appreciably for low games teams while TV++ is an instant swing of 100 (that's 5 skills or an extra positional!) for a 1 game team. These teams are nominally mechanically identical since they have had no chance to add anything yet.
And we're measuring performance. If the 100TV is too much then it will swing the other way. That's how the system works.
I'm not sure what 'missing half off all teams' means. And you can game the system however it's implemented, that really isn't a consideration in my opinion.
Typo - half of all teams. The median games played is 5, so any factor not implemented before 5 games misses half the teams. And gaming the system absolutely should be a consideration. It's the ability to game it which got us here.
I equate it to 12 wizards. If that's too much we'll just let the system self correct.
That would be great if you had any sort of self-correction happening. As it is you don't, which means you're talking nonsense. And you know it.
Image

morbidorbits
Posts: 17
Joined: 08 October 2009, 16:30

Re: New Matchmaking System

Postby morbidorbits » 23 March 2016, 22:30

I was warned that you were an idiot. I now have the proof I need, thanks Mike.
I'd take this to heart but I notice that you've only posted 8 times. Clearly your forum record indicates that your opinions should not be taken as equal to those of someone with, say, hundreds of posts on this forum. True, you may have posted elsewhere at various points in your posting career, but those don't count toward your posting experience here.

Given your original "thoughts" on the matter, this should make perfect sense to you.

If all of your posts are like the very few that I have read, they are almost certainly all hot stinky air, and therefore don't count either. :lol:


Return to “General”

Who is online

Users browsing this forum: No registered users and 9 guests

cron