Best criteria for Matchmaking

Share your ideas and Suggestions about Blood Bowl 2.
Antonlunau
Posts: 28
Joined: 24 August 2011, 15:32

Re: Best criteria for Matchmaking

Postby Antonlunau » 19 September 2013, 12:59

I usually have alot of respect for what Dode writes but this just seems wrong. Ill have to check my teams win-loss record when i get home to see what they would end up facing.

Edit: I realize that with this system in place i would most likely not get to a point where i would have a big + on the win-loss ratio in the first place.

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

Re: Best criteria for Matchmaking

Postby dode74 » 19 September 2013, 13:51

Koadah - I'm in the "whatever works best is best" camp.
I would still prefer that a match between teams with similar TV & similar zSum get a higher weighting than a match where largish TV + zSum differences cancel out to give similar TV+.
Either way you'd have an even match with a predicted chance of winning roughly the same as your opponent's.

Anton - why does it seem wrong? Bear in mind that, as you say, the win-loss record would be different to what you currently have were your teams to develop in the TV++ environment. Bear in mind further that your own zSum should not be compared with zero as a base. Most unsuccessful teams get deleted while successful teams get played, so the median zSum will likely be greater than zero as the number of games played (and therefore the TV) increases. I might take a look at FOL later to see if I can say what the median zSum is.
Image

User avatar
Koadah
Posts: 1211
Joined: 08 April 2009, 16:17
Contact:

Re: Best criteria for Matchmaking

Postby Koadah » 19 September 2013, 15:49

Koadah - I'm in the "whatever works best is best" camp.
Well, 90 might balance the books quicker and better but, if you win two close games and I lose two close ones I get a 360k TV advantage? With no inducements?

Have we really diverged that much in two games? Is that what you are calling fair?

Get off the fence Dode. ;)

This isn't non competitive school sports day. You don't have to give me a win because I'm 0-2.
I would still prefer that a match between teams with similar TV & similar zSum get a higher weighting than a match where largish TV + zSum differences cancel out to give similar TV+.
Either way you'd have an even match with a predicted chance of winning roughly the same as your opponent's.
Ah yes, trog, gut feel thinking I know. My gut goes with the similar TV & zSum not the big TV advantage/disadvantage. Not saying that I expect them to implement it but hey, if you don't ask...
CaRBB

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

Re: Best criteria for Matchmaking

Postby dode74 » 19 September 2013, 16:00

Well, 90 might balance the books quicker and better
Quicker, perhaps, but better? Not necessarily. Higher n means faster to get there but less accuracy when you do: 90TV either way doesn't allow for the specificity that 30TV does. Given that we don't know how the balance between accuracy and speed needs to be played out I don't know which is better, or if indeed it is either one of those. Mike suggested 30TV in his blog, but also the following:
  • "It doesn't matter what the value is... make it 30... make it 90... make it whatever you want. It is absolutely impossible to say what the "best" value is for a TVPlus environment until that environment has been in use for a while and we have data to analyze in order to create a "best" value based on actual data."
Ah yes, trog, gut feel thinking I know. My gut goes with the similar TV & zSum not the big TV advantage/disadvantage.
And you're usually so good at listening to your head ;)
Image

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

Re: Best criteria for Matchmaking

Postby VoodooMike » 19 September 2013, 18:48

The premise is that the inducements will not be enough to balance the teams, but that excludes coaching skill since the stats does not take into account if the inducements are given to the team with the better coach/elo rating.
There's no rating that measures coach skill. There's no way (that isn't constantly disputed and opposed) to separate out mechanical team strength from coaching skill in terms of outcomes, so it's a bit like complaining we can't separate out the eggs from an already baked cake before testing to see how many calories it's worth. TV is meant as a measure of team strength, but nobody ever really believed that a 1000 TV goblin team was the same strength as a 1000 TV dwarf team... and the data has supported that suspicion. Since there is no effective measure of team strength in absence of coaching skill., and no effective measure of coaching skill in absence of team strength, we instead look for an effective measure of team strength + coaching skill. TVPlus is the best we've found so far.
Really, is that how it works?! So a great coach with an 1100 TV team will face a 1800 team with a lousy coach? That cant be right
Yes, that's exactly right. More importantly, it might be possible for the 1100 TV and 1800 TV teams to play without inducements if the lower TV coach is extremely good and the 1800 TV coach is extremely bad. Why? Because the better coach is capable of winning the match without them, and we know it... if they were given 700k of inducements the game would probably be a massacre... maybe the better coach would enjoy a massacre, but the worse coach probably wouldn't. Good game players want to play a game that lets them exercise their skill... bullies want to take a crap on someone.
It wouldn't be as bad as Antonlunau thinks but I would still prefer that a match between teams with similar TV & similar zSum get a higher weighting than a match where largish TV + zSum differences cancel out to give similar TV+.
It shouldn't really matter if they have similar TV and similar zSum, or if they're different but collectively create the same TVPlus rating. From a statistical standpoint I'd rather see the latter, since that's what will really stretch the legs of the TVPlus system and test out its long-term ability, but realistically what you're saying is likely to be more common nomatter what.
Edit: I realize that with this system in place i would most likely not get to a point where i would have a big + on the win-loss ratio in the first place.
To use TVPlus in an existing environment, as I've said, we'd either need to deal with a painful adjustment period, or track zSum separately from when we started using TVPlus, such that the skewed W/L records from the previous environment don't require that adjustment period. That said, you can still use TVPlus as a solely predictive measure in an existing environment if you just want to know who is more likely to win a match, as that's what it cut its teeth on.
Well, 90 might balance the books quicker and better but, if you win two close games and I lose two close ones I get a 360k TV advantage? With no inducements?
Then you might lose the next one.... of course, in your analogy you're playing the same person over and over again, so the effects are doubled. In reality you probably wouldn't see that happening, and would only see the 90 change, and it wouldn't equate to a 90k change in inducements since the next person you play will likely be closer in TVPlus rating to you than the person you last played is. If you're playing the same person over and over you'd probably hit a point where you go back and forth between matches which are slightly in your favour and ones that are slightly in his favour, and sometimes just due to the dice or bad mistakes on someone's part, it'll move a bit farther than fair and there will be a match which favours one person more than usual that brings it back into its stable range.
Have we really diverged that much in two games? Is that what you are calling fair?
You didn't diverge during the games, your ratings just haven't proven effective in predicting those matches and thus, are not considered to be accurate yet because they're not relatively stable. The aim of TVPlus is perfectly even matches, but the reality of the game (and even of TVPlus) is that actual matches will always be a little bit skewed in one side's favour.. but much less than it is under TV, and who its in favour of will be unpredictable unlike now. Over time that averages out the same way dice luck does, and the "bias" is small enough not to be even close to as big an influence as the dice.
Quicker, perhaps, but better? Not necessarily. Higher n means faster to get there but less accuracy when you do: 90TV either way doesn't allow for the specificity that 30TV does. Given that we don't know how the balance between accuracy and speed needs to be played out I don't know which is better, or if indeed it is either one of those.
Yes, higher is quicker but less long-term accurate... or more specifically, more likely to have games where people have a slight advantage applied, even when they've reached stable ratings... but randomly applied. 30k takes longer to create those balanced matches, but creates more evenly balanced matches. The problem with "takes longer" is the median number of games MM teams currently play. I'd rather see the system push that median up by improving the experience of the average player immediately, rather than assuaging the fiddly nit-picking of long-term players who would be playing even if we were matching based on penis length.

I don't ultimately care which value the environment would start at. I'd want to run the numbers after 1000, 10,000, whatever.... games and then apply a much more accurate value at that point. Maybe we can split the difference and go with 60? Koadah seems to be OCD-focused on the number... maybe we could use his age as the value? I'll be happy with any implementation, frankly.
Friendly Reminder: Correlation does not equal Causation - tattoo it on the inside of your eyelids if it'll help.

Antonlunau
Posts: 28
Joined: 24 August 2011, 15:32

Re: Best criteria for Matchmaking

Postby Antonlunau » 19 September 2013, 19:24

I just checked my 2 most-played FOL team.

Undead 1660 Rating 40-5-9 (+31 Ratio)
Norse 1430 40-10-10 (+30 Ratio)

+30 is 2700 extra TV. Any ratio plus of more than 10 is going to be meaningless since it will not happen.

Eksample:
Orc team with 20 games played and and a +4 Win ratio. Rating 1600
Chaos Team 90 games played -6 win ratio. Rating 2300


No amount of coaching skill is going to stop the massacre here. The teams at TV 2100+ is very bash heavy. Even nuffle himself will not be able to do anything when you are down to 4 players. Also i cant wait for people to start making TV2000 Bash teams with enough negative ratio to face TV1000 Teams. Thats going to be fun for everybody.

I dont like a system that awards people for making dumb choices. Dont ever get better it will just make things more difficult for you. Play badly and make sure to ragequit alot and you will be rewarded with easy matches


The system is going to need a

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

Re: Best criteria for Matchmaking

Postby VoodooMike » 20 September 2013, 04:10

Undead 1660 Rating 40-5-9 (+31 Ratio)
Norse 1430 40-10-10 (+30 Ratio)
I'm so glad you keep on this straw man even after admitting you understand that such ratios would never come to pass under a TVPlus system.
No amount of coaching skill is going to stop the massacre here. The teams at TV 2100+ is very bash heavy.
Obviously that's not the case, or those teams wouldn't have the records they do. Before you reached the level of "massacre" you'd enter the realm of "victory is less likely" and "victory is very much less likely" at which point the zSum score would stop increasing. You (and often koadah) seem unable to understand the TVPlus system as progressive rating system, and you keep getting stuck on individual matches that you take not only out of context, but specifically implying there can't be a context. No team magically gets to its rating... the orc team in your example got there by playing 20 games and both gaining and maintaining a +4 zSum by playing increasingly difficult matches until the coach finally hit a point where the victories stopped being easy and a higher zSum couldn't be held.
Even nuffle himself will not be able to do anything when you are down to 4 players.
And yet, in the face of this supposed fact, the orc coach has managed to maintain that +4 ratio, which suggests that he knows a thing or two about his supposed limitations that you apparently do not.
Also i cant wait for people to start making TV2000 Bash teams with enough negative ratio to face TV1000 Teams. Thats going to be fun for everybody.
Whom they'd have to somehow allow to win every single time, just so they can play a bash game they're going to lose? Maybe you need to take a look at the "min/max" issue in TV matched play... you know, the one everyone cries about that, when the data is examined, happens so infrequently that it might as well be a bigfoot sighting? The point behind TV matched minmaxing is that you can "game the system" and increase your chances of winning by doing so. What you're suggesting is people playing all the games needed to gain the SPP to make a killer team, then play a throw a bunch of games to get a bad record, so they can sit there and deliberately lose every game they play.

Very, very few people waste the time and effort needed to win via min/maxing at low TV... your assertion is that a ton of people would put in all that time and effort to LOSE at low TVPlus. I guess we'd have to wait and see.
I dont like a system that awards people for making dumb choices. Dont ever get better it will just make things more difficult for you. Play badly and make sure to ragequit alot and you will be rewarded with easy matches
From the sound of your complaints and the look of your FOL teams, it sounds like you're someone who doesn't like to lose, and the prospect of not being able to find a niche in which you're unlikely to lose is frightening to you. If the idea of losing is anathema to you, its quite possible you're not interested in challenge, in which case you're not really someone who plays the game, you're someone who games the game system. That may be a "sweet spot" for a player, but it comes at the expense of other players.

To each their own, I suppose.

The design goal for the system is to make sure everyone is challenged at their own level, and to make sure the high ratings are given to the best players.. and to do everything we can to prevent the possibility of artificially raising one's rating past what one deserves. In TV MM your W/L/D record is meaningless... people can, successfully, min/max their way into an insanely high win ratio.. but they're winning all their games against much inferior teams. Similarly, high TV monster chaos teams can gain and maintain an equally insane record (go look at WMD in Box on FUMBBL). Only bash teams can reliably get and maintain high TVs in TV MM, too.... so both measures are mostly worthless.

BB already has a handicapping system. Handicapping systems always do exactly what you're complaining you don't like systems to do... it just happens that BB's handicapping system is really bad, and it has thus far worked in your favour. If you play hard, you'll be given challenges that suit your team development and level of skill when you play hard.... if you play sloppy and casual, you'll end up getting challenges that suit your sloppy and casual style of play. To top it off, if you can get and maintain a high TVPlus rating you can absolutely claim to be an excellent coach, because there's no way to game your way into a high rating without being an excellent coach, only into an artificially low rating... and nobody is going to care much about you if you're rated crap.
Friendly Reminder: Correlation does not equal Causation - tattoo it on the inside of your eyelids if it'll help.

User avatar
Koadah
Posts: 1211
Joined: 08 April 2009, 16:17
Contact:

Re: Best criteria for Matchmaking

Postby Koadah » 20 September 2013, 04:48

make sure the high ratings are given to the best players
By ratings do you mean the zSum?

TV is a big part of TV+ so that is as much about building a big team as coach skill. If you star player dies your TV+ rating goes with him.

A lot of coaches do not seem that keen on playing at high TV. I am wondering if this would encourage even more coaches to go the 'rinse & repeat' route as winning games at mid TV would push them closer to the heavy bash zone.

A high TV+ rating doesn't sound like something that would be that highly sought after. Unlike CR. Or maybe that is the plan. ;)
CaRBB

User avatar
Viajero
Posts: 287
Joined: 12 July 2009, 14:04

Re: Best criteria for Matchmaking

Postby Viajero » 20 September 2013, 07:50

Hey Mike,

Since my first post of this thread is not 100% clear or, put in another way, since I am not sure Cyanide will actually bother trying to make any sense out of all this... maybe it is a good idea for you to create a new thread where you try to describe the proposed TVPlus matching system/criteria in a simple and succint manner (adding all relevant links to your blogs and other discussions etc) so to make it super easy for Cyanide to understand in case they decide to try it out?

Then Dode feel free to copy paste all this thread into that new thread eventually so to avoid duplicity or some such etc?
Last edited by Viajero on 20 September 2013, 09:49, edited 1 time in total.
My general availability to start games is 20:00 - 21:30 GMT+1 during the week with some more flexibility on weekends.

erestor
Posts: 116
Joined: 27 January 2009, 20:42
Location: Debout sur le terrain à gueuler sur l'arbitre, et le sort qui s'acharne...

Re: Best criteria for Matchmaking

Postby erestor » 20 September 2013, 08:05

The rule that limited the minmaxing was the "No inducement" rule for the Major Tournament.

In my league, there are only 8 games concerned by season (4 Major Tournaments and Semi-Final & Final for each - no play off, only a selection based on different criterias for each tournament). And if you want to win these tournaments, you can not build an unbalanced team or you will be pulverised. And as long as it concerns only a very few number of games compared to the total amount of games played, it does not spoil the TV system that could be easily replaced by a zSum or a TV++ system.

So Major Tournaments with no inducements could be a good way to avoid minmaxing (because every player wants to participate to a MT, nope ? But no one wants to participate in pouch-front briefs...)
As the long departed Commissioner Three-horn famously said on many occasions “This is Blood Bowl, buddy, anything could happen!”.


Return to “Ideas and Suggestions”

Who is online

Users browsing this forum: No registered users and 4 guests

cron