FUMBBL :: Online Blood Bowl League

Hitonagashi

Joined: Apr 09, 2006

Posted: Apr 04, 2012 - 16:59

Christer wrote:

Something I would find interesting to do would be to take a very statistical approach to building a new team strength formula. For example, analyze win percentages based on the number of players on each team and see if there's a correlation. Them do the same for the number of block players, number of dodge vs tackle, etc. After coming up with a large number of weights / correlation functions, you'd combine it all into a best-estimate win percentage which would be fairly easy to apply to a pair of teams.

One major difference from TV, TW and TS would be that you couldn't give a strength to a single team, but instead require a pair of teams for evaluation.

This method could be extended to the number of expected TDs for each team, and the number of casualties as well resulting in what should be a very very strong match-making system which would be very difficult to "game".

It'd be a fairly large project to do, but it'd absolutely be possible to do and I have a feeling that the resulting system would be practically as fair as it could be.

This would be very interesting..but doesn't this get back to Chainsaw's blog, that to do this on any sort of large scale, we need to know the roster and skills at the time the match was played?

koadah

Joined: Mar 30, 2005

Posted: Apr 04, 2012 - 17:09

Christer wrote:

I didn't really think the blackbox would maintain popularity because of these killer teams dominating it, and I only did the blackbox development in the first place because there was a small chance it'd be a success. It turned out my instinct was wrong, and I'll happily admit it because in the end it means more people get more enjoyment out of the site.

Though it appears that under LRB4 you were dead right.

CRP Box is way too soft. Wink

_________________

New teams. Secret League or Official. Always recruiting!

stej

Joined: Jan 05, 2009

Posted: Apr 04, 2012 - 17:55

This sounds interesting. One could produce some sort of Generalised Linear Model to predict the chance of a team winning. Sounds like fun if you have all the data to play with Smile

VoodooMike

Joined: Nov 07, 2010

Posted: Apr 04, 2012 - 20:09

Ehlers wrote:

Now they play a match and following players get mng: 2,5,8 (round up to 11 players with JM)
Mean: 23,73 & Median: 8 & 2.97%
TV: 1620 -> 1668

The MNG players would still count in the calculation, which would make it Mean: 20.25, Median: 12.5 so there would still be no adjustment. If you fired some of your skilled players and replaced them with rookies you might trigger it, though you would, in fact, have a team that was looking like what the formula was meant to be triggered by.

Ehlers wrote:

Now we add one rotter to the team, because at that TV range and only 11 players is just equal to heavy beating and most likely another slaughter.

Ok, if we effectively add 4 0-spp players then yes, we're going to finally trigger the adjustment, as then it would be Mean: 19.06, Median: 8, so you'd have your TV increased by 2.2%

Ehlers wrote:

I might be wrong, my calculations might be wrong. IF so, please do and go and correct me.
BUT if my calculations are right, it seems this system force me to keep minmax and not add more players as that will increase my TV and each added rotter will be more and more expensive. So this system in my eyes forces to keep my team low.

Not profoundly wrong, no. All the players on the roster (whether they play or not) would be part of the calculation. If you suddenly add a ton of rookies then yes, you're going to get an upward adjustment, though I don't see how that forces you to "minmax".

As you played with your team, and those rookies started gaining SPP, the adjustment would flatten out and disappear. It's important to note that under such a system, other teams would be likewise affected - yours wouldn't be in a vacuum - so very minor % adjustments would be unlikely to affect your pairings in any significant fashion unless you're trying to maximize the effect of your high/low player differential, which you obviously wouldn't be.

Christer wrote:

I find it funny that after years of complaints on the Team Strength system we used here prior to LRB6, we now have a 12 page thread arguing to bring it back (sure, in some other shape but it's conceptually the same).

The problem is that nobody ever tries to create a TS measure that is based on anything but random opinion about the "objective" power of various things. An internally valid TS measure would be the holy grail of this sort of thing, but again, the data to create it is not easily available to the stats people =\

Pubstar wrote:

Again, maybe I'm not grasping the definition of min/maxing properly, but it seems like having a team with a decently high TV avoids them, and the only consequence of letting them min/max is that rookie teams have a hard time starting up. Which, in my personal experience, they do anyway, with a lack of rr's/block.

One of the major complaints about minmax teams is that they're bashy, CLPoMB teams that can attack low, undeveloped teams and result in a ton of permanent injuries and death. Getting your team past those teams, to a higher TV, requires that your players still be alive and useful such that they can gain skills and stats in order to increase that TV... of course, then you'll just run into the non-minmax, full roster CLPoMB teams that will send you back down...

I don't think the issue is restricted to CLPoMB teams, obviously, since it's not simply Chaos and Pact teams that are minmaxing, but I can certainly see some people's concerns.

Christer wrote:

Something I would find interesting to do would be to take a very statistical approach to building a new team strength formula. For example, analyze win percentages based on the number of players on each team and see if there's a correlation. Them do the same for the number of block players, number of dodge vs tackle, etc. After coming up with a large number of weights / correlation functions, you'd combine it all into a best-estimate win percentage which would be fairly easy to apply to a pair of teams.

I would suggest some form of data mining, maybe as simple as a factor analysis, using the information. Granted, the resulting formulas would not be something you could explain to people for them to use their gut to decide if they agreed with it or not, but it would certainly maximize the relationship between the resultant formula and whatever you were using as a criteria.

I still think that a per-team TS would be better than trying to create a system for matching specific pairs of teams, but that's a subjective position on my part. I have no objective argument against it.

Ehlers

Joined: Jun 26, 2006

Posted: Apr 04, 2012 - 20:47

@JimmyFantastic
ah oki, just thought it was aimed at me since it was posted right after my post Smile

sorry then

VoodooMike wrote:

The MNG players would still count in the calculation, which would make it Mean: 20.25, Median: 12.5 so there would still be no adjustment. If you fired some of your skilled players and replaced them with rookies you might trigger it, though you would, in fact, have a team that was looking like what the formula was meant to be triggered by.

EGH? Why the hell should your MNG player affect your next match?

VoodooMike wrote:

As you played with your team, and those rookies started gaining SPP, the adjustment would flatten out and disappear. It's important to note that under such a system, other teams would be likewise affected - yours wouldn't be in a vacuum - so very minor % adjustments would be unlikely to affect your pairings in any significant fashion unless you're trying to maximize the effect of your high/low player differential, which you obviously wouldn't be.

But then it shows you have no knowledge of how such as Nurgle perform as a team. Take a look at the nurgle top teams in the black box. Most of them have key players with lots of SPP and then lots of rookie rotters. It is near impossible to get rotters above 16 spp or just even get them over 6spp.

Christer wrote:

Now, don't get me wrong.. I'm pushing the issue to its extreme to make the point, and it's really not as black and white as I stated above, but there _is_ wisdom in the old saying "If it ain't broke, don't fix it". Is the Blackbox broken? I don't think so. Could it be improved? Sure, but it's difficult for me to alienate the avid B players in order to increase its popularity. It's really really hard to justify the risk when there are so many players who enjoy the format as it works now.

Christer you are doing great! And really no need in my viewpoint to change how the scheduler works. The only problem in this thread as others point out is the clawpomber and maybe even blodge. I am really not concerned with blodge at all, just that clawpomber is too strong and can really wreck havoc. Should you do any work on this now? No, focus your energy on the other awesome projects you have. It is just the time of month when FUMBBL have its period.

VoodooMike

Joined: Nov 07, 2010

Posted: Apr 04, 2012 - 21:00

Ehlers wrote:

EGH? Why the hell should your MNG player affect your next match?

You say that like they don't already affect your next matchup by contributing to your TV. It's about adjusting the TV value used for matching based on team composition, not about anything else, really... and in this case, using MNG players in the calculation AVOIDS the "broken" thing you were talking about, so again.. what's the issue?

Ehlers wrote:

But then it shows you have no knowledge of how such as Nurgle perform as a team. Take a look at the nurgle top teams in the black box. Most of them have key players with lots of SPP and then lots of rookie rotters. It is near impossible to get rotters above 16 spp or just even get them over 6spp.

Anything is possible. Link me to some of these teams you feel are totally reasonable and will break the system and I'll be happy to go in and run numbers and such.

As I said, the system is abstract, and is not meant as some knife-edge balance thing. I'm not super concerned with the rough edges because it's just a napkin solution, not something crafted with love. I'm certainly going to argue against any arguments that I don't find logical, though, even if I have little emotional investment in the system's perfection.

Ehlers

Joined: Jun 26, 2006

Posted: Apr 04, 2012 - 22:56

VoodooMike wrote:

Anything is possible. Link me to some of these teams you feel are totally reasonable and will break the system and I'll be happy to go in and run numbers and such.

As I said, the system is abstract, and is not meant as some knife-edge balance thing. I'm not super concerned with the rough edges because it's just a napkin solution, not something crafted with love. I'm certainly going to argue against any arguments that I don't find logical, though, even if I have little emotional investment in the system's perfection.

http://fumbbl.com/FUMBBL.php?page=teams&group=&order=&nav=10&race=55
Just begin from the top and work your way down.

You dont play any games on FUMBBL. So I cant really see how or why you are so concerned in trying to fix something that works.
There is no fact that point to that the black box has a problem. Most popular div.
People are not really complaining about minmax, but it is related to claw+po+MB being broken.

So all in all, hardly any reason at all to implement this napkin solution. So back to the drawing board and come up with something better.

PainState

Joined: Apr 04, 2007

Posted: Apr 04, 2012 - 23:09

Ehlers wrote:

There is no fact that point to that the black box has a problem. Most popular div.
People are not really complaining about minmax, but it is related to claw+po+MB being broken.

What da heck? Box as a div is a problem because the "problem" that is being discussed is primarily in the Box because of the scheduler.

CLPOMB is the problem? really?

CLPOMB is a problem because of the scheduler coupled with min/max.

Notice no one is accusing Chuck VS, WMD or Olun as being CLPOMB min/max cheese? We might not like CLPOMB but we can tell the diffrence between those teams and the teams that are taking advantage of the box scheduler in the form of min/max CLPOMB cheese. WHICH BTW IS WHAT THIS THREAD IS ALL ABOUT

popularity might also be part of the issue here. Meaning that is the way to get ahead in the box so might as well join them.

_________________
Comish of the:

VoodooMike

Joined: Nov 07, 2010

Posted: Apr 05, 2012 - 03:18

Ehlers wrote:

You dont play any games on FUMBBL. So I cant really see how or why you are so concerned in trying to fix something that works.

Ah, some true colours... forgive me my ignorance for trying to accommodate your concerns when it turns out you're just planning to be ignorant. Thanks for the heads up!

As a side note, I do think there are some ways to improve upon the base idea.. I think, for example, it would probably be better if only the most experienced 11 players on a roster were used, since anyone trying to keep their TV "artificially" low by padding a roster with rookies or using journeymen, would not bother putting more players on their roster than necessary.

Similarly, there might be something to be said for capping the SPP of any given player at 176 for purposes of the calculation, since any SPPs above 176 don't really do much.