Christer
Joined: Aug 02, 2003
|
So, things have been running for a while now, so I figure I'll publish information on how the blackbox system works. The games have been rasonably good so far, but there are a handful of matches that are a bit outside of what most people call fair. As you'll see, this trend might spiral further, so I'm probably going to rework some aspects of the system.
Either way, this is how it works right now:
All B coaches have a Blackbox Ranking (BR), and a Blackbox Bash Ranking (BBR), which work more or less the same as in R. BR is based on winning TDs, while the BBR is based on casualties.
The suitability score takes the win probability for both these figures based on respective ranking and strength, just like in R. Let's call these two probabilities P and bashP. These are as in R adjusted with the racial factor according to the charts available in the stats section.
Now the system subtracts 0.5 from both probabilities, making them both span from -0.5 to 0.5. This subtraction makes it symmetrical around the zero point, which is useful for the later steps of the system.
The bash probability number is now weighed to make it less important in the final figure:
bashP = bashP * 0.3
Now, the system calculates the "distance" as so:
distance = sqrt(P*P + bashP*bashP)
Which, of course, is the linear distance you all know from basic geometry.
At this point, a randomizer factor is applied:
distance += rand(0, 0.02)
Then the distance is normalized to (0, 1).
If the two teams are of the same race, apply a factor:
distance = distance * 0.97
If either of the two teams had their last match against the opponent, apply a factor:
distance = distance * 0.94
And finally, calculate the suitability:
suitability = round(1000*(1-distance))
This suitability score is calculated for every possible matchup for the participating coaches. This list is the total list of possible matchups in the system.
To generate the matchups, the system does the following:
1. Find the match with the lowest suitability score from the matches in the set.
2. Pair up each of the teams with their respective "best" matches, which must have suitability > 850.
3. Remove all matches involving either of these two coaches from the set of possible matches.
4. Repeat until there's less than two coaches left to pair up.
There you have it. The explanation is somewhat simplified, but covers every important detail of the scheduler. |
|
|
MSpanke
Joined: Jun 06, 2006
|
  Posted:
Nov 17, 2008 - 00:57 |
|
How does this work with extreme abuse? Lets say a coach with a 0% winning record on a team built to destroy. Norse with multiple DPs for example? Anyone game to test this out? |
|
|
Randompuppet
Joined: Jul 09, 2007
|
  Posted:
Nov 17, 2008 - 01:17 |
|
When do we get to see our (BR) and our (BBR)
Randompuppet |
|
|
DukeTyrion
Joined: Feb 18, 2004
|
  Posted:
Nov 17, 2008 - 01:30 |
|
I presume then that you have decided that BR will remain a crucial part of your formula, to aim for the golden 50% win rate?
Or is there still a possibility you would consider leaving BR outside the formula for matching teams?
Also, this seems to work on a coach basis, so there if a coach has 1 basher and 2 soft teams, you are in reality offering other coaches little protection against the bash team. |
|
|
fly
Joined: Aug 02, 2003
|
  Posted:
Nov 17, 2008 - 01:30 |
|
I've had bashP not calculated stronger - when my team is outnumbered - kick my rear several times. |
_________________ I play for fun. I play to win.
Do you play CPOMB 'cause you can't win otherwise?
No, that's a rhetorical question. |
|
SillySod
Joined: Oct 10, 2006
|
  Posted:
Nov 17, 2008 - 01:41 |
|
Duke, he has tweaked it since posting that. A very good tweak IMO.
<Christer> Anything beyond a TS difference of 5 is now worth three times as much
<pythrr> well, thats some BR shed, anyway
<AxeMurder> What did you tweak?
*** lugiak has quit (Quit)
<pythrr> ^^
<Cris> Good night all!
*** Cris has quit (Quit)
<Christer> So 10 difference now is considered as 20 difference before the tweak
<Kryten> that might make everyone happy
<Christer> 20 now is equivalent to 50 before
<SillySod> mm, nice change
<Christer> That will tighten the TS spread alot |
_________________ Putting the "eh?" back into Sexeh.
"There are those to whom knowledge is a shield. There are those to whom it is a weapon. Neither view is balanced." |
|
EnglishJake
Joined: Apr 06, 2007
|
  Posted:
Nov 17, 2008 - 01:50 |
|
Are the rosters going to be locked after the schedule is calculated? I've had last minute wizards hired on me. |
|
|
pythrr
Joined: Mar 07, 2006
|
  Posted:
Nov 17, 2008 - 01:55 |
|
Jake. That is illegal. Submit a support ticket. |
_________________
|
|
Mr_Foulscumm
Joined: Mar 05, 2005
|
  Posted:
Nov 17, 2008 - 02:00 |
|
I still don't get why you decided to use BR.
Glad to hear there has been tweaks anyway. Really hope this helps. |
_________________ Everybody's favorite coach on FUMBBL |
|
arw
Joined: Jan 07, 2007
|
  Posted:
Nov 17, 2008 - 03:24 |
|
DukeTyrion wrote: | Also, this seems to work on a coach basis, so there if a coach has 1 basher and 2 soft teams, you are in reality offering other coaches little protection against the bash team. |
...isn't that great in fact? It emphasizes to play soft teams doesn't it? Can't say I'ld feel sorry for the guy who is just playing Khem and having a disadvantage compared to the one playing Khem n Elves.
BUT it is not exactly like that since "bashP [is] adjusted with the racial factor according to the charts available in the stats section" (Christer). |
Last edited by arw on %b %17, %2008 - %03:%Nov; edited 1 time in total |
|
Mattybee
Joined: Mar 22, 2007
|
  Posted:
Nov 17, 2008 - 07:12 |
|
I think that tightening the TS spread will do a lot to cut down on the wacky matches. Like I've said before, I don't mind if BR (or BBR, apparently) is a factor; I just want it to be a very mild one when compared with "are these teams even"? The emphasis should be on even teams, and if there are a group of teams who are all fairly even, THEN match up by BR/BBR/BBQ/whatever. |
|
|
ploplo25
Joined: Aug 02, 2003
|
  Posted:
Nov 17, 2008 - 07:45 |
|
I love the idea of BBR but i think it needs to be a more important factor in the formula.
Lets face it, the annoying teams when playing BB are the ones that bash and dont try to win/score. I dont mind being stomped if the other coach is playing good and doesn't need my full team out to score.
imo, inverting the BR and BBR coefficients would be better, or maybe only make both at value 1
Other point is the racial factor, OK looks great in the principle (amazons vs dwarves is a classic example) but it has flaws.
My norse have been matched up against dwarves 8 times out of 13 games. This is indeed a fair matchup and I prefer 100 times dwarves than orcs at low TR with norse .... but a bit of diversity doesn't hurt
Is it feasible to include te racial history of the team in the formula (without complicated database queries) ? Maybe add a coefficient 1-(100-nb of team from one race)/100 to the calculation or a percentage of played races.
Just to illustrate the above, my norse team with good winning record (80% at the time) and decent bashing skills (3-4 per game including fouls) got paired against :
http://fumbbl.com/FUMBBL.php?page=match&id=2423341
+31 / +25 dwarves with 8 guards and 5 MB + 2 DP
I played one half and looked my remaining 6 players getting gang blocked/fouled during the rest of the game :/ |
_________________ Dancing Queen !!
Young and Sweet !!! ..... |
|
SillySod
Joined: Oct 10, 2006
|
  Posted:
Nov 17, 2008 - 08:54 |
|
I disagree, the last thing we want to do is to artificially modify the bash/agile balance of the division by giving significant advantages to teams that avoid causing casualties. Personlly I like to win while causing as much collateral damage as possible... I'm happy to keep winning however it affects my matchups because I value BR but it would annoy me immensley if my personal playing style got overtly punished for no apparent reason (and I play vampires more than anything else). I can see the point of making games have a more even chance of win/loss but evening up the chances of dying totally destroys a natural (and very interesting) gaming balance between races.
I think we should keep the BBR factor at a bare minimum, enough to avoid the kill-only coaches getting easy games for losing alot. |
_________________ Putting the "eh?" back into Sexeh.
"There are those to whom knowledge is a shield. There are those to whom it is a weapon. Neither view is balanced." |
|
DukeTyrion
Joined: Feb 18, 2004
|
  Posted:
Nov 17, 2008 - 11:05 |
|
SillySod wrote: | Duke, he has tweaked it since posting that. A very good tweak IMO. |
Well, I am just not ken on the inclusion of BR at all.
Anyway, I have done my part to help with the testing, but Blackbox with BR is obviously not for me, so I have deleted my teams and return to Ranked and Smacks.
I know the loss to the division is zero and I am sure it will go on to remain the most popular division on Fumbbl, but it is just not for me. |
|
|
SeraphimRed
Joined: Feb 01, 2004
|
  Posted:
Nov 17, 2008 - 11:23 |
|
It would be nice to see a results spread of how the change has effected the average TS difference of match ups (I'm sure Christer has run simulations based on changes), but that looks good. Though I'd consider that 5 TS is perhaps too little, and maybe the cutoff should be more like 8-10.
But I'm still yet to play a game myself to see how it works in practice, I know some of the links I've seen make me go "eek" though. |
_________________
Enhance YOUR FUMBBL
|
|
|
| |