38 coaches online • Server time: 19:53
Index Search Usergroups Profile
Log in
Recent Forum Topics goto Post A Very Steely Setup ...goto Post Help! My +ST wardanc...goto Post POLL: FUMBBLer of th...
SearchSearch 
Post new topic   Reply to topic
View previous topic Log in to check your private messages View next topic
Christer



Joined: Aug 02, 2003

Post   Posted: Oct 14, 2017 - 19:29
FUMBBL Staff
Reply with quote Back to top

In detail, the ranking system uses the following (ELO-like) formula:

CR' = CR + k * (S - p)

CR' = new ranking
CR = old ranking
k = a scaling factor that's based on the brackets of the coaches involved. Ranges from 0.52 to 6. Lower numbers are used for matches where the coach is in a higher bracket than the opponent. There are two variants of this:
k = 0.5 + (1 / [bracket difference]), ranging from 1 to 5. Used when a team is in a higher bracket than the opponent
k = 3 + [bracket difference], capped at 6. Used when a team is in a lower bracket than the opponent
k = 2 in the normal case, where coaches are in the same bracket.
A higher k value is used for the first few games (sort of like "placement matches").
S = result factor, ranging from 0.0 to 1.0:
S = 0 for a loss by 3 or more TDs
S = 0.1 for a loss by 2 TDs
S = 0.2 for a loss by 1 TD
S = 0.5 for a tie
S = 0.8 for a win by 1 TD
S = 0.9 for a win by 2 TDs
S = 1.0 for a win by 3 or more TDs

p = win probability based on CRs and TVs of the teams involved, with a racial factor applied. Basically, a 0.5 value here means the coaches, TVs and races makes the site think there's a 50% chance of winning.
Christer



Joined: Aug 02, 2003

Post   Posted: Oct 14, 2017 - 19:40
FUMBBL Staff
Reply with quote Back to top

To do more math, an example of a "cherry picker" coach losing a match against a low rankned coach, let's say 0-1:

k=6, S=0.2, p=0.5 => 6*(0.2-0.5) = -1.8

That single loss would require 1.8/0.26 roughly 7 4-0 victories to recover from. Now, even vs complete rookies, 4-0 is hard, so let's say the cherry picker wins by 1 TD:

k = 0.52, S=0.8, p=0.5 => 0.52*(0.8-0.5) = +0.16 CR

That's 11 matches to recover. That's a 92% win rate to break even, which seems unlikely to me.
JellyBelly



Joined: Jul 08, 2009

Post   Posted: Oct 14, 2017 - 19:57 Reply with quote Back to top

Christer wrote:
That's 11 matches to recover. That's a 92% win rate to break even, which seems unlikely to me.


I'm not going to name any names, but I've certainly seen picky teams in Ranked with win rates around that level.

_________________
"Opinions are like arseholes, everybody's got them and they all stink." - The protagonist, Fallout 2

"Go for the eyes, Boo! Go for the eyes!!" Razz
Antithesisoftime



Joined: Aug 20, 2014

Post   Posted: Oct 14, 2017 - 20:12 Reply with quote Back to top

JellyBelly wrote:
Christer wrote:
That's 11 matches to recover. That's a 92% win rate to break even, which seems unlikely to me.


I'm not going to name any names, but I've certainly seen picky teams in Ranked with win rates around that level.
I've only seen 2 or 3 couches that pick that hard, and even then they tend to have a less than 85% win rate.

One such skaven couch souls bee dripping in CR soon
Christer



Joined: Aug 02, 2003

Post   Posted: Oct 14, 2017 - 20:16
FUMBBL Staff
Reply with quote Back to top

Looking at some stats with 55% completion on the ranking recalc makes me think there will be a lot of discussion and complaints about the new formula.
mrt1212



Joined: Feb 26, 2013

Post   Posted: Oct 14, 2017 - 20:20 Reply with quote Back to top

Christer wrote:
Looking at some stats with 55% completion on the ranking recalc makes me think there will be a lot of discussion and complaints about the new formula.


So mission accomplished Wink
JellyBelly



Joined: Jul 08, 2009

Post   Posted: Oct 14, 2017 - 20:53 Reply with quote Back to top

As I've been saying in the other thread that's been going for the past couple of days (don't know if you've been following it, Christer):

My 2 cents is that it would be better to simply take away CR from Ranked entirely (or at least from open play in R). It's just not a very competitive environment, when players can pick and choose their games. And the fact that people can gain equal rewards/rankings in a more casual environment means that there is less incentive for coaches to play in Box (which is, as I understand it, aiming to be more hardcore/competitive).

Just my humble opinion.

_________________
"Opinions are like arseholes, everybody's got them and they all stink." - The protagonist, Fallout 2

"Go for the eyes, Boo! Go for the eyes!!" Razz
Joost



Joined: Mar 17, 2014

Post   Posted: Oct 14, 2017 - 22:06 Reply with quote Back to top

JellyBelly wrote:
As I've been saying in the other thread that's been going for the past couple of days (don't know if you've been following it, Christer):

My 2 cents is that it would be better to simply take away CR from Ranked entirely (or at least from open play in R). It's just not a very competitive environment, when players can pick and choose their games. And the fact that people can gain equal rewards/rankings in a more casual environment means that there is less incentive for coaches to play in Box (which is, as I understand it, aiming to be more hardcore/competitive).

Just my humble opinion.


isn't there a distinction between the 2 scores anyway? one for ranked and one for Box?
Cyrus-Havoc



Joined: Sep 15, 2006

Post   Posted: Oct 14, 2017 - 22:22 Reply with quote Back to top

Looks like it's finished.

_________________
Not Undead but perhaps the oldest living coach!
Christer



Joined: Aug 02, 2003

Post   Posted: Oct 14, 2017 - 22:26
FUMBBL Staff
Reply with quote Back to top

Yes, it is. However, I am making some adjustments and will run the script again (another 3 hour process).

The changes I'm adding are:

- Limit (S-p) in the formula to not go negative, as that would in extreme cases mean a CR loss even if you win.

- Adjust some internals of the win probability calculation to increase overall CR standard deviation (effectively, giving top coaches a higher CR). This will make it harder to climb to the top because of a random streak of good luck.
Craftnburn



Joined: Jul 29, 2005

Post   Posted: Oct 14, 2017 - 22:33 Reply with quote Back to top

And to think, all that's needed to avoid all this work is to eliminate CR. Wink

Elo is fitting for chess, because there is NO luck in chess..

BB 'aint Chess! (Though I love both!)
Christer



Joined: Aug 02, 2003

Post   Posted: Oct 14, 2017 - 22:33
FUMBBL Staff
Reply with quote Back to top

If you have nothing to add beyond "Remove CR", please feel free to not post in this thread; it's not intended for you.
Grod



Joined: Sep 30, 2003

Post   Posted: Oct 14, 2017 - 22:48 Reply with quote Back to top

Including point differentials rewards teams with a certain offensive playing style (elves) and punish those that focus in defence (dwarves).

_________________
I am so clever that sometimes I don't understand a single word of what I am saying.

Oscar Wilde
Kalamona



Joined: Apr 21, 2005

Post   Posted: Oct 14, 2017 - 23:24 Reply with quote Back to top

While I understand the reason behind the change scoring TD is not equally easy with all races. This will encourage coaches to use fast teams and avid playing with dwarf for example. You dont often see a dwarf team outscore their opponent by 3 TD.
Cloggy



Joined: Sep 23, 2004

Post   Posted: Oct 14, 2017 - 23:44 Reply with quote Back to top

So are you saying that a change that encourages people to play agile teams and try to score a lot instead of elfstalling is a bad thing?

Sounds pretty good to me to be honest Smile

_________________
mmmmmmmmmmm... Bloodbowl.........
Display posts from previous:     
 Jump to:   
All times are GMT + 1 Hour
Post new topic   Reply to topic
View previous topic Log in to check your private messages View next topic
Powered by PNphpBB2 1.2 © 2003 PNphpBB Group
Credits