54 coaches online • Server time: 13:13
Forum Chat
Log in
Recent Forum Topics goto Post Secret League Americ...goto Post DOTP Season 4goto Post Skittles' Centu...
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: Apr 21, 2004 - 19:42
FUMBBL Staff
Reply with quote Back to top

The difference in team strengths have been made into a relative measurement instead of a static difference:

dT = 100 * [s1/min(s1,s2) - s2/min(s1,s2)]

The ranking formula has been changed accordingly.
Sinner



Joined: Aug 02, 2003

Post   Posted: Apr 21, 2004 - 20:22 Reply with quote Back to top

uhhhm, when is the big "InRankDay"?
Will all so far unranked teams start at 150 or will their ranking be recalculated by games played? Will the new ranking influence the coach ranking at once or will only future games do so?

_________________
Sinner
Darkie's Dreams - successfully cherrypicking any race, any coach, any rating, any number of DP since 20/09/2003 ... and still winning!
Mully



Joined: Aug 02, 2003

Post   Posted: Apr 21, 2004 - 22:02 Reply with quote Back to top

Quote:

The difference in team strengths have been made into a relative measurement instead of a static difference:


Disagree with this change (suprised?). You are giving way TOO much weight to the coach ranking vs team strength. The old method was fine as it was.

Basically what the ranking is saying is <b>if</b> a 200 rated coach with a STR200 team wants to play a "fair" game (ie a 50/50 chance to win) he either ;

1) has to find a coach ranked 158 or higher
or if none available
2) <edited as I was wrong with this statement>

The difference in the quality of coaches between 210 rank and 160 rank is not that large. And is certainly not comparable to the difference between a 210 STR team and a 160 STR team. (which is what the new ranking insinuates)

To help clarify my point - in the following scenario do you think you have a 50/50 chance of winning this game ? :

Coach Christer (208) playing with this 197/196 team
http://fumbbl.com/FUMBBL.php?page=team&op=view&team_id=37680

vs

Coach Santa (155) playong with this 260/256 team
http://fumbbl.com/FUMBBL.php?page=team&op=view&team_id=44255

When the STR is adjusted for handicaps the "ranking" says you have a 53% chance of winning this game. But I'll bet that, depsite your 16-1-1 record with these skaven, you would be lucky to win 4 out of 10 games against Santa's team.

I know you want the ranking system to be meaningfull. But to be meaningfull, people have to understand it and buy into it. That's tough to do when it you can find flawed examples such as these.

Would you re-think this change or at least offer your reason why you think the new system better calculates how coaches/teams would fair against each other?

I look at it as what matchup do I consider a 50/50 game and how does ranking formula compare.

_________________
Owner of the REAL Larson
Come join the CCC League


Last edited by Mully on Apr 21, 2004 - 23:07; edited 3 times in total
Christer



Joined: Aug 02, 2003

Post   Posted: Apr 21, 2004 - 22:19
FUMBBL Staff
Reply with quote Back to top

First off, Str 200+ teams can play against other str200+ teams, regardless of the difference between them.

Second, the 40 str diff is NOT normalized, but a simple difference to make it easy for coaches to realize what games are allowed and what games are not.

There are other people who strongly believe that strength should be taken out from the ranking formula entirely and leave it with comparing coach rankings only, just like they do in Chess, Magic the Gathering, Othello, etc etc.

The reasoning behind this change is that str 100 vs 140 is more one-sided than str 200 vs 240. This change reflects that.
thmbscrws



Joined: Aug 02, 2003

Post   Posted: Apr 21, 2004 - 22:21 Reply with quote Back to top

I don't pretend to understand this but i have feeling it will only make it even harder for those of us with high coach rankings to get games than it already is. Crying or Very sad

_________________
"If God really existed it would be necessary to abolish him." - Mikhail Bakunin
Christer



Joined: Aug 02, 2003

Post   Posted: Apr 21, 2004 - 22:31
FUMBBL Staff
Reply with quote Back to top

thmbscrws wrote:
I don't pretend to understand this but i have feeling it will only make it even harder for those of us with high coach rankings to get games than it already is. Crying or Very sad


I'd say it's the other way around. High strength teams will have easier to find opponents as losing against a relatively equal team doesn't lower you as much as before. If you minmax, you want to play as few games as possible where the ranking system overestimates your chances of winning. This will make it more worthwhile to play against slightly higher teams at high strength values.

It will also reduce the effectiveness of playing games which the ranking system underestimates the probability of winning. Overall, I think this is a good change for alot of reasons.
Mully



Joined: Aug 02, 2003

Post   Posted: Apr 21, 2004 - 23:06 Reply with quote Back to top

Quote:

There are other people who strongly believe that strength should be taken out from the ranking formula entirely and leave it with comparing coach rankings only, just like they do in Chess, Magic the Gathering, Othello, etc etc.


HUGE difference between Bloodbowl and those games. In Chess and Othello both coaches start off with EQUAL sides. In bloodbowl you do not. Beating a STR200 team will always be tougher than beating a STR170 team no matter who the coach is.

A coach that beats an equal STR team would gain in ranking the same if he beat a team 20 STR points lower than him? Change it to that and the ranking whores would NEVER play an equal STR game.

Quote:

The reasoning behind this change is that str 100 vs 140 is more one-sided than str 200 vs 240. This change reflects that.


I AGREE with that concept totally and now understand your point. So I guess in the end where we disagree is I feel the dt constant should be much higher than 100. (I think the game matchup I showed in the previous post shows that point).

Look at the following scenarios with a 25 pt coach ranking difference and see which one <i>feels like</i> it makes more sense (I realize the scenarios are VERY general).

The following would be considered 50/50 games at various dts

........................................................dt100.......dt150........dt200
Coach A.....Rank 180.... Team STR......100..........100..........100
Coach B.....Rank 155.....Team STR......112..........108..........106


The following would be considered 50/50 games at various dts

........................................................dt100.......dt150........dt200
Coach A.....Rank 180.... Team STR......150..........150..........150
Coach B.....Rank 155.....Team STR......167..........161..........159


The following would be considered 50/50 games at various dts

........................................................dt100.......dt150........dt200
Coach A.....Rank 180.... Team STR......200..........200..........200
Coach B.....Rank 155.....Team STR......223..........215..........211

To me, it feels that dt100 put's too much weight on the coach ranking and a number between 150-200 would make more sense.

<b> FYI : With a dt of 175 the Skaven-Necro game above would yield a 40% chance of you winning which, to me, sounds pretty accurate.</b>



Thanks , as always

_________________
Owner of the REAL Larson
Come join the CCC League
Christer



Joined: Aug 02, 2003

Post   Posted: Apr 22, 2004 - 00:16
FUMBBL Staff
Reply with quote Back to top

I've said it before, and I'll say it again: I'm always open for constructive and insightful ideas on how to improve the formula. I have never said it's perfect and I would never say that's inherently impossible.

However, I DO listen to people who have comments and I have been known to update the formula after input from coaches.

As for the "No Str influence" suggestion, it _would_ work in Blood Bowl if the game is set up as tournaments where you don't have a way to choose your opponents. This is obviously not the case in FUMBBL and as such I don't think it's a reasonable change.

I don't really see what you mean with the dT100 - 200 in your tables but I'm guessing it's the factor 100 in the normalizing formula. The funny thing about ELO type ranking systems (which this is) is that they're self-regulating. The actual constants that are used in the formula (100/70 and 1/150 for CR) are really irrelevant in the long run. The only thing that happens when you change them is that you change the rate at which people reach their "actual ranking" so to speak.

Having said that, I have no problem changing the multipliers to make it more reasonable. The factors were taken straight from the NAF ranking formula and I have no idea where they got them. I'm guessing they were just picked out of thin air... Either way, if you believe it's necessary to change the denominator of the strength part of the formula, feel free to suggest a reasonable value for it. I'd rather keep the CR denominator at 150 as it looks nice compared to the initial 150.0 ranking.

Finally, please avoid the "Boo, you suck.. This change is useless.. Scrap the whole thing!" attitude (that's what most your posts in this matter sound like to me). It just makes me less likely to take anything in the post seriously. Again, I'm willing to listen to ideas but there's a limit to the amount of crap I will take before walking away.
Mully



Joined: Aug 02, 2003

Post   Posted: Apr 22, 2004 - 00:40 Reply with quote Back to top

Quote:

I don't really see what you mean with the dT100 - 200 in your tables but I'm guessing it's the factor 100 in the normalizing formula


yes dt = 100 * [whatever] > The "100" number is the one that seems off. I am not sure what the "range" of coach rankings in the NAF is, but that could be a reason why "100" makes more sense for that league than FUMBBL. My guess (based on that) is their coach rankings may be grouped in a much closer range than ours.

Quote:

if you believe it's necessary to change the denominator of the strength part of the formula, feel free to suggest a reasonable value for it


Based on this statement I think you may have misinterpreted my points above. I have no problem with the "dr" or "dt" <b>denominators</b>. I think they work fine. My suggestion is to change the "dt" numerator.

Based on the examples in previous posts my suggestion for the dt calculation would look like this :
Quote:

dT = 175 * [s1/min(s1,s2) - s2/min(s1,s2)]


I looked at about 30 scenarios and this formula looked like it gave the best 50/50 matchups for various coach and team STR ratings.


Quote:

Finally, please avoid the "Boo, you suck.. This change is useless.. Scrap the whole thing!" attitude (that's what most your posts in this matter sound like to me).


Sorry - you wouldn't be able to tell it from my posts but I HATE to write. Thus I usually get right to the point without sugarcoating or thoughts about how the recipient might interpret it. But be known that I a have a great appreciation for what you do here and am in awe of your mathematical and computer expertise. If you ever make it over to Cincinnati I'll treat ya to free beer for your troubles.

_________________
Owner of the REAL Larson
Come join the CCC League
Christer



Joined: Aug 02, 2003

Post   Posted: Apr 22, 2004 - 00:58
FUMBBL Staff
Reply with quote Back to top

Mully wrote:

yes dt = 100 * [whatever] > The "100" number is the one that seems off. I am not sure what the "range" of coach rankings in the NAF is, but that could be a reason why "100" makes more sense for that league than FUMBBL. My guess (based on that) is their coach rankings may be grouped in a much closer range than ours.


Actually, NAF uses a straight TR difference. Most tournaments are TR100 based, so most games are played with this TR part = 0, making coach ranking the only thing that matters. They also have separate rankings for each race (No, I don't want that in FUMBBL for technical reasons). They also have higher K values for their tournaments (K = 2*sqrt(N) where N is the number of NAF coaches participating).

The variable K is something I am looking into for official R based tournaments.

Mully wrote:

iBased on this statement I think you may have misinterpreted my points above. I have no problem with the "dr" or "dt" <b>denominators</b>. I think they work fine. My suggestion is to change the "dt" numerator.

Based on the examples in previous posts my suggestion for the dt calculation would look like this :
Quote:

dT = 175 * [s1/min(s1,s2) - s2/min(s1,s2)]



175/70 = 100/40. It makes no difference really. I just like the 100 nominator to make it slightly more readable.

That would mean that 1 difference in str would equal roughly 3 points in CR meaning that a CR 200 coach would need to play up around 7 strength against someone at CR180.

The consequence of a change like this would be that any inaccuracies in the strength formula would have a greater effect and therefore more prone to abuse. How to fix these inaccuracies is a quite different (and huge) topic.

Mully wrote:

Sorry - you wouldn't be able to tell it from my posts but I HATE to write. Thus I usually get right to the point without sugarcoating or thoughts about how the recipient might interpret it. But be known that I a have a great appreciation for what you do here and am in awe of your mathematical and computer expertise. If you ever make it over to Cincinnati I'll treat ya to free beer for your troubles.


Fair enough. I'll try to remember that in the future. However, it never hurts to have a slightly more positive introduction to your posts.. Smile

Oh, and I'll have to turn that beer down.. Can't stand the stuff Smile
Mully



Joined: Aug 02, 2003

Post   Posted: Apr 22, 2004 - 05:20 Reply with quote Back to top

Quote:

175/70 = 100/40. It makes no difference really. I just like the 100 nominator to make it slightly more readable


Ya - I've looked at around 30 scenarios of coach and team ratings and that equation yields what looks to be the most realistic win%s

Quote:

The consequence of a change like this would be that any inaccuracies in the strength formula would have a greater effect and therefore more prone to abuse. How to fix these inaccuracies is a quite different (and huge) topic.


Ya - but we need to assume the STR formula is accurate (or at least as accurate as it can be). And the innaccuracies become smaller and smaller above STR200 which is where (I think) some of the STR formula flaws are more exposed.

Mully

_________________
Owner of the REAL Larson
Come join the CCC League
Pmg



Joined: Nov 24, 2003

Post   Posted: Oct 10, 2004 - 04:08 Reply with quote Back to top

Now that everyone has played a lot more games, I have two things to say about the ranking system.

First it creates like an invisible barrier around 170 ranking that rquires a ridiculous number of games to surpass in normal open season. Games agaisnt equal rated players gives you a slightly less than decent increase, and the same with playing higher rated teams, but still you need around 15 games to recover from a loss against a lower rated player, and for someone who actually enjoys playing bloodbowl, only acepting games agaisnt high rated coaches with stronger teams and spending 3 hours waiting to find a game like that is quite boring.

Second, the ovious way to overcome this is with oficial fumbbl tournaments with higher K ranking value, that not only randomize the matches, but give a true feeling of victory in bloodbowl. What brings me to my final question what happened to those? If i recall correctly, we were promised those tournaments more often , both to create some adrenaline and to give some emotion to the top of the ranking system, than by now is beyond boring, beeing dominated by mix of interested players with loads of games and a couple of guys that were lucky in the two oficial cups a couple of monts ago.
Csonti



Joined: Aug 02, 2003

Post   Posted: Oct 30, 2004 - 00:34 Reply with quote Back to top

Pmg wrote:
First it creates like an invisible barrier around 170 ranking that rquires a ridiculous number of games to surpass in normal open season.


Well, i think this is just the greatness of the new system, not it's flaw. Every ranking system has a theoretical (but quite precisely estimated) limit where even the best must say: i couldn't do better. In the previous ranked system this was around 240 or so, and u needed tons of matches to get there (90 points from 150). As u mentioned, in the new system, the limit is around 170 or maybe a bit more, which is only 20-25 points from the start, so within 50-60 good matches u could be in the top. What's the problem with that?

And if u wanna play vs high ranked coaches, I'm more than glad to match with u in an even fight. Wink (I'm annoyed that high ranked coaches don't want to play vs each other altough the ranking system just encourages that - experienced it far too many times Sad )

On the other hand I can just agree with u about the official tournaments. Where are they?
Azurus



Joined: Aug 02, 2003

Post   Posted: Oct 30, 2004 - 00:39 Reply with quote Back to top

Csonti wrote:
On the other hand I can just agree with u about the official tournaments. Where are they?


Evo posted a bit about them in this thread.

http://fumbbl.com/index.php?name=PNphpBB2&file=viewtopic&t=3643

looks like we're mostly waiting for updated code to get them going again.
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