MattDakka
![](./modules/PNphpBB2/images/avatars/upload/7018948714f7352955a174.gif)
Joined: Oct 09, 2007
|
  Posted:
Oct 20, 2017 - 14:21 |
|
PurpleChest wrote: |
Why not decide that scoring in the central 4 squares of the td zone displays 'positional/field mastery' and award bonus points for that? there are plenty of sports scoring systems as precedent. |
I use to score in the central square of the End Zone, whenever I can. ![Razz](./modules/PNphpBB2/images/smiles/icon_razz.gif) |
|
|
thoralf
Joined: Mar 06, 2008
|
  Posted:
Oct 20, 2017 - 17:05 |
|
Grod wrote: | Stall! Don't be nice to your to your opponent, you are now penalised for it. |
That kind of argument doesn't change with a CR that takes number of TDs into account, so the "now" is unjustified.
Any R player could attest that being nice is its own reward. |
_________________ There is always Sneaky Git. |
|
happygrue
![](./modules/PNphpBB2/images/avatars/upload/92775368258a791da06371.png)
Joined: Oct 15, 2010
|
  Posted:
Oct 20, 2017 - 18:35 |
|
Christer wrote: |
The point here is that the rating system does its best to indicate what level you are playing at, and not necessarily what level of coach you are. It also inherently puts more weight into recent results than older ones, which makes sense as the option would be that the system would assume you couldn't improve as a player.
In addition to that, another thing many people overlook when talking about the CR system is that the design of it will average many many games over time. With more games played, the inherent randomness of the dice will have a lower and lower significance to the end result. Sure, a single match can be lost or won purely because of dice rolls. Even two or three. However, after 10 matches, you've simply rolled too many times for the statistical distribution of them being anything but average. And at that point, it's your ability to utilize the good dice rolls well, and reduce the negative effects of the bad ones. Thus, over time, the rating system will converge your rating towards the rough place indicating what level you play at. The individual luck and bad luck streaks will make you move up and down the rating somewhat but in the end everyone has the same dice with an equal amount of lucky and unlucky streaks. |
This (emphasis mine) gets at the heart of my interest in the rating system. I am curious why the system is designed to favor current trends rather than the larger body of coaching experience/skill? What is the rational for trying to move CR around a lot after upsets if one of the coaches has thousands of games and is likely quite accurately ranked for the majority of the matches they are likely to play?
I can see advantages, as well as disadvantages, of this approach. I am curious as to your thoughts on why lean toward faster CR movement that risks volatility over slower, more stable movement? Personally, most of what I use CR for is to quickly get a read on my opponent's level of skill. I'm much more interested in their overall level of ability than "what they've done recently". That's especially true in tournaments. I can see how it might be more predictive from a statistical point of view by taking into account recent trends, but I was just curious about your own thoughts on the matter. |
_________________ Come join us in #metabox, the Discord channel for HLP, ARR, and E.L.F. in the box!
![Image Image](https://fumbbl.com/i/465979) |
|
JackassRampant
Joined: Feb 26, 2011
|
  Posted:
Oct 20, 2017 - 19:07 |
|
Well, part of it is that coaches grow and develop over time. So if CR doesn't change quickly (like, say, win% doesn't), it doesn't take into account recent coach development. |
_________________ Lude enixe, obliviscatur timor. |
|
licker
![](./modules/PNphpBB2/images/avatars/upload/166509707555ef89b25b521.jpg)
Joined: Jul 10, 2009
|
  Posted:
Oct 20, 2017 - 19:11 |
|
Should probably have 3 different levels of CR.
Global CR
Racial CR
Team CR
But really, I'm on the 'who cares' about CR band wagon anyway. At least R CR ![Wink](./modules/PNphpBB2/images/smiles/icon_wink.gif) |
|
|
Sp00keh
![](./modules/PNphpBB2/images/avatars/upload/15329720395a2c5d2287897.png)
Joined: Dec 06, 2011
|
  Posted:
Oct 20, 2017 - 19:22 |
|
Also, the actual players on a team can affect outcomes, and if you lose key player your performance could go down
Fast CR changes would include that variation by updating fast enough
If you want overall indication, there's winrates available for coaches |
|
|
JackassRampant
Joined: Feb 26, 2011
|
  Posted:
Oct 20, 2017 - 19:31 |
|
You can match CR to win rate. If they don't jive, there's a reason. Fortunately, when you encounter a coach whose CR isn't up to his win rate, there's the CR history graph, which will tell you if you're playing a good coach coming off a couple bad games, or if you're up against a dirty picker. |
_________________ Lude enixe, obliviscatur timor. |
|
koadah
![](./modules/PNphpBB2/images/avatars/upload/11621913544284994cadd9f.gif)
Joined: Mar 30, 2005
|
  Posted:
Oct 20, 2017 - 20:10 |
|
|
mrt1212
![](./modules/PNphpBB2/images/avatars/upload/20457904475440aaf205319.jpg)
Joined: Feb 26, 2013
|
  Posted:
Oct 20, 2017 - 20:30 |
|
koadah wrote: | We used to get top coaches trying to sign up to 145 Club when their CR hit 145. (not many)
How well does the system deal with people going on long runs with no tree flings, or all skellie/zombie teams? |
Probably not well at all.
I've done a few 0 RR theme teams in Box and I wonder how many other coaches would be able to deal with that kind of restriction long term.
There is not 'pause' button for when you're goofing around in Box and Ranked though. |
|
|
JackassRampant
Joined: Feb 26, 2011
|
  Posted:
Oct 20, 2017 - 20:37 |
|
I use Ranked Amateurs to "cull" my CR without playing less competitively. I found, at least under the old system, that I'm good enough to earn a Legend tag playing top races in the ways they were meant to be played, but not good enough to earn one with a hand behind my back. Which is fine by me, it's easier to get games as a Super Star than as a Legend. Looking back at my CR over time, this philosophy seems to hold under the new system too. |
_________________ Lude enixe, obliviscatur timor. |
|
koadah
![](./modules/PNphpBB2/images/avatars/upload/11621913544284994cadd9f.gif)
Joined: Mar 30, 2005
|
  Posted:
Oct 20, 2017 - 20:47 |
|
|
JackassRampant
Joined: Feb 26, 2011
|
  Posted:
Oct 20, 2017 - 20:50 |
|
Yeah. I've noticed that. Too many? How's that work, Christer? |
_________________ Lude enixe, obliviscatur timor. |
|
Christer
![](./modules/PNphpBB2/images/avatars/upload/212093185457642fe6c6ee4.jpg)
Joined: Aug 02, 2003
|
happygrue wrote: | I am curious why the system is designed to favor current trends rather than the larger body of coaching experience/skill? |
Great question. I've started to reply to this a couple of times only to delete what I wrote to start over. The thing is: When I come up with designs for things like this, I spend a lot of time on researching the topic and compiling concepts from other pre-existing ways of solving the problem.
Coming up with the rating system we use here was something I did very very long ago. It obviously uses the Elo system as its foundation, but with some modifications to account for TV differences. A typical Elo implementation (e.g. chess) uses all matches for a tournament as its basis for rating changes (more or less changing S to the actual win rate of the matches played during a tournament). Doing so more or less reduces some of the volatility of the calculation and reduces the effect of "luck". Thus, they usually use a relatively high k value (10-40 seems to cover most common values). With a per-match rating change, I felt it would be better to go relatively low. Wikipedia states this:
Wikipedia wrote: | If the K-factor coefficient is set too large, there will be too much sensitivity to just a few, recent events, in terms of a large number of points exchanged in each game. Too low a K-value, and the sensitivity will be minimal, and the system will not respond quickly enough to changes in a player's actual level of performance. |
To come up with a number for the rating system we use on the site, I mostly did a bunch of iterations with different k values to see what produced a reasonable overall distribution of ratings. "Reasonable" in this context is pretty much me looking at a graph of the CR distribution and considering the overall shape of where people are located on the rating and looking at the min and max values. Inherently, a higher k value spreads out the ratings more, while a lower k value makes the ratings narrower.
In the end, it comes down to finding something that "feels right". I don't always spend too much time dwelling on the whys and why nots of this. If it feels right to me, it often pretty much is. The same process follows with the revision of the formula I just completed.
Thus, we arrive at your question. Actually taking the time to consider the fundamentals of the rating system, my primary drivers for keeping it as volatile as it is are:
- It's more fun if it moves around. Having a number that takes forever to move makes the system feel very grindy. To me, it is more rewarding to know that "it takes 2 wins for me to pass this other guy" instead of having to play 25 matches to catch up.
- If I want to make it move slower, I could calculate a rolling average of the last X games and show that number instead (and yes, I realize there could be strange effects by doing so). That, as Wikipedia states, does slow down how fast the system reacts to actual changes in skill level (or current level of play). It also reduces precision to an extent. Going from the inherently slower rating changes to something with more volatility and responsivity would be considerable harder.
- With a fair percentage of people playing multiple teams, and probably varying what they do every now and again, it makes sense to have a system that is able to respond relatively quickly to these changes. Most people have a favoured style of play, and moving out of that "safe zone" will inherently make their level of play go lower; at least in the short term. If the rating system reacted to this too slowly, more opponents would get "unfair" changes in rating when playing these types of games.
And yeah, I've been writing long posts lately ![Razz](./modules/PNphpBB2/images/smiles/icon_razz.gif) |
|
|
Christer
![](./modules/PNphpBB2/images/avatars/upload/212093185457642fe6c6ee4.jpg)
Joined: Aug 02, 2003
|
koadah wrote: | There are a lot of Super stars. You can be Super Star with only around 154.
I'd have thought you'd need nearer 160. |
The brackets. Ah, yes. There's a reason I have a reserved post at the top of the thread for this..
The short version is that they're currently set up to be X number of legends in a bracket, and the rest of the players put linearly into each of the other 5 brackets.
It's not a good way though, because the middle brackets are very very narrow in terms of rating (like 2 CR), whereas the super star one is extremely wide (like 12 CR).
I need to go through and make the CR bands based on the sigma (standard deviation) of the CRs instead. I'll let you figure out the details of what that means to keep the post short. |
|
|
JackassRampant
Joined: Feb 26, 2011
|
  Posted:
Oct 20, 2017 - 21:09 |
|
Using current “all divisions†CR:
175+: 33 coaches
170-174.99: 87
165-169.99: 119
160-164.99: 131
155-159.99: 223
150-154.99: 525
145-149.99: 492
140-144.99: 103
143.99-: 29
Hmmm. 87+119 ~ 200. If there are about 33 legends at 175+ CR, then 165 sounds like a good threshold for super stars, at least at the moment. And Super Star means "top 10% of active non-legend coaches" or "top 13.5% of coaches, minus the number of legends" in that case.
Hmmm.
2% legends... (33 coaches; 175.26 min)
and then of the remaining 98%:
10% super stars (171; 166.32)
20% stars (342; 155.9)
40% emerging stars (684; 149.26)
20% veterans
10% experienced |
_________________ Lude enixe, obliviscatur timor.
Last edited by JackassRampant on %b %20, %2017 - %21:%Oct; edited 1 time in total |
|
|
| |