15 coaches online • Server time: 05:41
Index Search Usergroups Profile
Log in
Recent Forum Topics goto Post UW Goblins - Disturb...goto Post SWL Fantasy League S...goto Post Is kicking first wit...
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 21, 2017 - 16:39
FUMBBL Staff
Reply with quote Back to top

I have done a revision of the first post detailing the exact formulas used in the current iteration of the CR system, which is likely to remain as-is. I am happy with the current state.

I also did the write-up of how the brackets function, now that I have updated them to use a standard deviation based distribution rather than forcing a set number of coaches into each bracket.

There's a lot to read, and it's math heavy. You have been warned Smile
Rags



Joined: Nov 09, 2008

Post   Posted: Oct 21, 2017 - 16:50 Reply with quote Back to top

MattDakka wrote:
Probably a coach playing vs opponent coaches of comparable skill with average dice would not try/risk/be able to win by large margins.


Exactly. The very reason why rating should be higher for those that are confident/dare/succeed.
MattDakka



Joined: Oct 09, 2007

Post   Posted: Oct 21, 2017 - 17:27 Reply with quote Back to top

Then the rating would put an emphasy on the will of taking risks and on teams with one turners/stat freaks, not necessarily on higher coach's skill.
Good coaches are able to score a lot, in theory (well, assuming they are using a fast enough race), they just don't do it because trying to score a lot increases the randomness of the match.
Good coaches want to control the match as much as they can, this is why they stall.
When a good coach is sure enough of having the win in the bag (with a good TD difference or numbers' advantage) then he may score a lot, but at that point the game is already decided, so winning 2-0 or 4-0 won't make a meaningful difference in terms of coach's skill.

_________________
Please vote the Diving Catch and Break Tackle/Diving Tackle bug reports, thanks!
Rawlf



Joined: Jul 15, 2007

Post   Posted: Oct 24, 2017 - 11:50 Reply with quote Back to top

Thank you very much, Christer, for spending so much time on once more improving your fantastic site for all of us! And if you are happy with the current result, so am I.

Without being demanding or ungrateful in any way, I would still like to add some of my thoughts, because some details of the new CR leave me with a worrying feeling.

I read the bit about leaving out the racial factor from calculating the win probablity of a match a few times now, but I still cannot really understand it.

Christer wrote:
The intent of the racial factor was to build a system that would map the win rates of the respective races versus other races. [...] The race ratings were very very close to 50% across the board. Almost no variation at all.


It sounds as if you are saying: TV1100 Dwarves vs TV1100 'Flings, TV1400 Wood Elves vs TV1200 Ogres, TV1000 Amazons vs TV1000 Dwarves - is all pretty much fifty-fifty. But surely that is not the case?! Wood Elves have win rates around 65% across all TVs and vs all races, Ogres have maybe 30%.
I have a very strong suspicion I thoroughly misunderstood something here, and I would appreciate very much if you could help me out.

Ignoring races for calculating the win probablity of a match seems like ignoring what kind of car a driver brings to a car's race. But everyone in a Ferrari can lap others in their FIAT Cinqucentos.

Christer wrote:
This meant that the overall win probability of the CR system was being pulled closer to 50% than felt correct.


Isn't this more a question of how you include the racial factor? If you did it in a multiplicative way, like maybe: p_adj = (p_cr + 0,5) * p_race, a racial factor of 50% would not influence p_adj at all.

Why I am posting this is that i feel very awkward about leaving out the (imo) most important factor of predicting a games outcome (luck and coach skill aside) with 'races used' (as a proxy for a more individual factor like 'team build' or even 'TS').

And at the same time, TV gets a lot of attention, while in my experience, this is a very meaningless factor for predicting the outcome of a match. Using 'TV before inducements' as a factor for win probablilities is assuming that inducements close the TV gap by 0%, no matter how you spend them. While this would have been a good guess in the LRB4 handicap system, I think it vastly underestimates the power of inducements. They can be tailored to beat your opponent, like get a saw, dp or tackle mercs vs Elves or babes vs Norse. Not to mention the cheap bribes/chef for the stunties. Or overdog inducings. Back when there was the wizard, a negativ TV gap > 150k would very often turn the underdog into the actual favourite as inducement value was much higher than 100% of their cost in TV.

If anything, using TV after inducements (and hence rating their value at 100% of their cost in TV) would be a lot closer to the truth imo.
Even leaving TV out completely might be favourable in my mind, crazy as it sounds. (Reasoning that the TV gap after inducements is mostly near zero anyway.)

I hope I did not come across as complaining or unhappy. I am really just interested.
koadah



Joined: Mar 30, 2005

Post   Posted: Oct 24, 2017 - 12:47 Reply with quote Back to top

Rawlf wrote:

If anything, using TV after inducements (and hence rating their value at 100% of their cost in TV) would be a lot closer to the truth imo.
Even leaving TV out completely might be favourable in my mind, crazy as it sounds. (Reasoning that the TV gap after inducements is mostly near zero anyway.)


The team with the bigger TV tends to win more often than they lose. That is something that is not difficult to calculate. Worrying about the inducements is not necessary.

Extreme flings vs dwarves type matchups are still may be a point though.

_________________
Image
Secret League - SLUMBBL Shield - 16th November!
Christer



Joined: Aug 02, 2003

Post   Posted: Oct 24, 2017 - 17:19
FUMBBL Staff
Reply with quote Back to top

The last time I did a calculation of win rates based on TV differences, I more or less ended up with a 55% win rate for the higher team at anything above trivial differences.
Rawlf



Joined: Jul 15, 2007

Post   Posted: Oct 25, 2017 - 12:16 Reply with quote Back to top

Looking at http://fumbbldata.azurewebsites.net/stats2.html global win rates seem to range from about 60% (Lizards) to 25% (Ogres), disregarding CR, TV etc.
For each race individually the win rates vs other races seem to vary roughly from 80% to 30%, with some outsider matchups being even crazier (WE vs Ogres 97%).
I don't know which games are compiled at that site; the sample size isn't huge. And i have no idea how these numbers relate to Christers 'very very close to 50% across the board'.
But it seems very likely that the races have a stronger effect on a games outcome than TV.

My worry is that the CR might be deluding by overlaying how strong a coach is with how well he choses successful races. Turning the 'Coach Rating' into a 'Choice Rating' so to say. And it might also present a strong incentive for coaches to leave unsuccessful races unplayed.
Dominik



Joined: Oct 29, 2004

Post   Posted: Oct 25, 2017 - 13:54 Reply with quote Back to top

I am a top 30 coach but mainly coach top races. I dont think that I deserved that. Many coaches do outstanding things with races like Goblin, Undead (on a TV level of above 1800) and so on and dont get the fame for it.
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