Averaged Asymptotical Scoring System for LLC2007

[Vertigo]

"Averaged Asymptotical Scoring" could be completely wrong as name (Im no math buff), but it sounds way too cool, so I'm sticking with it . Maybe even better: "Averaged Asymptotical Scoring with Hospitalization Handicap"

Anyway, despite the name, its utterly simple and elegant I think. Just read the red/bold part below, and there you have it. Bare with me:

Requirements of the scoring:

- pilots should have the ability to fly as frequently as they can/want, each result should matter, but flying every single race rather than the large majority of races shouldn’t give an unfair advantage. OTOH, flying only a tiny fraction of the races shouldn’t give you reasonable chances to win either.
- no situations should occur where a pilot’s best option would be to not fly.
- Taking risks and gambling should not be rewarded, crashes should always be penalized, landing out safely instead should be rewarded
- reflect FAI scoring, but accommodate the fact that unlike RL almost no one will fly all races, and big differences between number of raced days will exist.

Proposed Solution: Keep day scores (‘1000â€

[Panther]

Hi Vertigo,

a very interesting suggestion.


The big advantage of the 'asymptotical' system im my opinion is that it will always give an advantage for outlanding compared to crashing and that it will be always beneficial to participate and not to take too much risks (very much more than in real life). The crashing penalty is a very good proposal, I'd even make it somewhat harder by adding the 'hospital day' factor to the divison factor. For the example used above this would mean:

Pilot flies 4 races, 900 points on average: 3600/5= 720 points
Fifth day he crashes -> 0 points so he goes down to 3600/(6+1 punishment )= 514 points
Sixth race is not counted (hospital) so still: 3600/7 = 514 points


The biggest weakness compared to the old system in my opinion is that
frequent fliers will always have at least a small advantage above those who cannot participate all races. Example: two pilots who participate very often and are quite good (lets say average is 900 pts): If one of them flies just one race more (lets say 8 races) while the other one misses one race more due to 'private time restrictions', the one with the race more will win:

900 * 8 / 9 = 800
900 * 7 / 8 = 787,5

With the old system, they would have equal scores (which is in my opinion fairer to those who cannot participate so often).
The difference is small, but always present.
To make this a bit more clearly: even if one of our above pilots beats the other one clearly in two races (lets say one he gets 950, other one 930 points while the other one stays at 900 pts), the one who flew one race more would still win:

900 * 8 / 9 = 800
(900 * 5 + 950 + 930) / 8 = 797,5

Although there is just a small difference in resulting points to be seen, a pilot who joins more races will always have it much easier to win against a pilot who is better, but only misses a few races.
If you compare pilots who fly e.g. 4 vs. 5 races per month, the difference is even bigger - even if the one with 4 races beats the other one every race, he will lose: e.g. the one with 4 races gets 930 pts every race, the other one 900 pts:

930 * 4 / 5 = 744
900 * 5 / 6 = 750 -> winner, although he was always behind and participated only one race more (which in my opinion should weight less than their difference in 'average points per race').

This is a very significant imbalance in my opinion!



Weighting out advantages and disadvantages, in terms of scoring equity I'd prefer the old system (50%) with a good crash penalty method - as you described in the previous thread. For me as someone who can participate max. about 50% of the tasks, this would grant better scores.

But the score is for me actually not the most important thing: I like Condor and LLC for being similar to real-world soaring, and therefore I clearly prefer the 'asymptotic' system because of its potential to make Condor more like the real thing!

It would only be nice if there would be a way not to give up the advantages of this system, but to eliminate the 'frequent-flyer-bonus'.

[Vertigo]

Hi Panther,

I guess it depends what you call "extreme" in extremely frequent. If you can only fly half the races while allowing others to fly all of them AND somhow eliminating the "take risks to win" factor of dismissing bad results, there are no good solutions, at least none that I can think off.

Still, the problem isnt so bad I believe. Some more examples: say you can only fly 4 races, but you are pretty good, scoring 950 points average:
4x950/5=760

SWT and JJJ fly almost each race, and for 7 races getting the same score requires them to average 869:
7x869/8=~760

That may seem unfair, but do consider this as well: when flying that many races, there is also a larger chance of them messing up a race, or even landing out. Say JJJ generally flies as well as you, scoring 950 on average, seven races on a row, but 8th raceday he lands out and scores just 180 points:

(7x950 + 180) /9= 758

Bingo, you won!

IMO, not landing out (let alone crashing) will be much more important with this scoring, than flying all 8 races; although I agree you should be aiming at 5 or 6 races per month

[EA]

@ Vertigo :

Though I'm in favor of the day factor you just cannot use it with your system, because of the average ! Example...

Pilot 1 flies three races. First and second have a day factor = 1, and 0.5 for third one. He wins all the races, and so his score will be :

(1000+1000+500) / 4 = 625

Pilot 2 flies only the two first races and then the fourth one (all day factor 1), which Pilot 1 didn't have time to fly, finishing with 950, 950, and then a bad result, let's say 600 pts. He'll get (950+950+600)/4 = 625 pts, though ALL his results were below those of Pilot 1 !

So, you just cannot use dayfactor when averaging. Each race must here have the same weight in the average. Day factor is made for scoring systems were you ADD results.

Seb

[Vertigo]

EA, I totally agree, but FWIW your logic also applies to the LLC 2006 scoring (which is was I was in favor of abolishing dayfactor there too).

In LLC 2006, I could fly first 4 races of the month (DF=1), winning all, and getting 4000 points. You could win the next 4 (DF=0,5) and have only 2000 points. Makes no difference if you add or average scores, dayfactor makes no sense if you dont use the same days for each pilot.

That said, dayfactors where usually between 1 and 0,95 so the difference wasnt all that big.

Still, I suspect dayfactor will be a thing of the past in LLC 07 (speed/distance points will still apply)

[yvesf2]

Excellent proposal.
I agree with the double penalty for the crash (0 for the day and 1 day in the hospital) - In RL it's "game over" even if you survive unhurt...
I would suggest also to have a minimum task flown of "let's say" 15 on the possible 50 because, as you noticed above, the risk you take by flying every task is in fact much greater than saying "I won the first 10 then I stop hoping the others will make an outlanding that day.
The task are not similar at all (like in RL) and you could have the tendancy not to fly a specific day if the wx is marginal and the risk of outlanding - or crashing with very high turbulence - is high.
All pilots should have flown a certain percentage of these difficult tasks where you can loose a lot.
I do not know how you could handle this? Maybe by giving a rating to the tasks? 10 or 15 would be considered as "difficult" and you have to fly 5 of them minimum?
Anyway thanks to the LLC team, you are doing a great job

Yves

[Vertigo]

I would suggest also to have a minimum task flown of "let's say" 15 on the possible 50 because, as you noticed above, the risk you take by flying every task is in fact much greater than saying "I won the first 10 then I stop hoping the others will make an outlanding that day.

What I proposed above was again using month place scores, so you'd need to get a good score each month to win. To get a good score each month, you need more than 1 or 2 flights (one will give you 500p at best, 2 will give you only 666 where 800+ is possible). Realistically you will need 4-5 good flights and no bad ones to be able to win a month to make up for the "zero flight" everyone get (the +1 in the formula).

If you abolish the month place score, and make the overall score based on all 50 flights instead, the same thing would apply with 5 or 10 zero score flight, so you will need about 30-40 flights to make that up and get a representative average.

The task are not similar at all (like in RL) and you could have the tendancy not to fly a specific day if the wx is marginal and the risk of outlanding - or crashing with very high turbulence - is high.
All pilots should have flown a certain percentage of these difficult tasks where you can loose a lot.
I do not know how you could handle this? Maybe by giving a rating to the tasks? 10 or 15 would be considered as "difficult" and you have to fly 5 of them minimum?

I think speed/distance points already takes that into account. On a difficult/slow day, you will score relatively high even with an outlanding if you achieved a long distance. If no one finishes you might even get 1000p for an outlanding halfway if we abolish dayfactor. High turbulence or other difficulties shouldnt really increase the change of crashing though, not if you fly safely. Task should only really increase the chance of landing out early which is already being accounted for.

[Panther]

Hi Vertigo,

when flying that many races, there is also a larger chance of them messing up a race, or even landing out.

I disagree on this one. If I fly 4 instead of 8 races, I have only half the probability to mess up. But if I mess up, it will count twice as much. In LLC, counted for the whole cup this means that the 'expectancy value' (derived from the german word, I don't know the correct english expression) is the same.
Expectancy value as I use it is the product of
Probability * Value of Damage.

But nevertheless, I already said that I prefer the asymptotic method because of its benefits although it will always favour those who fly more often as it is proposed now. It would be great if something could be done about that, but nevertheless, the scoring system which grants the 'most realistic behaviour' wins for me!

[Vertigo]

Hi Vertigo,
But nevertheless, I already said that I prefer the asymptotic method because of its benefits although it will always favour those who fly more often as it is proposed now. It would be great if something could be done about that, but nevertheless, the scoring system which grants the 'most realistic behaviour' wins for me!

Not true, it depends up the +1 factor which you can varry. I picked 1 per month because I think it makes sense, but for arguments sake, lets put this at 0,01 (or even 0). What you get is that you can easily win a month with just 1 race (all you need to do is win it).
1000x1 / 1+0,001= 999,9

You'd be nuts flying another race then if you have 999.9 average.

THe other extreme is setting this to 10 per month, in which case you really can't win if you don't fly all races that month, and having more results will be just as important (if not more) than having good results.

If I understand you correctly, you think 1 "zero flight" per 8 races is too much already.. I might redo some math using 0,8 or so instead later and see what it gives.

[Panther]

If I understand you correctly, you think 1 "zero flight" per 8 races is too much already.. I might redo some math using 0,8 or so instead later and see what it gives.

Good thought.

I made a small calculation, for my two examples the 'break-even' would be around 0.8 (still small advantage for the 'worse pilot with more races'). By using 0.7 or 0.6, the 'better pilot with less races' would win with a very, very small advantage (maybe within rounding tolerances). See numbers for my two examples below (hope the formatting works )

Factor_______________0.8_________0.7________0.6
4*930:______________775,00______791,49_____808,70
5*900:______________775,86______789,47_____803,57

8*900:______________819,09______828,51_____838,14/
7: 5*900 + 930 + 950__818,85______829,48_____840,39

Of course it is still possible to construct examples where the worse pilot wins because he flew more races. But with a factor below 1, the whole system seems to be a lot more balanced to me.

From what I see, 0.7 for a month could be a good compromise.
What do you think?

[EA]

Without DF and with the "zero-flight factor" set as you'll judge appropriate, these scoring rules seem almost perfect imo...

Seb

[Vertigo]

If I understand you correctly, you think 1 "zero flight" per 8 races is too much already.. I might redo some math using 0,8 or so instead later and see what it gives.

Good thought.

I made a small calculation, for my two examples the 'break-even' would be around 0.8 (still small advantage for the 'worse pilot with more races'). By using 0.7 or 0.6, the 'better pilot with less races' would win with a very, very small advantage (maybe within rounding tolerances). See numbers for my two examples below (hope the formatting works )

Factor_______________0.8_________0.7________0.6
4*930:______________775,00______791,49_____808,70
5*900:______________775,86______789,47_____803,57

8*900:______________819,09______828,51_____838,14/
7: 5*900 + 930 + 950__818,85______829,48_____840,39

Of course it is still possible to construct examples where the worse pilot wins because he flew more races. But with a factor below 1, the whole system seems to be a lot more balanced to me.

From what I see, 0.7 for a month could be a good compromise.
What do you think?

Leaning more to 0.8+ to encourage frequent participation, but with we have a few more months to figure that one out Think it will be a lot better than what we had anyway.

[mac]

As posted in the other thread presenting this solution, I found it elegant and effective and I support it fully but for the crashes part.

Still I disagree in general with heavy penalisation of crashes for many reasons:

1. Penalising Crashes means to intervene also on the way disconnections and Software/PC failures have to be considered. The risk is that we make a bit more fair the treatment of crashes making much more unfair the treatment of failures and disconnections.

2. Crashes can be generated by third parties. It happened to me recently that while ridging close to a slope somebody tried to overtake me, crashed into me, and the slope was closer to the glider than my finger to the "Q" key. Limited view, and high pings do contribute to these kind of incidents.

3. Situation in Condor is different than in RL. An example: in RL only qualified pilots are in the air, in Condor whoever got 50 euros for buying a licence is. I think that LLC should aknowledge the broader scope of its public.

I disagree specifically with the solution proposed as I think it's unfair to ask a pilot to run a race that gives no points: time is limited for everybody, let's everybody get the best out of it!
Also, practical issue: if the flight get no points, I would start the task and outland after 5 km... why should I risk a second crash for nothing?
It simply doesn't work logically, IMO.

Alternative proposal, as I however agree 100% that a crash generated by too risk shouldn't count as an outlanding:

Simply count crashes in the average with negative values, and do it incrementally (exponentially, even), so that one or two crashes in the entire year will basically have little weight (this should leave room for a PC failure, somebody smashing you on the rocks...), while systematic crashing will be heavily penalised. Ex: -10, -30, -100, -250, -500... (but I trust LLC in finding out the right figures for that.[/u]

[Vertigo]

1. Crashes in Condor occour in different situations than in RL. An example: in RL only qualified pilots are in the air, in Condor whoever got 50 euros for buying a licence.

Yup, and they are free to fly LLC even with no skill, but I am not going to design a scoring system that will let them win Seriously, if you can't safely land out or avoid mountains, how would you succeed in staying behind the towplane ?? I'm still convinced >95% of the crashes we saw where due to reckless flying, not lack of skill. It doesnt require a lot of skill to see keep a safety margin, it takes skill/luck to fly on the edge of it (which is where it goes wrong).

2. Penalising Crashes means to intervene also on the way disconnections and Software/PC failures have to be considered. The risk is that we make a bit more fair the treatment of crashes making much more unfair the treatment of failures and disconnections.

I disagree. Disconnection will have no impact (continue offline, submit your IGC, and all is well). Only caveat is that we will have to judge from the IGC if someone landed out or crashed, but thats ok (we"ll be liberal).

as for PC crashes, once again:
- pc crash on final glide will be treated as in LLC '06, ie, full points (with IGC or eye witness reports)
- pc crash anywhere else will be scored as outlanded; all that is required is a lasttrack.ftr (converted into IGC) that condor almost always writes since patch 8 (7?), even when it crashes.
- pc crash with no witnesses and no lasttrack, well that is tough luck. Then again, no scoring or rule set could ever make this "fair", if your pc crashes after 10km, and that is after you got disconnected first, just how many points would you expect ?

3. Crashes can be generated by third parties. It happened to me recently that while ridging close to a slope somebody tried to to overtake me, crashed me, and the slope was closer to the glider than my finger to the "Q" key. Limited view, and high pings amongst the more prominent factors can influence that.

Thats why we still have a jury. If something like that happened, send us your IGC, and we"ll decide. If what you describe above is what happened, we'll likely treat it as an outlanding (can't do better than that anyhow), and we might give a warning or penalty to the offending pilot.

Really, this is no different than LLC '06 where we assumed each disconnection to be a crash as well, and the above example would also be zero points until judged.

I disagree specifically with the solution proposed as I think it's unfair to ask a pilot to run a race that gives no points: time is limited for everybody, let's everybody get the best out of it!

He's not obliged to race the day he is in hospital, he can revalidate there as well

Also, practical issue: if the flight get no points, I would start the task and outland after 5 km... why should I risk a second crash for nothing? It simply doesn't work logically, IMO.

Again, you can fly the race that you are hospitalized, we may even give you what-if scoring in the day results, but you don't have to. You will be hospitalized eg "for day 7" (after crashing on day 6), not for "your next race". And yes, if you only fly on sunday's, that means your crash won't be penalized so heavily (only zero points that do count), but you will have less chance to make up for it also.

Simply count crashes in the average with negative values, and do it incrementally (exponentially, even), so that one or two crashes in the entire year will basically have little weight (this should leave room for a PC failure, somebody smashing you on the rocks...), while systematic crashing will be heavily penalised. Ex: -10, -30, -100, -250, -500... (but I trust LLC in finding out the right figures for that.[/u]

Three things: one, if you do that, to some extend you open the door again to risk taking ('I havent crashed yet, so I may as well gamble this time"). Secondly, I did propose something alike: 1 day in hospital first crash, 2 day for the second, etc. Thirdly, don't overestimate the effect of a day in hospital with the AAS scoring. example:
You flew 5 races, average of 800 points (medium good pilot)
5x800/ 5.8 =690 points
Then you crash on day 6:
5x800/6.8= 588 points

crash costs you >100 points here, regardless of hospitalization penalty.

Now, Scenerio 1: hospitalisation:
day 7 is not counted (hospital), day 8 you have family reunion, your stuck with 588p

Scenerio 2: no hospitalisation:
day 7 you score 800 again as usual, day 8 you have family reunion:
6x800/7.8=615 points

As you see, in this case this penalty only adds another 28 points over the 100 you'd have anyway. Maybe you thought we'd add the "hospital" day to the number of flights ? Wasn't my intention, and doesnt make sense since you didnt fly it officially, hospitalization is only 1 less chance to improve your average that month.

[TimKuijpers]

Ok, before I went to the city/pub I tried to make a post...
But unfortunately it (or I) didn't got (it) on the forum.

As I drank a few beers I'll summerise (how do you spell that?) the post I ment to post...
First point was my complements of thinking out something like that.
I'm really looking forward to try a scoring system like that.

The only disadvantage will be the days that (still) don't count.
That way taking risks on the last races can only result in advantage by flying fast but if you land out it doesn't hurt either.

I do think that those last flights should count, else it will make the last days too easy or too difficult.
But it still should exclude the risk taking factor.
I doubt if this is possible.
But as it is written in the opening post of this thread I think it will be at least twice as good as LLC 2006.