68 coaches online • Server time: 22:52
Forum Chat
Log in
Recent Forum Topics goto Post ramchop takes on the...goto Post NBFL Season 32: The ...goto Post Creating a custom to...
SearchSearch 
Post new topic   Reply to topic
View previous topic Log in to check your private messages View next topic
sk8bcn



Joined: Apr 13, 2004

Post   Posted: Jul 29, 2009 - 13:56 Reply with quote Back to top

This has a serious consequence Smile

I'd better be aware that team that are hurt are better activated alone. As it picks you the best game (well with a random factor added), you might have trouble to get your 180/140 team scheduled.

Which makes me think: Couldn't the random factor be weighted by the player instead of the "prefered team" activation method?

_________________
Join NL Raises from the Ashes
DukeTyrion



Joined: Feb 18, 2004

Post   Posted: Jul 29, 2009 - 13:56 Reply with quote Back to top

Unstoffe wrote:
Excellent work here - looking forward to getting some [B] games in at the weekend.
My thoughts on the points raised so far :
1) I like the way the scheduler tries to maximise the number of games, I would certainly rather get a slightly uneven match than for me or another coach to get no game at all.


I agree with this.

I would prefer the 4 games with an average suitability of 910, to 3 games (and 2 coaches missing a round) with an average suitability of 950.
CircularLogic



Joined: Aug 22, 2003

Post   Posted: Jul 29, 2009 - 14:38 Reply with quote Back to top

But would you prefer 4 games @750 to 3 games at 880?
Chingis



Joined: Jul 09, 2007

Post   Posted: Jul 31, 2009 - 09:02 Reply with quote Back to top

I think there are two separate scheduling questions here:

1) Would you prefer 4 games at 750 average to 3 games at 880 average, say?

2) Given an average suitability of 750, does it matter if this is made up of:
a) 3x900 and 1x300
b) 4x750
Currently the scheduler doesn't distinguish between the two, it simply goes for the highest total, which if we changed it ever so slightly to 3x900 and 1x301 would be schedule a).
TheCetusProject



Joined: May 25, 2004

Post   Posted: Jul 31, 2009 - 10:31 Reply with quote Back to top

One thing you could do to solve scenario 2) is, if suitability is lower than some value, give the coach with the inferior team in the poor matchup the absolute right to cancel the game and not play.
sk8bcn



Joined: Apr 13, 2004

Post   Posted: Jul 31, 2009 - 10:37 Reply with quote Back to top

CircularLogic wrote:
But would you prefer 4 games @750 to 3 games at 880?


personnaly yes too

_________________
Join NL Raises from the Ashes
sk8bcn



Joined: Apr 13, 2004

Post   Posted: Jul 31, 2009 - 11:06 Reply with quote Back to top

I wanted to prove that 300 in suitability is impossible (tough actully, it seems it is.

I hence followed the procedure with an extreme exemple:

two teams: one chaos TR/TS 100/90 vs zons 100/105.

Christer wrote:
1. Calculate the win probability (p) of the match as follows:
- Get the team strengths of the two teams.
- Adjust the strength of one of the teams with 5 TS per handicap.
- Calculate the win probability for the teams using the normal ranking formula. In this step, use equal rankings (ie, dR = 0).


The calculation gives me 17.3% in win probability

Quote:
- In the above ranking formula, if the normalised strength difference is above 5, use the following instead:
dT = (dT*3) - 10*dT (properly adjusted for who is higher or lower)


With the formula change, the proba is modified to 0.2%.

Quote:
- Apply racial factor as normal.


knowing that Zon vs Chaos is the worst matchup in this TS zone at 33,6%, I get:

5,1% under normal conditions, but 0.003% under the formula change.

Quote:
2. Subtract p by 0.5 (transposing the value to be centered around 0 rather than 0.5).


ok -4.997

Quote:
3. Define the distance as the absolute value of p


p distance= 4.997

Quote:
4. Apply a small random factor (adding 0 - 0.02)


skipped

Quote:
5. Normalise the distance to 0-1


That's were I'm blocked. I don't have the standart deviance (of what, actually? games in the round? possible TS-matchup of fictive teams? ...)

Quote:
6. Get the base suitability as 1-normalised distance

7. If the two teams are of the same race, multiply suitability by 0.97

8. If either team played the other in their last game, multiply suitability by 0.94

9. If there are any handicaps in the game, multiply suitability by (1-numHandicaps * 0.03)

10. Scale suitability to 0-1000 (ie, multiply by 1000 and round to an integer).


unfortunately, I can't go further.

_________________
Join NL Raises from the Ashes
CircularLogic



Joined: Aug 22, 2003

Post   Posted: Jul 31, 2009 - 11:09 Reply with quote Back to top

sk8bcn wrote:
CircularLogic wrote:
But would you prefer 4 games @750 to 3 games at 880?


personnaly yes too

That is your preference - I rather wait 30 more minutes instead having a really onesided game for one hour.
Rijssiej



Joined: Jan 04, 2005

Post   Posted: Jul 31, 2009 - 11:16 Reply with quote Back to top

CircularLogic wrote:
sk8bcn wrote:
CircularLogic wrote:
But would you prefer 4 games @750 to 3 games at 880?


personnaly yes too

That is your preference - I rather wait 30 more minutes instead having a really onesided game for one hour.


I would rather wait too.
CircularLogic



Joined: Aug 22, 2003

Post   Posted: Jul 31, 2009 - 11:22 Reply with quote Back to top

sk8bcn wrote:

5,1% under normal conditions, but 0.003% under the formula change.

Quote:
2. Subtract p by 0.5 (transposing the value to be centered around 0 rather than 0.5).


ok -4.997

Quote:
3. Define the distance as the absolute value of p


p distance= 4.997

Quote:
4. Apply a small random factor (adding 0 - 0.02)


skipped

Quote:
5. Normalise the distance to 0-1


That's were I'm blocked. I don't have the standart deviance (of what, actually? games in the round? possible TS-matchup of fictive teams? ...)

Quote:
6. Get the base suitability as 1-normalised distance

7. If the two teams are of the same race, multiply suitability by 0.97

8. If either team played the other in their last game, multiply suitability by 0.94

9. If there are any handicaps in the game, multiply suitability by (1-numHandicaps * 0.03)

10. Scale suitability to 0-1000 (ie, multiply by 1000 and round to an integer).


unfortunately, I can't go further.


The bolded part is the first error. if you subtract 0.5, you get -0.4997. The absolute is 0.4997. The absolute value can range from 0 to 0.5 so to normalize it, you just have to double the value, so that a spread between 0 and 1 is possible. So we get 0.994 - or a siutability of 6. Though I seriously doubt that a win% of 0.003 is right.
treborius



Joined: Apr 05, 2008

Post   Posted: Jul 31, 2009 - 22:52 Reply with quote Back to top

Rijssiej wrote:
I would rather wait too.

me three
Igvy



Joined: Apr 29, 2007

Post   Posted: Aug 01, 2009 - 07:48 Reply with quote Back to top

Christer wrote:
.. and as a happy coincidence, this just happened:

Code:
Generating schedule #1
- Balle2000 vs peikko  (Norse 120/120  vs 129/121 Chaos) @965
- haenschel vs Woodstock  (Dwarf 163/161  vs 156/151 Dark Elf) @936
- Fathernurgle vs luigi  (Ogre 172/146  vs 167/140 Nurgle's Rotters) @974
New best score: 2875
(...)
Generating schedule #3
- Woodstock vs Fathernurgle  (Lizardmen 131/128  vs 172/146 Ogre) @887
- haenschel vs Balle2000  (Orc 165/149  vs 137/144 Amazon) @886
- peikko vs Lochlainn  (Chaos 111/94  vs 130/99 Ogre) @963
- Men-o-maN vs luigi  (Orc 204/193  vs 222/194 Khemri) @904
New best score: 3640
Generating schedule #4
- Balle2000 vs Woodstock  (Norse 120/120  vs 131/128 Lizardmen) @963
- haenschel vs luigi  (Khemri 155/139  vs 160/147 Lizardmen) @973
- peikko vs Lochlainn  (Chaos 111/94  vs 130/99 Ogre) @963
- Fathernurgle vs Men-o-maN  (Orc 203/194  vs 204/193 Orc) @959
New best score: 3858


The first attempt was with three matches scheduled with an average suitability of 958. A four match schedule was then found with an average suitability of 910. A reduction of suitability average by 48 points. Then another four match schedule was found with an average of 965, an improvement by 7 points. None of the three match schedules generated in any pass had a higher average than the first (in fact, most of them were the same schedule).

The worst case scenario is most likely very very uncommon though due to the minimum number of coaches and the fact that most people have multiple teams. Still, it may be worth considering the average method instead of using the total sum to avoid the extremes.


Maybe you vary the formula by coach depending on the number of teams activated. Calculate both the sum and average.
Then weight the number of teams a coach activated to the formula. This would mean you would really only have the chance of missing out if you activated only 1 or 2 teams that can't be matched eaisly.

However this is a pain when you are trying to play a team that doesn't match well. Activating any other teams will just usally get them picked.

Just a thought.
Christer



Joined: Aug 02, 2003

Post   Posted: Aug 01, 2009 - 09:41
FUMBBL Staff
Reply with quote Back to top

So far, the worst suitability scores that the scheduler has seen since the changes look like this:

Code:
- Wreckage vs Kord  (High Elf 100/98  vs 138/122 Orc) @581
- KingStevenW vs ger0nim0  (Halfling 100/85  vs 100/97 Chaos Dwarf) @590


Neither of these were actually scheduled, but I'd say this is approaching the worst case that the scheduler considers.
sk8bcn



Joined: Apr 13, 2004

Post   Posted: Aug 03, 2009 - 12:15 Reply with quote Back to top

CircularLogic wrote:
sk8bcn wrote:
CircularLogic wrote:
But would you prefer 4 games @750 to 3 games at 880?


personnaly yes too

That is your preference - I rather wait 30 more minutes instead having a really onesided game for one hour.


yes I know Smile

Christer wanted our opinion. I just gave mine.

_________________
Join NL Raises from the Ashes
CircularLogic



Joined: Aug 22, 2003

Post   Posted: Aug 03, 2009 - 13:03 Reply with quote Back to top

Now that the box is out of testing, could the inactive coaches taken out of the blackbox top10?

There was already a poll and a vast majority was in favor of some kind of cleanup.
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