52 coaches online • Server time: 18:47
Index Search Usergroups Profile
Log in
Recent Forum Topics goto Post Blocking in turn 16goto Post kill, apo, killgoto Post Any Elves but Wood? ...
SearchSearch 
Post new topic   Reply to topic
View previous topic Log in to check your private messages View next topic
SzieberthAdam



Joined: Aug 31, 2008

Post   Posted: Feb 08, 2019 - 11:59 Reply with quote Back to top

I was trying to go over the Blackbox scheduler sustainability calculation to get a better view of its method but it seems that the documented way is either wrong or I made a bug which I doubt as I checked everything many times. I would be good to get a values with a factor of 0.5.

Here is my code:
https://github.com/FUMBBLPlus/blackbox/blob/master/sust.py

It seems that the Point 5 (Apply a small random factor, and normalise) returns a value in range of 0..1000 instead of 500..1000 plus somehow the sustainability result is inverted. I get higher scores when the site scheduler got lower and in the other way.

I was particuarly interested in this draw:
https://fumbbl.com/p/blackbox?op=scheduler&r=2019-02-08+01%3A30

Smallman complained about the high TV difference and about the scheduler in general and my mono activation but that goes to another post.

For the top 3 matchups I get the following results:

FUMBBL:
TV 1030k Necromantic Horror (3)
TV 1060k Norse (1)
@971

My script:
Code:
> python sust.py 1030 1060 3 1
dTV_0 = 29.126213592232997
dTV = 29.126213592232997
p_0 = 0.4760663100888059
factor = 0.5
p = 0.4760663100888059
d_0 = 0.0239336899111941
d = 0.0525472527395459
S_0 = 52.5472527395459
r_opp = 1
r_games = 1
S = 52.5472527395459


FUMBBL:
TV 1240k Skaven (5)
TV 1150k Human (6)
@896

My script:
Code:
> python sust.py 1240 1150 5 6
dTV_0 = 78.26086956521738
dTV = 134.78260869565213
p_0 = 0.3909418562318328
factor = 0.5
p = 0.3909418562318328
d_0 = 0.10905814376816719
d = 0.23520884343857418
S_0 = 235.20884343857418
r_opp = 1
r_games = 1
S = 235.20884343857418


And finally our matchup with 920 TV difference:

FUMBBL:
TV 2550k Dwarf (70)
TV 1630k Elven Union (33)
@510

My script:
Code:
> python sust.py 2550 1630 70 33
dTV_0 = 564.4171779141105
dTV = 1593.2515337423315
p_0 = 0.005267839562496554
factor = 0.5
p = 0.005267839562496554
d_0 = 0.49473216043750345
d = 0.9880109899154592
S_0 = 988.0109899154592
r_opp = 1
r_games = 1
S = 988.0109899154592


I will update if I know more.

_________________
"I love him who is ashamed when the dice fall in his favour, and who then asketh: 'Am I a dishonest player?' - for he is willing to succumb."
-- Friedrich Nietzsche: Thus Spake Zarathustra, Zarathustra's Discourses, 4.
CircularLogic



Joined: Aug 22, 2003

Post   Posted: Feb 10, 2019 - 10:14 Reply with quote Back to top

What I noticed in the linked description of the suitability score:
The graph in the bottom as well as steps 7-9 seem backwards as low suitability scores are the best. A rookie mirror match comes out at p=.5 and so without random factor you arrive at a distance d=0 which gives you a suitability score of 0 - which should be the best match.

There is your inversion @SzieberthAdam
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