| 1 |
I'm
aware
of
the
problem
as
[url=https://zero-k.
info/Forum/Thread/36555]described[/url]
by
@dunno
(
and
linked
by
@malric
)
.
There
are
multiple
possible
solutions.
I
just
wanted
to
exclude
those
that
would
obviously
not
work.
|
1 |
I'm
aware
of
the
problem
as
[url=https://zero-k.
info/Forum/Thread/36555]described[/url]
by
@dunno
(
and
linked
by
@malric
)
.
There
are
multiple
possible
solutions.
I
just
wanted
to
exclude
those
that
would
obviously
not
work.
|
| 2 |
\n
|
2 |
\n
|
| 3 |
[q]If you can point me at the exact balancing logic we use I would be happy to take a look - I'll probably go find it myself at some point.[/q]Essentially, it minimizes the difference between the teams' WHR averages. The balancer uses ladder rating, the predicter uses actual rating. For the details of the code, look at the links at the bottom of [url=https://zero-k.info/Forum/Post/239694#239694]this post[/url].
|
3 |
[q]If you can point me at the exact balancing logic we use I would be happy to take a look - I'll probably go find it myself at some point.[/q]Essentially, it minimizes the difference between the teams' WHR averages. The balancer uses ladder rating, the predicter uses actual rating. For the details of the code, look at the links at the bottom of [url=https://zero-k.info/Forum/Post/239694#239694]this post[/url].
|
| 4 |
\n
|
4 |
\n
|
| 5 |
@Helwor's 2/3 suggestion would also violate against the invariance of the elo scale under additive shifting transformation. According to @dunno's data, @Helwor's suggestion would overcompensate the effect for some cases and undercompensate it for other cases.
|
5 |
@Helwor's 2/3 suggestion would also violate against the invariance of the elo scale under additive shifting transformation. According to @dunno's data, @Helwor's suggestion would overcompensate the effect for some cases and undercompensate it for other cases.
|
| 6 |
\n
|
6 |
\n
|
| 7 |
I suggest that those who want to make useful suggestions for solutions should first understand @dunno's analysis.
|
7 |
I suggest that those who want to make useful suggestions for solutions should first understand @dunno's analysis.
|
| 8 |
\n
|
8 |
\n
|
| 9 |
[q]@dunno's analysis already had a conclusion:
|
9 |
[q]@dunno's analysis already had a conclusion:
|
| 10 |
[q]Summary: Uneven team games should probably be balanced such that the smaller team has about 50-200 higher average Elo than the larger team[/q]
|
10 |
[q]Summary: Uneven team games should probably be balanced such that the smaller team has about 50-200 higher average Elo than the larger team[/q]
|
| 11 |
If there would be agreement to try implementing that conclusion, I guess the code change will not be extremely hard. I looked through the analysis but I did not do a through review, someone working more with statistics could probably do it easier than me ( @Brackman ?..)[/q]
|
11 |
If there would be agreement to try implementing that conclusion, I guess the code change will not be extremely hard. I looked through the analysis but I did not do a through review, someone working more with statistics could probably do it easier than me ( @Brackman ?..)[/q]
|
| 12 |
Yes, this would be a decent solution and rather easy to implement. The question is only how the epad vector should be calculated exactly.
|
12 |
Yes, this would be a decent solution and rather easy to implement. The question is only how the epad vector should be calculated exactly.
|
Matchmaker balancing for uneven team counts seems broken