Latest Posts

Topic: Rating system

WorldSavior
Avatar
Joined: 2016-10-15, 04:10
Posts: 1257
Ranking
One Elder of Players
Location: GER
Posted at: 2019-09-09, 09:11

The last three tables don't seem to make sense.

trimard wrote:

Maybe I could just put directly 8 slots for each games? Widelands doesn't support much more players yet anyway AFAIK.

Thoughts?

There are maps for sixteen players and you can play on them, the problem is just that the UI doesn't allow it to make settings for all players. With last trunk it was possible for some time, but then that feature was unfortunately removed. Maybe 8 should be changed to 16 in your model?

Edited: 2019-09-09, 09:11

“It's a threat to our planet to believe that someone else will save it.” - Robert Swan

Top Quote
einstein13
Avatar
Joined: 2013-07-29, 00:01
Posts: 1088
Ranking
One Elder of Players
Location: Poland
Posted at: 2019-09-09, 10:01

WorldSavior wrote:

The last three tables don't seem to make sense.

I agree. I suggest that standard deviation should not be higher significantly than initial one. For example if SD > 350 or 400, you should reject the test.

What abut number of games per round: 2 and all factors as the first one? Can you simulate that?

(...) Maybe 8 should be changed to 16 in your model?

I guess that this can be fixed after "alpha-release" to cover all the cases face-wink.png . 8 is enough for now and probably never used.


einstein13
calculations & maps packages: http://wuatek.no-ip.org/~rak/widelands/

Top Quote
trimard
Joined: 2009-03-05, 22:40
Posts: 200
Ranking
Widelands-Forum-Junkie
Location: Paris
Posted at: 2019-09-09, 23:23

What abut number of games per round: 2 and all factors as the first one? Can you simulate that?

Rank Player Rating Deviation
1 worldsavior 2014 139
2 king of nowhere 1704 137
3 the-x 1666 124
4 mars 1616 164
5 kaputtnik 1606 186
6 tando 1600 164
7 hessenfarmer 1562 168
8 Hasi50 1502 171
9 trimard 1429 158
10 gunchleoc 1428 164
11 laza 1384 167
12 ektor 1328 195
13 ModellbahnerTT 1297 177
14 watchcat 1283 177
15 janus 1104 179

But there is a problem I think the factor ITERATION_LIMIT (number of iteration max on step 5) should be set to 2, whatever we do. Otherwise volatility never changes. That's why I started messing around with the number tbh, I didn't understand why the volatility was static. There wouldn't be all that fuss about step5 if could just ignore it right away!

On a side note, I think we need more games too face-tongue.png

There are maps for sixteen players and you can play on them, the problem is just that the UI doesn't allow it to make settings for all players. With last trunk it was possible for some time, but then that feature was unfortunately removed. Maybe 8 should be changed to 16 in your model?

Yeah sure, why not? Actually much easier to do than I though at first

PS: Ok, but after that, I'm starting back my studies this week, so I'll be much slower in the work and the replies. I'm sorry, it's a special, shall I say "bloody" year in my cursus face-tongue.png But I'll try to continue develop on some weekends. This project is a really good way to enjoy in spare times, it's a real pleasure having these discussions with you all! Thanks a lot face-grin.png face-grin.png

Edited: 2019-09-09, 23:29

Top Quote
kaputtnik
Avatar
Joined: 2013-02-18, 20:48
Posts: 1765
Ranking
One Elder of Players
Location: Germany
Posted at: 2019-09-10, 20:28

GunChleoc wrote:

A tribe table sounds good.

We have already the one for the encyclopedia: https://github.com/widelands/widelands-website/blob/master/wlhelp/models.py

Creating a form in django is easy, handling a form in django can be tricky face-smile.png


Top Quote
einstein13
Avatar
Joined: 2013-07-29, 00:01
Posts: 1088
Ranking
One Elder of Players
Location: Poland
Posted at: 2019-09-12, 14:05

trimard wrote:

BOtherwise volatility never changes. That's why I started messing around with the number tbh, I didn't understand why the volatility was static.

As I understand the model, volatility doesn't have to change. The most significant are Rating and Rating Deviation. The rest can be static and the points will change anyway.

Also note that our situation is not the best fit for the model: we are playing not too often, while the model works the best with multiple games for each player in every calculation step.

Ok, but after that, I'm starting back my studies this week, so I'll be much slower in the work and the replies.

You already did great job!


einstein13
calculations & maps packages: http://wuatek.no-ip.org/~rak/widelands/

Top Quote
trimard
Joined: 2009-03-05, 22:40
Posts: 200
Ranking
Widelands-Forum-Junkie
Location: Paris
Posted at: 2019-09-15, 21:20

Ok, so I'm starting to understand the whole thing a bit better.

Volatility

It should take 3 to 19 iteration for step 5. If we put 3 as a limit for now, we can have a reasonnable change of the volatility.

I will however make the decimal stored bigger so that even small cahnge can be taken into account.

Rating period

The number of game per round is our current rating period. But it's not done to work like that. It should be a time period in which we can hope player make from 10 to 15 games.

However if we applied this rule it would mean we wouldn't even have a complete rating deviation with current games. It means all the games would be calculate at the same time, which would mean this score:

wl-ico-64.png

Which doesn't seems so bad? Does it?

What I propose as an alternative is we make some kind of rule, like "at least 5 games per user rated in this period". We are far from the suggested 10 to 15 games in glicko. But it's already kind of a hard requirement. It would mean max 2 rating period I think for the current number of games we have, well maybe after a few more game.

Of course I'm talking average. So if some players play 20 games, it would make it up for the player that played only 2 or 3 games.

Rating period for player that had no games

I forgot about it. It is not yet implemented, and it shows in the last test I made. Of course the rating deviation should be much higher for most player!

Indeed that would mean that for 30 games, with 5 games rated by rating period, in total 6 rating period.

Hence in many of these rating period, some players wouldn't play and their standard deviation would get much higher.

Edited: 2019-09-15, 21:22

Attachment: Screenshot_2019-09-15_21-12-16.png (154.5 KB)

Top Quote
einstein13
Avatar
Joined: 2013-07-29, 00:01
Posts: 1088
Ranking
One Elder of Players
Location: Poland
Posted at: 2019-09-16, 00:03

trimard wrote:

Ok, so I'm starting to understand the whole thing a bit better. (...) What I propose as an alternative is we make some kind of rule, like "at least 5 games per user rated in this period". We are far from the suggested 10 to 15 games in glicko. But it's already kind of a hard requirement. It would mean max 2 rating period I think for the current number of games we have, well maybe after a few more game.

I forgot about it [no games played]. It is not yet implemented, and it shows in the last test I made. Of course the rating deviation should be much higher for most player!

How hard in your models and coding is to create one-game glicko recalculation? I think that official rank should be as good as possible (so it can be taken every week or two weeks), but players does want to know what are their current results too. That means changes every played game.


einstein13
calculations & maps packages: http://wuatek.no-ip.org/~rak/widelands/

Top Quote
trimard
Joined: 2009-03-05, 22:40
Posts: 200
Ranking
Widelands-Forum-Junkie
Location: Paris
Posted at: 2019-09-16, 08:54

Sorry that wasn't clear. Having a rating period doesn't mean we can't update the score after each game. Just tells us how to group the games.

If the game you just played created a new rating period it changes nothing except other's players standard deviation will get higher

Edited: 2019-09-16, 08:56

Top Quote
einstein13
Avatar
Joined: 2013-07-29, 00:01
Posts: 1088
Ranking
One Elder of Players
Location: Poland
Posted at: 2019-09-16, 11:15

Current usage of Elo system in chess doesn't update ranks all over the players, only to participants of the one particular game. So my idea is to apply Glicko the same way: only participants of the one game we are targeting changes the rank values.


einstein13
calculations & maps packages: http://wuatek.no-ip.org/~rak/widelands/

Top Quote
trimard
Joined: 2009-03-05, 22:40
Posts: 200
Ranking
Widelands-Forum-Junkie
Location: Paris
Posted at: 2019-09-16, 16:55

Rank value value won't be change though. Only standard deviation if, and only if, there is a change of rating period


Top Quote