I have to second Brisk about Gnuffone. thx Fizzer

Are you saying we'll need another poll? ;)

If the ladders changed to Trueskill, there would be different options on how to do the transition.

One option is to re-run every game ever played through Trueskill and use the resulting ratings. This would be as if Trueskill was used from day 1. This makes some sense, since games in a Trueskill world wouldn't expire.

Another is to simply start Trueskill with the current ratings and continue. This makes some sense, since players before Trueskill were playing under bayeselo so they should be rated under bayeselo.

Another option is to run the last 3 months in Trueskill. This doesn't seem to make any real-world sense since the 3-month expiration wouldn't apply to Trueskill.

I like the first option a lot, but I think the second would make most sense for players and coder.

As mentioned, why no 3v3 europe option?

Also, last template on RT suggestions says map does not exist (slow earth).

slow earth is the ME template of szew :P

I'm actually a little curious about the practical implementation of the second option. After all, the standard deviations in TrueSkill play a major role, while those of BayesElo are mostly hidden. I wonder if they show the same features and if BayesElo SD's can be simply used as TS SD's without some really weird things happening.

@ Arun: for player yes, but for coder option 1 is actually easier.

Option 3 doesn't make sense indeed as your first unexpired game counts as if you and your opponent are both playing someone unranked (which wasn't the case when the game was created).

((Funny how I do like option 1 much better myself while I would be almost certainly be better off with option 2.))

I don't like option 2

People with inflated rating would start with an inflated rating instead of the place they should be

I thought that myself Timi, then I thought about it some more. Wouldn't those people who gamed the system be hurt by using all games in TruSkill? Afterall, they probably had a high rating based on a small sample of unexpired games. I would think re-running it on all games would benefit those who did not game the system, as all games would be taken into account. Also, if you were a staller, those stalled losses which really did not effect your ladder rating before (since the player probably just left the ladder and let the losses happen), would now count against them.

I'd like to hear more from Math Wolf on this comparison, but it would seem to me that we'd want to count all old games so that real wins/losses are factored in, without the time element that ELO cares about (which is where the manipulation is happening).

isn't best reset the ladder? i thought was obvious. If we change setting/rating sysstem/etc, why old game should count?

Afterall, they probably had a high rating based on a small sample of unexpired games. I would think re-running it on all games would benefit those who did not game the system, as all games would be taken into account

It would also benefit those who abandon old account(s) with worse record and punish those who play continuously with same account

Why does it punish those who play continuously with the same account? I'm not saying you are wrong, I am just trying to completely understand the options before voting.

because when you start playing you are usually losing more - and those loses will be included in TS rating (i know that old games weight less than new one, but still are counted)
So, instead of staller there will be more alts in ladders as some people my try to get good result with more than one approach. That's why i don't like TS. Increasing required number of games in bELO should be enough - for example:[20 consecutive (not necessery first) or 30 total finished games].

Edited 9/2/2014 16:01:20

I didn't think old games weight any more or less. I thought timing only mattered in that your win counts based on the opponent rating at the time. So lets say I beat a player that was #1 at the time, then retired. My win doesn't expire (of course), but it also doesn't get diminished by the fact that the retired person's rating tanked from going inactive. A good win remains a good win.

In the same breath, a bad win is a bad win (people who got #1 by beating nobody worthwhile).

I am not sure about your alt comments. Why would TruSkill result in more alts? IMO, there would be less since you cannot make these 'fake' runs anymore.

I would think the only negative would be for a person new to the ladders it would take a long time to get high on the ladder.

Yes as far as I understood, there is NO weighting of games. The Trueskill was affected once, then never again by each game. More like a Chees ELO - more fair in my opinion.

The alts he was talking about would be because when we all started the ladder (most of us), we had a lower rating because we were newer and not as good. Take, for example, myself. Look at the graph on mine and you can clearly see an upward trend.

Now if the ladder flips to TrueSkill, all those months at the bottom end will hurt me. If I were to join with an alt, I could only play at my current skill, getting rid of all those low losses from the beginning.

Hmm, I agree with you there. It does punish people who were considerably worse to start out.

It does seem most fair to just start over in that regard.

about weighting games:

Also, games never expire like they do in the other ladders, but the TrueSkill algorithm weights newer games more highly, so you still have the ability to move your rating over time.

from http://blog.warlight.net/index.php/2014/03/website-update-2-5-real-time-ladder/


about alts:
games never expire -> you can't get rid of your old loses -> it's easier to play with new account without loses than recover from lower rating
Gnuff was angry that i mentioned this in another thread and probably will be again, but i can't do anything about that, as he is excellent example why i don't like TS. He already played RT ladder with (at least) 3 accounts. Two of them are already abandoned - no new game for a while(Gnuff and Killua). He play only with Marquis now. And new question of the day - compare a bunch of first games of all those accounts and tell me which has most wins.
There will be a lot of such behaviour if we use TS for 1v1 ladder.

Anyway - i found that:
http://blog.warlight.net/index.php/2012/01/trueskill/
there's a link for WLTrueSkill, but this file is not avaialble now. Maybe fizzer could reupload it, as it would be nice to do some simulations. (including end date for games API would be helpful too:P)

Edit: starting over is only temporary solution, there are still new people joining and players improve over time.

Edited 9/2/2014 16:51:28

But the problem remains... for new players. As they have to start from scratch.

Ok you quoted something but it doesn't make it correct. The Trueskill algorithm is much like ELO I thought - can you actually show where it takes old games into account?

No, i can't show anything, but can you show it doesn't take old games? I just assume Fizzer know what he use:P

Also, you can check - http://research.microsoft.com/en-us/projects/trueskill/faq.aspx
They've created know, so if you don't trust Fizzer, maybe you will trust them :P

Edited 9/2/2014 17:13:09