The Official Unofficial Multiwinia Ladder

[Cyan.]

Does ladder calculate points when you have become second in a mulitplayer match?

[bert_the_turtle]

Yes. You won't get credited a win for the counts, but for the ranking, it'll almost be as if you beat all those below you in a duel.

[Cyan.]

That's a reliev heh, haha

[Imeanunoharm]

http://math.bu.edu/people/mg/research/gdescrip.pdf

Just on the issue of ranking movement. In the last paragraph there is a practical problem outlined where RD's should not move below 30. Sorry it wouldn't copy and paste properly. Is this implemented in the ladder?

[Lelouch]

Is it possible that after winning 5 ranked matches your score stays the same? Or now you can move up only if you beaten opponent that is higher than you on ladder?

[Imeanunoharm]

I had a 10 streak immediately before the end of the last rating period (2) and didn't move an inch wehn the rating period switched. One of those matches has Mas and N0rd in it, but most were with lower ranked players. I suppose a couple of my losses earlier that week could have dragged me down.

It does make me wonder how olypiacos, mach, mrmalone, cheesmoo0 and maybe a couple of others have the scores they do even though you look at their matches and they might have had one match won with a slightly higher ranked player No offence to those guys. That's why I asked about how the scoring worked really early in the piece before rating period 2, because it seems to have elevated some people arbitarily early on, and I can't see why others are not pushing them out quicker. Those guys only have a few games, so jelco's explanation doesn't really answer it properly.

Still seems weird to me, but then again, I don't know how the data is collated and pumped through the scripts. Jelso has said that the data is run from scratch every rating period, but does that mean:
RP1 scores at 1500, that weeks game data is fed into RP2, RP2 data is fed into RP3,
OR
RP3's script data is taken from all 3 weeks and crunched again.

You see, I don't ever recall seeing everyone's score on 1500 (like it should have been in week 1), but that was because jelco was running daily updates. As far as I understand it, those daily updates are largely meaningless, because at the end of the week all the weekly data should be recruched again, otherwise the daily data will introduce compounding errors as the glicko system only works on a rating period basis (in this case 1 week). I can see why he does it the way he does, as daily updates are kind of fun, but unless the daily updates are "discarded" at the end of the rating period then there will be a problem.

Just some thoughts out loud. Sorry if this is getting a little anal, but I'm one of thos silly people that likes to understand (at least basically) everything they use, computers, cars, human bodies, etc.

[Lelouch]

I hardly understand how that ladder works aswell...meh... guess its just one of those mysteries - to be uncovered in next century by our grandchildrens. Bet it turns out to be a fluke too.

[The GoldFish]

Hm, well. The raned/unranked pie chart currently labels both as unranked. Could do with fixing?

[bert_the_turtle]

I guess the daily recalculations are discarded in the weekly update. It doesn't make sense any other way, really.

About you not moving a lot despite grand victories: that's the rating deviations at work. The lower your deviation, the less you move for each game you play. The more you play, the lower your rating deviation. Also, the number of games you play essentially cancels out. If you want to advance fast, play only a few ranked games per period, and make them good games. If your week has a bad start, play some more games to compensate. It would be useful if your profile would show you
a) your rating deviation
b) the number of games played in the current rating period
the saturation effect where more games won't move you any further hits when the number of games this period is about (200/RD)^2, if Jelco stuck to the default constants.

[bert_the_turtle]

I did stick to those constants, although I'm not sure how that equation of yours would work but it sounds alright.

It's simple: The shift your rating gets each period is scaled by a factor of 1/(RD^-2 + d^-2), where d^-2 is bound from above by q^2 * number of games/4. The point where RD^-2 roughly equals d^-2 is when your number of games, and that's when the number of games is roughly (RD * q/2)^-2. And the default value of q is roughly 1/200, hence the threshold number of games is (400/RD)^2. Yeah, I missed that factor of 4 in my first calculation

[Imeanunoharm]

... the other thing is that we started on a Sunday night, hence the first official RP was just 24 hours. The thing you regarded as week 1 was technically week 2.

Finally! The crucial missing piece of the puzzle. I hypothesise that anyone that played in the first 24 hours in all probability has a slightly skewed score depending on who they played and how they went, as compared to those who didn't play in those 24 hours.

I know I may have sounded a bit repetitive (and if you really read my last post, it's more a musing on thoughts than a personal gibe, and didn't require that sort of coverage, though again I appreciate it), but I've come into this blind and you've come in with prior knowledge, and something like this is what I have been implying all along if you go back and have a good read yourself. I'm not a statistician, but I have studied some stats and am a scientist. Things like this can have compound effects on stats, as a very small number of small games in RP1 will have a butterfly effect that cascades down the chain on the data until much much larger data sets come in and average the results. The reason I sound repetitive is because no-one except me seems to have see this (and I admit I still may be quite wrong ).

I mean, as a scientist, I have to ask, how do you really know this? "The big avalanche of matches that followed will greatly negate the effects of this weird start anyway."

I don't mean to be confrontational, really I don't, and I do appreciate all the effort that has gone into this (and your limited time to deal with these things), genuinely, but do me favour if you can, if it's easy and possible. Take the first days (or RP1's) data out and see what happens. Do the experiment, and satisfy the scientist (in all of us ). My hypotheisis, now that I know that RP1 is 1 day, is that with that data removed, the ladder will make much more sense, and anomolies will be less prevalent.

If it's particularly hard though, just as likely ignore me, it's really not that huge a deal. The stat's will start to sort themselves out eventually, but I hypothesise it will take longer. This is science though. You're a youngish guy. Science is about explaining yourself in the best way possible with evidence. Sure this is a game ladder though, so ... anyway, it's late here is Aus, and I now the beer is rambling.

[Phelanpt]

I understand though that the scoring system is difficult to comprehend so I keep working on it.

On other discussions about rating systems, I've seen a phrase that seems relevant: "Fair or easy to understand: Pick one"

[Imeanunoharm]

Don't start sprouting nonsense now just because - I'm willing to admit I missed something but there is no way you can convince me that this is what you were talking about.

It wasn't?

But what seems quite strange to me is that there are people with way worse win loss ratio's than me that seem to have higher scores, eg. #670316_(4)_0:4, and there are a tonne of people that have won one game and got 27 points straight off the bat. My first ranked game I lost, does that instantly set me up as a "weak" player (RE the "strength" system metioned in the faq), and thus it is much harder for me to get points for wins or something? (Another eg. #670305_(2)_1:1 where his first game was a win and what looks like a quick 57 points.)

So how was the ladder sorted/ranked originally if everyone started on 1500? First to play a game (and win)?

What I mean is, how did some people that have much fewer wins than others early on get such high scores/ratings early in the peice (ie rating period 1&2)? I mean everyone was equal strength, so how do they rate high if they have few wins?

Anyway, I apologise for that last paragraph as it came out quite wrong. I don't think I've ever said you are wrong, it was just a late night poorly worded attempt on the nature of science as an experimental medium where evidence speaks volumes. I've always found that the best scientific defence is to plonk the data on the table (something I've noticed younger people seem to forget, instead, taking it personally, though come to think of it there's a few oldies that do that too, so I apologise again for that ). However, since I can't run your experiment, I have to take your word for it, which is fine, as I said, it's a gaming ladder, not dark matter research.

If you have any more questions, don't hesitate to ask.

You might regret that

Cheers.

[The GoldFish]

Jist is: The ladder's responce time can be slow. Give it a couple of weeks and hopefully things will even out.

This is just how the Glicko system that is being used works - the underlying function of it is that it expects you to do as well in one week as another, and that in every subsequent rating period in which you play, your performance is a further refinement of your rating (and so the change each victory or loss makes is smaller on a per period basis). If you perform below your eventual average in the first week that you happen to play a lot of games, it may be harder to destabalise that reputation in the long run, compared with someone who has as shakey start but plays only a few games. Odds are, though, that things will normalise in a couple more rating periods and begin to make sense.

It's my opinion that the ladder could do with being a little more, well, let's say forgetful in that regard, but that's only relavent over longer periods.

Plus, if for some reason they don't start to make sense, Jelco wil have a lot more data to actually check against, and people will have better comments to make on the ladder in general. The fact that it even works in the first place is some sort of miracle. Relax that you're in the top 20 (if you happen to be in the mix)

[Imeanunoharm]

What happened with the latest script update? I jumped about 120 points and 15 spots in one night, and the ladder is making a lot more sense. No not just because I moved There was a huge amount of movement and many genuinely deserving players have made large movements in the right direction.