Warrior General Games

Originally posted by Corndog
It will actually probably be pretty weird for rookie teams.

All teams start at 1000 elo. Losing games make your ratings go down...so the teams with awful records will be below 1000 when new teams are added, despite being sophomore.

I'm not sure how long it will take to work itself out for sure. We may do some weird thing when teams go from Rookie to Sophomore where we add 200 or something to break them up from the new teams, though that creates other weird issues as well.

Or we could just let it work itself out, though a lot of rookie teams next season would probably lose their first few ladder games at least.

So um...what did you guys do here? I see that all the rookie/new sophomore teams have the same rating (tied for rank 155) and are below every existing team. I'm certain that if it's normal ELO rating that a team like http://glb2.warriorgeneral.com/game/team/208 that went 6-22-2 would have a below-starting ELO rating, but they're ahead of all the new teams. So something funky is happening. And Bort made it sound like he didn't do anything with the midseason teams in S1, but those all started out below all the existing teams too.

We have the unique problem that newly released sophomore teams (which is a S2 oddity that won't happen again) are going to be mixed in with the rookie teams. We'll see how long that keeps affecting things, but at least all the new Soph teams will be playing higher rated existing sophomore teams most of the time in league play and getting a boost in rating that way, and almost certainly will be crushing rookie teams, especially early on. It might impact Rookie Ladder winners, as those teams lucky enough to avoid "New Sophomore" teams in the early weeks may have an edge on the ladder.

I'm also a little worried that something is wrong with the elo rating function and every team is getting +rating points every game, win or lose. It seems to be working out about correctly so far, but if there's a math error happening now it will probably get worse as we pile up more games and quite possibly no longer look sensical.

Originally posted by NiborRis
I'm also a little worried that something is wrong with the elo rating function and every team is getting +rating points every game, win or lose. It seems to be working out about correctly so far, but if there's a math error happening now it will probably get worse as we pile up more games and quite possibly no longer look sensical.

I was wondering something along those lines as well. Shouldn't a team that went 12-4 in ladder games be higher ranked than a team that went 8-7-1?

Originally posted by bdnannac
I was wondering something along those lines as well. Shouldn't a team that went 12-4 in ladder games be higher ranked than a team that went 8-7-1?



Got links? A few things:
1) Any team that reset down to Rookie also had their ladder rating reset, which should put them in the massive tie for 155th place.
2) Ladder AND league games count for the ladder ranking, so you'd also have to look at the league.
3) ELO sort of has a "strength of schedule" effect in it - basically you lose less rating points for losing to a higher ranked team and you gain less rating points for beating a lower ranked team, and so forth. So it's possible for a team with a slightly better record to end up slightly lower in rating. However, we didn't really play too many games, and there was a round robin inside each league, so that effect should be fairly small for now and not likely to be bigger than about 2 wins/losses.

#2 Answers most of my question. The other team I was talking about was 12-2 in the league and I was only 6-6-2. I wondered if there was some kind of strength of schedule effect too. My wins were mostly against higher teams, but, I could never crack the top 30.

Originally posted by NiborRis

2) Ladder AND league games count for the ladder ranking, so you'd also have to look at the league.

Can anyone confirm this?

Because that doesn't make any type of sense.

Originally posted by Jampy2.0
Can anyone confirm this?

Because that doesn't make any type of sense.

Im pretty sure those are 2 completely seperate, unrelatable things

when I won or lost a league game it affected my ladder ranking, like it was a ladder match. Maybe if they separate the league games from the ladder games it probably won't affect the rankings.

Originally posted by Reaper Of The Dead
when I won or lost a league game it affected my ladder ranking, like it was a ladder match.

Wow, didn't know that, that is very stupid.

Originally posted by bdnannac
I was wondering something along those lines as well. Shouldn't a team that went 12-4 in ladder games be higher ranked than a team that went 8-7-1?



It's not just about your record, it's about WHO you beat (or lost to).

The ladder is about the strength of your team. League games are a measure of that as well as ladder games. Ladder games just give you a chance to play a competitive game with someone you might not otherwise play. I think that NOT having league games affect your ladder rating would be stupid.

Also, PLEASE let bad sophomore teams play good rookie teams in the ladder. It's an awesome chance for bragging rights when the rookie wins. We just had a league champ with only 6 human bots on the team. If he can do that, we'll surely see some good rookie teams who are on par with the poorer sophomore teams.

Originally posted by NiborRis
So um...what did you guys do here? I see that all the rookie/new sophomore teams have the same rating (tied for rank 155) and are below every existing team.

Oooookay...

Now the new sophomore teams are rank 155 and the new rookie teams are "something below that" and thus tied for rank 181.

I still think it would be nice for you guys to just tell us what you're doing with the ratings, so we could actually, you know, provide meaningful feedback. Right now we have no idea if the new sophomore teams are rated 1 point higher than the new rookie teams or 10,000 points higher, and why the new teams are lower than ALL of the existing teams.

Right now, there's something wrong. Basically, no S2 team is EVER going to catch up to the good S1 teams. There's no reason at all for the "midseason" teams from S1 to have been so far back from the full season teams, either - they showed up in like week 6 and good teams should have been moved up well toward the top, but weren't. Good teams were just barely catching the mediocre teams.

Seriously:

#103 - http://glb2.warriorgeneral.com/game/team/128 - full season team, 15-15 record.
#104 - http://glb2.warriorgeneral.com/game/team/297 - "midseason" team, 15-7 record.
#105 - http://glb2.warriorgeneral.com/game/team/10 - full season team, 14-15-1 record.

A team that LOST 8 more games somehow has a higher rating?
Under an Elo rating system, with the parameters on matchups - impossible. Flat out, impossible.

Ratings are BROKEN.

If a rank 10 team is beating and losing to teams that are rank 5-15, then of COURSE they will be ranked higher than a team that is beating and losing to teams ranked 155. Most of my experience with ELO is in chess, so I'll use a chess example. Imagine I am ranked 1600, and I go to a tournament where I win 6 and lose 7 games. My resulting rating afterwards might be 1560, or 1570. A player ranked 900 goes to a tourney and wins all 13 of his games. Do you think his ELO should be higher than mine, just because he won more games?

We shouldn't be separating ladders by age. Just let teams play other teams that they match up with.

Originally posted by Jampy2.0
Wow, didn't know that, that is very stupid.

That is absolutely 100% logical. Of course league games have to count in the ladder.

Originally posted by NiborRis

Ratings are BROKEN.

No doubt about that.


Would have been great to get the ELO score and a + / - on a win loss on the matchup page

Originally posted by cavalier

Would have been great to get the ELO score and a + / - on a win loss on the matchup page

This. A little transparency wouldn't hurt.

Originally posted by Badhands
If a rank 10 team is beating and losing to teams that are rank 5-15, then of COURSE they will be ranked higher than a team that is beating and losing to teams ranked 155. Most of my experience with ELO is in chess, so I'll use a chess example. Imagine I am ranked 1600, and I go to a tournament where I win 6 and lose 7 games. My resulting rating afterwards might be 1560, or 1570. A player ranked 900 goes to a tourney and wins all 13 of his games. Do you think his ELO should be higher than mine, just because he won more games?

We shouldn't be separating ladders by age. Just let teams play other teams that they match up with.

In the real world there are a lot of factors to consider with how ELO moves, yes. But your example is not what happened here. Also, you keep freely swapping the words "ranking" and "rating" and it's an important difference.

Take your experience. Now imagine 200 people in a room that all start at rating 1600. They play 8 games against each other in more or less random fashion. Now add 70 more people to the room starting at rating 1600. Now have ALL of those 270 people play each other 22 more times, and half the games people are paired up against an evenly rated opponent so their rating moves about the same amount each time. THAT is more like what happened here. Now, when you look at one of those 200 people who went 15-15, and compare against one of the 70 who went 15-7...yes, the person who went 15-7 would be noticeably higher in rating when it's all done.

Because of what we know of the starting point of all the teams ratings, and how the games were matched up, we have a very reasonable idea of how the rating movement should have worked. It's not the same as looking at a chess tournament full of people with varied ratings to begin with.