Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Emeric]

Monsieur X is a good friend and I don't think he played 50.000 games in the last few days :

[SCSN]

I wouldn't be so sure about that, based on the limited amount I've played him he seems quite the beast.

[ragingduckd]

Monsieur X is a good friend and I don't think he played 50.000 games in the last few days :

Thanks.  This is the same old leaderboard bug that's been happening on and off for months.  It's been really hard to replicate and track down, so I've just been fixing the damage by regenerating the leaderboard from scratch when it happens.

I've done that again (so there are no 50k+ players now), and I think I've solved the underlying problem too this time.  It was happening when the log parser encountered two pro games with identical timestamps (down to the microsecond).  I had some faulty code that resulted in parsing the two games over and over until the parser hit it's 100k max-games-per-query.  So the winning player's rating and variance would go through the roof, as would his number of games played.

[ThaddeusB]

The log search is updating (i.e. new games appear when searched for), but the ratings aren't updating currently (and for the last 16 hours or so).

[ragingduckd]

The log search is updating (i.e. new games appear when searched for), but the ratings aren't updating currently (and for the last 16 hours or so).

Thanks.  I broke something else while fixing the 50k game issue.  Ratings are updating again.

[theblankman]

Conclusive proof that Goko Pro has it all wrong compared to Isotropish  

[Elanchana]

Alright I'm a bit confused. I just installed Salvager a few days ago and I'm enjoying it a lot but I don't understand the Iso ratings. I saw someone who had a pretty high pro rating and a good win ratio but his Iso level was in the -20 range. Meanwhile, I've leveled up twice today - once by losing a game and once from my opponent quitting. Isn't that the kind of stuff that's not supposed to happen? I think I'm going to need a more detailed explanation of the variables so I know what's going on.

[sudgy]

Alright I'm a bit confused. I just installed Salvager a few days ago and I'm enjoying it a lot but I don't understand the Iso ratings. I saw someone who had a pretty high pro rating and a good win ratio but his Iso level was in the -20 range. Meanwhile, I've leveled up twice today - once by losing a game and once from my opponent quitting. Isn't that the kind of stuff that's not supposed to happen? I think I'm going to need a more detailed explanation of the variables so I know what's going on.

First, pro mode and Iso aren't the same, that's why we have Iso ratings.  I'm at level 25 in Iso ratings and just checked my pro rating recently and it was 2700 (I haven't played in a while) (I started playing again, pushing it up to 4500).  I don't know why you got rating for losing a game, and when your opponent quit you should definitely get rating for that.

[Joseph2302]

Alright I'm a bit confused. I just installed Salvager a few days ago and I'm enjoying it a lot but I don't understand the Iso ratings. I saw someone who had a pretty high pro rating and a good win ratio but his Iso level was in the -20 range. Meanwhile, I've leveled up twice today - once by losing a game and once from my opponent quitting. Isn't that the kind of stuff that's not supposed to happen? I think I'm going to need a more detailed explanation of the variables so I know what's going on.

I can't give a massively detailed explanation, however:

(1). Goko Pro includes games against bots and games with 3+players, Iso only counts 2 players, no bots . Therefore, if someone has a good Pro rating and bad Iso rating it probably means they play most of their games against bots, or 3+players.
(2). If your opponent quits/resigns, you still gain Iso rating, unless the game is really short (i.e. it doesn't load for someone).
(3). Iso rating is calculated by taking a skill part and an uncertainty part, and subtracts the uncertainty from the skill, and rounds down to get your level i.e. according to the Iso leaderboard, I'm currently 31.26±9.94, so level 21, as 31.26-9.94 = 21.32, which become 21. So, I guess (not 100% sure about this) that your uncertainty must have increased decreased more than the skill bit decreased, so get higher despite losing a game. Especially if you don't play much, your uncertainty can be high, and then gets lower quite quickly if you play lots of games.

Hope this was helpful

[qmech]

Note that Iso doesn't update instantly, so the changes you observed might have been caused by games different from those that you guessed.

[Awaclus]

your uncertainty must have increased decreased more than the skill bit

I hate to nitpick but this has a potential to make things super confusing, so FTFY.

Note that Iso doesn't update instantly, so the changes you observed might have been caused by games different from those that you guessed.

Doesn't it update almost instantly anyway?

[Joseph2302]

Note that Iso doesn't update instantly, so the changes you observed might have been caused by games different from those that you guessed.

Doesn't it update almost instantly anyway?

If it's working properly, then it updates every 1-2 minutes I believe.

[qmech]

Yes, it's pretty quick, but as it's not living on the MF servers it has to download all of the logs, which it doesn't do constantly so as not to overload their server.  (Let's face it, they need all the help they can get.)

[Mic Qsenoch]

(1). Goko Pro includes games against bots and games with 3+players, Iso only counts 2 players, no bots . Therefore, if someone has a good Pro rating and bad Iso rating it probably means they play most of their games against bots, or 3+players.

Isotropish includes bot games for sure.

[Joseph2302]

(1). Goko Pro includes games against bots and games with 3+players, Iso only counts 2 players, no bots . Therefore, if someone has a good Pro rating and bad Iso rating it probably means they play most of their games against bots, or 3+players.

Isotropish includes bot games for sure.

I should really stop talking, seems everything here is wrong.

[GeoLib]

Both isotropish and goko pro are implementations of Microsoft TrueSkill (though goko has not admitted it, AI reverse-engineered the system). They use different parameters however.

[DG]

Isotropish has me down for 78 games. Those are all my league games and it is not counting any of my bot games (which is good).

[Mic Qsenoch]

Isotropish has me down for 78 games. Those are all my league games and it is not counting any of my bot games (which is good).

Do you play Pro or Casual against the bots? I am pretty sure it counts bots, as there was a thread about Canton Pilots Association (who just plays bots, and is highly rated), also Tao Chen (hdu88 on the forums I think?) has been almost exclusively playing bots recently while reaching his current high level.

[DG]

I select the 'play bots' option rather than going into the kingdom creation screen. It adjusts the Goko score for both casual and pro (I think). To be honest I'd rather nobody investigated this thoroughly enough to make any changes!

[Elanchana]

I select the 'play bots' option rather than going into the kingdom creation screen. It adjusts the Goko score for both casual and pro (I think). To be honest I'd rather nobody investigated this thoroughly enough to make any changes!

I think that the "play bots" option just goes to your casual score (which is weird because it doesn't give you the option of picking the kingdom). That's what it was last time I checked. There's an option to play a pro game with a bot in the multiplayer server and that's probably what Iso counts.

Weird thing though - when I played casual games with bots on the multiplayer server it didn't do anything to my rating. I wonder if it's the same thing for pro because that would kind of defeat the purpose of pro vs. bots.

[Beyond Awesome]

Sometimes when I'm bored, I play bots through the main screen. I can assure you that it counts for pro rating.

[Elanchana]

Sometimes when I'm bored, I play bots through the main screen. I can assure you that it counts for pro rating.

Okay then, I stand corrected.

[Polk5440]

Sometimes when I'm bored, I play bots through the main screen. I can assure you that it counts for pro rating.

Okay then, I stand corrected.

It used to count for casual. They recently changed it.

[-Stef-]

I have a request: please no longer count bot games for Isotropish.

I'm starting to get the feeling the bot games seriously mess with the ratings now.
My theory is this: It's just too easy to beat them once you get the basic concepts of the game. Their ratings won't be really low because with their reasonably good executions of mediocre strategies they beat up a lot of new players.

Maybe some players that actually play bots can weigh in on this one.

[WanderingWinder]

I have a request: please no longer count bot games for Isotropish.

I'm starting to get the feeling the bot games seriously mess with the ratings now.
My theory is this: It's just too easy to beat them once you get the basic concepts of the game. Their ratings won't be really low because with their reasonably good executions of mediocre strategies they beat up a lot of new players.

Maybe some players that actually play bots can weigh in on this one.

My theory is that the rating system is actually improperly calibrated, because it forces a normal distribution. If Tommy is good enough against Suzy that he wins 70% of the time, and Suzy is good enough that she beats Pat 70% of the time, then it expects Suzy would beat Pat 85.3% of the time. (Well, it tries to correct for this somewhat via the sigma parameters, but this is limited). The thing is, in Dominion, after a point, it's very very hard to achieve such a high threshhold. But it will try, via the Sigma, and over a wide enough scale, it screws things up.

tl;dr there are ranges of rating differences I expect people to outperform expectations for, and bots are probably the most egregious problem here.