Dominion Strategy Forum

Please login or register.

Login with username, password and session length
Pages: 1 ... 6 7 [8] 9 10 ... 13  All

Author Topic: Isotropish Leaderboard (alternative to Goko Pro)  (Read 144068 times)

0 Members and 1 Guest are viewing this topic.

Qvist

  • Mountebank
  • *****
  • Offline Offline
  • Posts: 2400
  • Shuffle iT Username: Qvist
  • Respect: +4085
    • View Profile
Re: Isotropish Leaderboard (alternative to Goko Pro)
« Reply #175 on: November 16, 2013, 07:17:37 pm »
0

I think the leaderboard stopped updating. If you have time, it would be great if you have a look at it.

"Last recorded game finished 132 min, 43 seconds ago"

ragingduckd

  • Board Moderator
  • *
  • Offline Offline
  • Posts: 1059
  • Respect: +3527
    • View Profile
Re: Isotropish Leaderboard (alternative to Goko Pro)
« Reply #176 on: November 16, 2013, 09:36:45 pm »
+1

I think the leaderboard stopped updating. If you have time, it would be great if you have a look at it.

"Last recorded game finished 132 min, 43 seconds ago"

I restarted it.  Dunno what went wrong.
Logged
Salvager Extension | Isotropish Leaderboard | Game Data | Log Search & other toys | Salvager Bug Reports

Salvager not working for me at all today. ... Please help! I can't go back to playing without it like an animal!

Kirian

  • Adventurer
  • ******
  • Offline Offline
  • Posts: 7096
  • Shuffle iT Username: Kirian
  • An Unbalanced Equation
  • Respect: +9411
    • View Profile
Re: Isotropish Leaderboard (alternative to Goko Pro)
« Reply #177 on: November 16, 2013, 09:39:14 pm »
+1

A crazy question about the calculations:  is there something intrinsic to the system that creates a floor for sigma?  There's literally no one--no matter how many games played--with a sigma under 9.92.  What's up with that?
Logged
Kirian's Law of f.DS jokes:  Any sufficiently unexplained joke is indistinguishable from serious conversation.

ragingduckd

  • Board Moderator
  • *
  • Offline Offline
  • Posts: 1059
  • Respect: +3527
    • View Profile
Re: Isotropish Leaderboard (alternative to Goko Pro)
« Reply #178 on: November 16, 2013, 10:01:46 pm »
0

A crazy question about the calculations:  is there something intrinsic to the system that creates a floor for sigma?  There's literally no one--no matter how many games played--with a sigma under 9.92.  What's up with that?

Honestly, I have no idea.  It seems suspicious to me too.

The code is here if anyone wants to look at it.  Basically I'm just outsourcing the job to this TrueSkill package.
Logged
Salvager Extension | Isotropish Leaderboard | Game Data | Log Search & other toys | Salvager Bug Reports

Salvager not working for me at all today. ... Please help! I can't go back to playing without it like an animal!

flies

  • Minion
  • *****
  • Offline Offline
  • Posts: 629
  • Shuffle iT Username: flies
  • Statistical mechanics of hard rods on a 1D lattice
  • Respect: +348
    • View Profile
    • ask the atheists
Re: Isotropish Leaderboard (alternative to Goko Pro)
« Reply #179 on: November 17, 2013, 09:07:20 am »
0

A crazy question about the calculations:  is there something intrinsic to the system that creates a floor for sigma?  There's literally no one--no matter how many games played--with a sigma under 9.92.  What's up with that?
This could be a natural property of the game.  I mean sometimes you do everything possible and still lose.
Logged
Gotta be efficient when most of your hand coordination is spent trying to apply mascara to your beard.
flies Dominionates on youtube

ragingduckd

  • Board Moderator
  • *
  • Offline Offline
  • Posts: 1059
  • Respect: +3527
    • View Profile
Re: Isotropish Leaderboard (alternative to Goko Pro)
« Reply #180 on: November 18, 2013, 10:52:04 pm »
0

A crazy question about the calculations:  is there something intrinsic to the system that creates a floor for sigma?  There's literally no one--no matter how many games played--with a sigma under 9.92.  What's up with that?
This could be a natural property of the game.  I mean sometimes you do everything possible and still lose.

Possibly, but the isotropic leaderboard didn't look quite like that.
Logged
Salvager Extension | Isotropish Leaderboard | Game Data | Log Search & other toys | Salvager Bug Reports

Salvager not working for me at all today. ... Please help! I can't go back to playing without it like an animal!

Warfreak2

  • Saboteur
  • *****
  • Offline Offline
  • Posts: 1149
  • KC->KC->[Scavenger, Scavenger, Lookout]
  • Respect: +1324
    • View Profile
    • Music what I do
Re: Isotropish Leaderboard (alternative to Goko Pro)
« Reply #181 on: November 19, 2013, 07:11:14 am »
0

Maybe there's a lower signal-to-noise ratio on Goko, with more games being lost because of bad connections, abandoned due to slowplaying, &c.?
Logged
If the only engine on the board is Procession->Conspirator, I will play it.

2.71828.....

  • Saboteur
  • *****
  • Offline Offline
  • Posts: 1290
  • Shuffle iT Username: irrationalE
  • Respect: +1322
    • View Profile
Re: Isotropish Leaderboard (alternative to Goko Pro)
« Reply #182 on: November 19, 2013, 07:55:09 am »
0

A crazy question about the calculations:  is there something intrinsic to the system that creates a floor for sigma?  There's literally no one--no matter how many games played--with a sigma under 9.92.  What's up with that?
This could be a natural property of the game.  I mean sometimes you do everything possible and still lose.

Possibly, but the isotropic leaderboard didn't look quite like that.
Could it possibly have something to do with the games played that count toward your rank?  The isotropic leaderboard includes all eligible games, so lespeutere, for example, has over 10,000 games counting toward his ranking.  Meanwhile since the isotropish includes only games played in the last month, lespeutere has just under 3000 games counting toward the isotropish leaderboard.  I haven't followed all of the implementation talk and such, so this is just a bystander's casual look at the rankings, but that seems to be the biggest difference in the two leaderboards.
Logged
Man. I had four strips of bacon yesterday. Was one automatically undercooked, one automatically overcooked? No, let's put a stop to that right here, all four strips were excellent.

Mic Qsenoch

  • 2015 DS Champion
  • *
  • Offline Offline
  • Posts: 1709
  • Respect: +4329
    • View Profile
Re: Isotropish Leaderboard (alternative to Goko Pro)
« Reply #183 on: November 19, 2013, 10:27:59 am »
0

Could it possibly have something to do with the games played that count toward your rank?  The isotropic leaderboard includes all eligible games, so lespeutere, for example, has over 10,000 games counting toward his ranking.  Meanwhile since the isotropish includes only games played in the last month, lespeutere has just under 3000 games counting toward the isotropish leaderboard.  I haven't followed all of the implementation talk and such, so this is just a bystander's casual look at the rankings, but that seems to be the biggest difference in the two leaderboards.

I think the isotropish leaderboard includes all eligible games in its ranking, it just doesn't display your name on the leaderboard if you haven't played a game in the last month.
Logged

yed

  • Minion
  • *****
  • Offline Offline
  • Posts: 620
  • Shuffle iT Username: yed
  • Respect: +571
    • View Profile
Re: Isotropish Leaderboard (alternative to Goko Pro)
« Reply #184 on: November 20, 2013, 07:18:31 pm »
0

AI: please restart it again: Last recorded game finished 40 min, 21 seconds ago
Logged

ragingduckd

  • Board Moderator
  • *
  • Offline Offline
  • Posts: 1059
  • Respect: +3527
    • View Profile
Re: Isotropish Leaderboard (alternative to Goko Pro)
« Reply #185 on: November 20, 2013, 09:16:22 pm »
0

AI: please restart it again: Last recorded game finished 40 min, 21 seconds ago

Ok, and I've turned on massive debugging too.  Until now I haven't even been able to find an error in the log.  The loop seems to just stop silently.
Logged
Salvager Extension | Isotropish Leaderboard | Game Data | Log Search & other toys | Salvager Bug Reports

Salvager not working for me at all today. ... Please help! I can't go back to playing without it like an animal!

soulnet

  • Mountebank
  • *****
  • Offline Offline
  • Posts: 2142
  • Respect: +1751
    • View Profile
Re: Isotropish Leaderboard (alternative to Goko Pro)
« Reply #186 on: December 03, 2013, 08:33:16 pm »
0

Is it possible that log gathering is not working? I have two games played about 10 minutes ago that do not appear in the search and the leaderboard was updated more than 30 minutes ago according to its own displayed info (every time I remember before was less than a minute ago).
Logged

ragingduckd

  • Board Moderator
  • *
  • Offline Offline
  • Posts: 1059
  • Respect: +3527
    • View Profile
Re: Isotropish Leaderboard (alternative to Goko Pro)
« Reply #187 on: December 05, 2013, 10:35:14 pm »
0

Is it possible that log gathering is not working? I have two games played about 10 minutes ago that do not appear in the search and the leaderboard was updated more than 30 minutes ago according to its own displayed info (every time I remember before was less than a minute ago).

Yes... it shut off again.  I restarted it.

Even with my in-depth logging, I can't figure out why this is happening.  I don't mind restarting it whenever people notice it's busted tho.
Logged
Salvager Extension | Isotropish Leaderboard | Game Data | Log Search & other toys | Salvager Bug Reports

Salvager not working for me at all today. ... Please help! I can't go back to playing without it like an animal!

Qvist

  • Mountebank
  • *****
  • Offline Offline
  • Posts: 2400
  • Shuffle iT Username: Qvist
  • Respect: +4085
    • View Profile
Re: Isotropish Leaderboard (alternative to Goko Pro)
« Reply #188 on: December 06, 2013, 06:19:43 pm »
0

Sorry to bother you AI. But there is something wrong for sure. The # of games isn't updating anymore.
I don't know if this is only a display bug or if the rating is also calculated wrong, but the # games doesn't rise anymore.

ragingduckd

  • Board Moderator
  • *
  • Offline Offline
  • Posts: 1059
  • Respect: +3527
    • View Profile
Re: Isotropish Leaderboard (alternative to Goko Pro)
« Reply #189 on: December 06, 2013, 10:07:43 pm »
0

Sorry to bother you AI. But there is something wrong for sure. The # of games isn't updating anymore.
I don't know if this is only a display bug or if the rating is also calculated wrong, but the # games doesn't rise anymore.

Restarted it again.

This is probably the same bug that's stopped the system the last several times.  I've looked for it to no avail and I don't expect to have time for a serious bug hunt any time soon.

Just PM me if/when this happens again, rather than posting here.  No need to clutter up the thread.
Logged
Salvager Extension | Isotropish Leaderboard | Game Data | Log Search & other toys | Salvager Bug Reports

Salvager not working for me at all today. ... Please help! I can't go back to playing without it like an animal!

rspeer

  • Witch
  • *****
  • Offline Offline
  • Posts: 469
  • Respect: +877
    • View Profile
Re: Isotropish Leaderboard (alternative to Goko Pro)
« Reply #190 on: March 31, 2014, 10:52:57 pm »
0

A crazy question about the calculations:  is there something intrinsic to the system that creates a floor for sigma?  There's literally no one--no matter how many games played--with a sigma under 9.92.  What's up with that?
This could be a natural property of the game.  I mean sometimes you do everything possible and still lose.

Possibly, but the isotropic leaderboard didn't look quite like that.
Could it possibly have something to do with the games played that count toward your rank?  The isotropic leaderboard includes all eligible games, so lespeutere, for example, has over 10,000 games counting toward his ranking.  Meanwhile since the isotropish includes only games played in the last month, lespeutere has just under 3000 games counting toward the isotropish leaderboard.  I haven't followed all of the implementation talk and such, so this is just a bystander's casual look at the rankings, but that seems to be the biggest difference in the two leaderboards.

It's pretty much a natural property of TrueSkill. There is a point at which you can't decrease your uncertainty any more, because the uncertainty increase per game balances the uncertainty decrease you get by playing.

This happened on Isotropic too. It had two different sources of uncertainty, though, one that was per-game and one that was per-day. The players with games >> days could get their uncertainty down to 6-ish.
Logged

Holger

  • Minion
  • *****
  • Offline Offline
  • Posts: 736
  • Respect: +458
    • View Profile
Re: Isotropish Leaderboard (alternative to Goko Pro)
« Reply #191 on: April 01, 2014, 04:32:58 pm »
+1

A crazy question about the calculations:  is there something intrinsic to the system that creates a floor for sigma?  There's literally no one--no matter how many games played--with a sigma under 9.92.  What's up with that?
This could be a natural property of the game.  I mean sometimes you do everything possible and still lose.

Possibly, but the isotropic leaderboard didn't look quite like that.
Could it possibly have something to do with the games played that count toward your rank?  The isotropic leaderboard includes all eligible games, so lespeutere, for example, has over 10,000 games counting toward his ranking.  Meanwhile since the isotropish includes only games played in the last month, lespeutere has just under 3000 games counting toward the isotropish leaderboard.  I haven't followed all of the implementation talk and such, so this is just a bystander's casual look at the rankings, but that seems to be the biggest difference in the two leaderboards.

It's pretty much a natural property of TrueSkill. There is a point at which you can't decrease your uncertainty any more, because the uncertainty increase per game balances the uncertainty decrease you get by playing.

This happened on Isotropic too. It had two different sources of uncertainty, though, one that was per-game and one that was per-day. The players with games >> days could get their uncertainty down to 6-ish.

But the per-day source only ever increased the uncertainty on Isotropic, according to Doug; so how could people get uncertainties substantially below 9.9 on Isotropic, but not on Isotropish?
(I don't think this can only be due to the number of games played either - e.g. jog now has more games on oko, but a much lower uncertainty (8.6) on Isotropic.)
Logged

7string

  • Baron
  • ****
  • Offline Offline
  • Posts: 56
  • Shuffle iT Username: 7string
  • Respect: +57
    • View Profile
Re: Isotropish Leaderboard (alternative to Goko Pro)
« Reply #192 on: April 03, 2014, 03:35:28 pm »
+2

I noticed that as of today, the Isotropish Leaderboard only displays the top 100 ranked players.  It only displays the top 100 now regardless of whether you select partial or full displays.  As a newer player, I have enjoyed being able to reference the Isotropish Leaderboard to review my stats as I try to work my way up the leaderboard, but I'm still down around the 400 rank or so, and now I can no longer see my stats or those of my other family and friends who are also not in the top 100.  Anybody know if this is a known issue, or if there is someplace else I should report it.  I don't really do any coding myself, so do not have a sign-on on Github.  Thanks.
Logged

ragingduckd

  • Board Moderator
  • *
  • Offline Offline
  • Posts: 1059
  • Respect: +3527
    • View Profile
Re: Isotropish Leaderboard (alternative to Goko Pro)
« Reply #193 on: April 03, 2014, 03:41:36 pm »
+2

I noticed that as of today, the Isotropish Leaderboard only displays the top 100 ranked players.  It only displays the top 100 now regardless of whether you select partial or full displays.  As a newer player, I have enjoyed being able to reference the Isotropish Leaderboard to review my stats as I try to work my way up the leaderboard, but I'm still down around the 400 rank or so, and now I can no longer see my stats or those of my other family and friends who are also not in the top 100.  Anybody know if this is a known issue, or if there is someplace else I should report it.  I don't really do any coding myself, so do not have a sign-on on Github.  Thanks.

Whoops.  Sorry about that.  It's back to normal now.

Thanks for letting me know.
Logged
Salvager Extension | Isotropish Leaderboard | Game Data | Log Search & other toys | Salvager Bug Reports

Salvager not working for me at all today. ... Please help! I can't go back to playing without it like an animal!

7string

  • Baron
  • ****
  • Offline Offline
  • Posts: 56
  • Shuffle iT Username: 7string
  • Respect: +57
    • View Profile
Re: Isotropish Leaderboard (alternative to Goko Pro)
« Reply #194 on: April 03, 2014, 04:06:56 pm »
0

I noticed that as of today, the Isotropish Leaderboard only displays the top 100 ranked players.  It only displays the top 100 now regardless of whether you select partial or full displays.  As a newer player, I have enjoyed being able to reference the Isotropish Leaderboard to review my stats as I try to work my way up the leaderboard, but I'm still down around the 400 rank or so, and now I can no longer see my stats or those of my other family and friends who are also not in the top 100.  Anybody know if this is a known issue, or if there is someplace else I should report it.  I don't really do any coding myself, so do not have a sign-on on Github.  Thanks.

Whoops.  Sorry about that.  It's back to normal now.

Thanks for letting me know.

Wow...fast response!  Thanks much!!...I rarely got that fast of service even when I was an IT manager. ; )
Logged

WanderingWinder

  • Adventurer
  • ******
  • Offline Offline
  • Posts: 5275
  • ...doesn't really matter to me
  • Respect: +4381
    • View Profile
    • WanderingWinder YouTube Page
Re: Isotropish Leaderboard (alternative to Goko Pro)
« Reply #195 on: April 04, 2014, 07:25:14 am »
+1

Here are a couple of videos from one of the three guys who developed TrueSkill. He talks about some other stuff, too.
(The first link has three videos - second video there is where most of the TS stuff is for that link, though he sets up mathematical principlas in the first part).

http://videolectures.net/acml2013_herbrich_real_time_bayesian_learning/

http://videolectures.net/acml2013_herbrich_technology_transfer/

Anyway, might post some more thoughts on this later, but thought people here might be interested.

Polk5440

  • Torturer
  • *****
  • Offline Offline
  • Posts: 1708
  • Respect: +1788
    • View Profile
Re: Isotropish Leaderboard (alternative to Goko Pro)
« Reply #196 on: April 04, 2014, 07:38:35 am »
0

Here are a couple of videos from one of the three guys who developed TrueSkill. He talks about some other stuff, too.
(The first link has three videos - second video there is where most of the TS stuff is for that link, though he sets up mathematical principlas in the first part).

http://videolectures.net/acml2013_herbrich_real_time_bayesian_learning/

http://videolectures.net/acml2013_herbrich_technology_transfer/

Anyway, might post some more thoughts on this later, but thought people here might be interested.

Related:

Microsoft Research has a lot of good material on TrueSkill. For instance, here is an overview, a detailed description, a lengthy article, and an academic-style article on the system.
Logged

Holger

  • Minion
  • *****
  • Offline Offline
  • Posts: 736
  • Respect: +458
    • View Profile
Re: Isotropish Leaderboard (alternative to Goko Pro)
« Reply #197 on: April 04, 2014, 05:16:32 pm »
0

A crazy question about the calculations:  is there something intrinsic to the system that creates a floor for sigma?  There's literally no one--no matter how many games played--with a sigma under 9.92.  What's up with that?
This could be a natural property of the game.  I mean sometimes you do everything possible and still lose.

Possibly, but the isotropic leaderboard didn't look quite like that.
Could it possibly have something to do with the games played that count toward your rank?  The isotropic leaderboard includes all eligible games, so lespeutere, for example, has over 10,000 games counting toward his ranking.  Meanwhile since the isotropish includes only games played in the last month, lespeutere has just under 3000 games counting toward the isotropish leaderboard.  I haven't followed all of the implementation talk and such, so this is just a bystander's casual look at the rankings, but that seems to be the biggest difference in the two leaderboards.

It's pretty much a natural property of TrueSkill. There is a point at which you can't decrease your uncertainty any more, because the uncertainty increase per game balances the uncertainty decrease you get by playing.

This happened on Isotropic too. It had two different sources of uncertainty, though, one that was per-game and one that was per-day. The players with games >> days could get their uncertainty down to 6-ish.

But the per-day source only ever increased the uncertainty on Isotropic, according to Doug; so how could people get uncertainties substantially below 9.9 on Isotropic, but not on Isotropish?
(I don't think this can only be due to the number of games played either - e.g. jog now has more games on Goko, but a much lower uncertainty (8.6) on Isotropic.)

I think I've found the explanation (right in this thread  ;)):
so I still don't understand what γ = σ0 / 100 (applied daily) means.

In the early days Iso would increase the variance after each game.  Since that put a hard floor on how low the variance could go, it was later changed to happen once a day.  So "applied daily" means that you only fudge the variance once a day, rather than recalculating the (constant) gamma daily.

So apparently Isotropic only increased the variance per day, not also per game. Therefore its variances varied much more, depending on the number of games played per day. On Isotropish, the increase per game leads to an absolute floor, as we currently observe. Players with >1 game/day usually had lower variances on Isotropic, players with <1 game/day usually have lower variances now.
Logged

ragingduckd

  • Board Moderator
  • *
  • Offline Offline
  • Posts: 1059
  • Respect: +3527
    • View Profile
Re: Isotropish Leaderboard (alternative to Goko Pro)
« Reply #198 on: April 04, 2014, 07:15:33 pm »
0

I think I've found the explanation (right in this thread  ;)):
so I still don't understand what γ = σ0 / 100 (applied daily) means.

In the early days Iso would increase the variance after each game.  Since that put a hard floor on how low the variance could go, it was later changed to happen once a day.  So "applied daily" means that you only fudge the variance once a day, rather than recalculating the (constant) gamma daily.

So apparently Isotropic only increased the variance per day, not also per game. Therefore its variances varied much more, depending on the number of games played per day. On Isotropish, the increase per game leads to an absolute floor, as we currently observe. Players with >1 game/day usually had lower variances on Isotropic, players with <1 game/day usually have lower variances now.

I don't think I understand.  Isotropish doesn't increase variance after each game either.  I would expect that the more games you play per day under dougz's system, the closer you get to the floor your variance hits on Isotropish.  I don't get how your variance could actually get lower.
« Last Edit: April 04, 2014, 07:19:17 pm by ragingduckd »
Logged
Salvager Extension | Isotropish Leaderboard | Game Data | Log Search & other toys | Salvager Bug Reports

Salvager not working for me at all today. ... Please help! I can't go back to playing without it like an animal!

Holger

  • Minion
  • *****
  • Offline Offline
  • Posts: 736
  • Respect: +458
    • View Profile
Re: Isotropish Leaderboard (alternative to Goko Pro)
« Reply #199 on: April 05, 2014, 12:09:43 pm »
+1

I don't think I understand.  Isotropish doesn't increase variance after each game either.  I would expect that the more games you play per day under dougz's system, the closer you get to the floor your variance hits on Isotropish.  I don't get how your variance could actually get lower.

I haven't looked at your code, but you said previously that you just used the vanilla TrueSkill algorithm, which does increase the variance (by GAMMA^2) once for each game:

Quote from: http://research.microsoft.com/en-us/projects/trueskill/faq.aspx
So, what is going on here? Between any two games of a gamer, the TrueSkill ranking system assumes that the true skill of a gamer, that is, μ, can have changed slightly either up or down; this property is what allows the ranking system to adapt to a change in the skill of a gamer. Technically, this is achieved by a small increase in the σ of each participating gamer before the game outcome is incorporated.

Quote
gamma is a small amount by which a player's uncertainty (sigma) is
  increased prior to the start of each game.  This allows us to
  account for skills that vary over time; the effect of old games
  on the estimate will slowly disappear unless reinforced by evidence
  from new games.

This is a plausible algorithm, but I don't see it in dougz's code
So this line
Quote
sigma=sqrt(pl.skill[1] ** 2 + GAMMA ** 2)
https://github.com/dougz/trueskill/blob/master/trueskill.py#L345
does something else?

Yeah, that updating is a normal part of the Vanilla TS algorithm. It happens once per game, not once per day. But going back to the original quote:

Quote
(For those interested in the details, I've set β = 25, γ = σ0 / 100 (applied daily), and the draw probability at 5%.)

... I understand your original statement now. I agree that dougz may have meant that he's doing the once-a-day updating in addition to or instead of the once-per-game updating. I think I'll wait until he chimes in to make any such changes though.

(Bold by me.) So you apply GAMMA once per game, while Doug (according to qmech's quote above) ended up applying it only once per day, not per game.
Logged
Pages: 1 ... 6 7 [8] 9 10 ... 13  All
 

Page created in 0.065 seconds with 21 queries.