Dominion Strategy Forum

Dominion => Dominion Online at Shuffle iT => Dominion General Discussion => Goko Dominion Online => Topic started by: GeoLib on August 28, 2014, 08:47:08 pm

Title: Update 28 August 2014
Post by: GeoLib on August 28, 2014, 08:47:08 pm
http://forum.makingfun.com/showthread.php?5097-Release-Notes-8-28-2014

Quote from:  Jeff (Developer)
This update includes these improvements:

Added Prince promo card. Now available in the store. Notes:

    You can now choose the order of cards that take effect at the start of the turn if the ordering would make any difference. The cards that do this are Prince, Horse Traders played as a reaction, and duration cards. For example, you may choose to play your princed Menagerie, then your Caravan duration card, then your princed Chapel.
    A princed Hermit follows the same pattern as is already used for a schemed Hermit: if you didn't buy anything on a turn where a princed Hermit is played, the Prince's on-discard effect for the Hermit takes place before the Hermit's own effect. This will cause you to gain a Madman, but the Hermit won't be trashed because the Hermit will have lost track of itself.


Fixed bug: Changing email in "Edit Profile" was not working correctly.

Fixed bug causing guests to sometimes not be able to play games.

Adventures

    Changed adventures so each player starts with the same starting hand, so you do not need to zap down your opponent. There are also no starting zaps.
    Fixed: Adventures in Dark Ages begin with three estates AND the three shelters.
    Other improvements involving required cards.


Bots

    Play vs. Bots now uses the Pro rating system, which is used for fair games. If you want to play bots under the Casual system, you still can by creating a table with the casual system and inviting a bot.
    Fixed bug when AI trashed a Catacomb during the Buy Phase (via Farmland or Doctor).
    Fixed bug: AI would sometimes waste coin tokens when it could buy a Grand Market but decided not to.
    Improvements for Highway and Courtyard.


Fixed bug: Coppersmith wouldn't change coin display of Coppers until the Buy Phase.

Changed sound for Gold/Platinum to occur only when gaining the card, not when playing it.

Messages

    Changed status message and log message for Philosopher's Stone to show the count of your deck plus discards pile.
    Fixed discard message for Count.


"Support" link on home screen replaced by "Forum" and "Report a problem" links.

Optimizations, especially in the size of internal messages during a game.

So how does making "Play Bots" pro make things more fair? I really don't see any reason why you shouldn't get an option when clicking "Play Bots" to choose pro, casual, or unrated. Opening the multiplayer, setting up a game, and adding a bot is actually significantly more work (mostly because the lobby is so absurdly slow).

And finally semi-fixed Adventures!

Also, you should be able to choose the order of the schemed/princed hermit's effects. If you want to trash it you should be able to. There probably aren't many situations when you do, but what if you don't want to gain any more 3-costs? Ah well; a minor issue and in most cases this way of doing it makes the game more playable with fewer decisions.
Title: Re: Update 8 August 2014
Post by: shark_bait on August 28, 2014, 09:33:42 pm
Well that explains why my Goko rating dropped 400 points.  All those "I'll just resign" 4-player bot games.
Title: Re: Update 8 August 2014
Post by: pst on August 28, 2014, 10:43:09 pm
http://forum.makingfun.com/showthread.php?5097-Release-Notes-8-28-2014

Quote from:  Jeff (Developer)
    Play vs. Bots now uses the Pro rating system, which is used for fair games. If you want to play bots under the Casual system, you still can by creating a table with the casual system and inviting a bot.

So how does making "Play Bots" pro make things more fair?

It's the other way round. Games against bots are usually played as "fair" (that is random) games, so they should use the rating that is used for fair games.

This is a good step. The previous maintainer thought that the game of Dominion was about creating interesting tables that you wanted to play, which you could save to play over and over again. "We" knew that the game is about adapting to a board that is unknown to all players before the game, and that a rating system for games played with a pre-set tables isn't serious. As an afterthought the special "Pro" rating was created for "us". Really it should be the only rating, and I see this as a move towards making "Pro" the default rating. Many casual players now don't have "Pro" ratings at all, assuming from the name that "Pro" games are for the "Pros". With this change many more will get "Pro" ratings.
Title: Re: Update 8 August 2014
Post by: dondon151 on August 28, 2014, 10:49:28 pm
I'm still waiting for a way to get enough VP tokens to buy Prince.
Title: Re: Update 8 August 2014
Post by: pst on August 28, 2014, 11:04:11 pm
This is a good step.

Except it's not working. At least I just tried a couple of bot games, and they were played with "Casual" rating.
Title: Re: Update 8 August 2014
Post by: DG on August 28, 2014, 11:10:51 pm
Bots should be casual. They are not pros. They can be manipulated easily for ranking points.

If casual players want to know how good they are compared to the bots then the casual rankings should give them that. Experienced players want to know how good they are compared to other humans and the pro ranking should give them that. No ranking legitimately does that now within Goko.

I played a few adventurers earlier and it changed my casual rank, pro rank, and left my adventure rank unchanged (maybe provisional).
Title: Re: Update 8 August 2014
Post by: pst on August 28, 2014, 11:26:49 pm
Bots should be casual. They are not pros. They can be manipulated easily for ranking points.

No one is a pro and the "pro" games are only called that because Goko didn't understand what the game was about, so they already had a bogus rating and had to call the real rating and the real way of playing something special that is intimidating to some.

I don't understand what you mean with manipulation. On many boards you can win very easily against a bad player. That is regardless of whether that bad player is a human or a bot, and that isn't manipulation. The main difference between bots and other not-so-good players are that the bots play way more which has several effects, one being that it's easier to know about their particular weaknesses.
Title: Re: Update 8 August 2014
Post by: Beyond Awesome on August 28, 2014, 11:40:18 pm
I'm still waiting for a way to get enough VP tokens to buy Prince.


Never gonna happen.
Title: Re: Update 28 August 2014
Post by: michaeljb on August 29, 2014, 12:27:58 am
Update 8 August 2014

ಠ_ಠ
Title: Re: Update 28 August 2014
Post by: GeoLib on August 29, 2014, 01:18:26 am
Update 8 August 2014

ಠ_ಠ

So I just looked at my post several times to confirm that it said 28 before realizing that it's only correct because you edited it. Haha, woops! Thanks for catching that.
Title: Re: Update 8 August 2014
Post by: serakfalcon on August 29, 2014, 03:50:00 am
I'm still waiting for a way to get enough VP tokens to buy Prince.


Never gonna happen.

I bought my cards late and I saved 100 shields specifically for prince. But it costs 200  :'(
Title: Re: Update 8 August 2014
Post by: Donald X. on August 29, 2014, 06:05:27 am
No one is a pro and the "pro" games are only called that because Goko didn't understand what the game was about, so they already had a bogus rating and had to call the real rating and the real way of playing something special that is intimidating to some.

I don't understand what you mean with manipulation. On many boards you can win very easily against a bad player. That is regardless of whether that bad player is a human or a bot, and that isn't manipulation. The main difference between bots and other not-so-good players are that the bots play way more which has several effects, one being that it's easier to know about their particular weaknesses.
I am sympathetic to this viewpoint. The bots may seem less fair because they never learn - you can beat one up with Dukes and then just do it again next game. There are people like that too though. The bots aren't cheating. They are opponents of a particular skill level, that only changes due to programming. They're ranked. Beating up on bots should be no easier than beating up on similarly low-ranked players. And when you know they lose to Ambassador-Rats or whatever, well you don't get to pick the cards.
Title: Re: Update 28 August 2014
Post by: Tables on August 29, 2014, 07:29:14 am
Well, I'm glad I saved my VP tokens. Time to go test out Prince I suppose...

Edit: Or it could cost 200 VP tokens. Wow. I'm like 40 short and can't earn any more.
Title: Re: Update 28 August 2014
Post by: werothegreat on August 29, 2014, 09:00:22 am
I just coughed up $4 to buy it outright.
Title: Re: Update 28 August 2014
Post by: Breezy D on August 29, 2014, 11:43:13 am
Same here.

And with the spare change, maybe one day I'll be able to purchase walled village rather than go all out to finish those adventures.

Title: Re: Update 28 August 2014
Post by: Holger on September 10, 2014, 08:15:47 am
Quote from:  Jeff (Developer)
Bots

    Play vs. Bots now uses the Pro rating system, which is used for fair games. If you want to play bots under the Casual system, you still can by creating a table with the casual system and inviting a bot.

This is not true, unfortunately. In the multiplayer lobby, when I create a game and choose "Add bot", I am shown an empty list, no matter whether I choose Pro, Casual or Unrated.
(It makes some sense not to allow Casual bot games since you can exploit their weaknesses on specific boards to reach the top of the Casual leaderboard, but at least Unrated bot games should be possible.)
Title: Re: Update 28 August 2014
Post by: ThaddeusB on September 10, 2014, 12:19:02 pm
Quote from:  Jeff (Developer)
Bots

    Play vs. Bots now uses the Pro rating system, which is used for fair games. If you want to play bots under the Casual system, you still can by creating a table with the casual system and inviting a bot.

This is not true, unfortunately. In the multiplayer lobby, when I create a game and choose "Add bot", I am shown an empty list, no matter whether I choose Pro, Casual or Unrated.
(It makes some sense not to allow Casual bot games since you can exploit their weaknesses on specific boards to reach the top of the Casual leaderboard, but at least Unrated bot games should be possible.)

That happens occasionally - likely the bots are offline.  It has nothing to do with the update.
Title: Re: Update 28 August 2014
Post by: Holger on September 10, 2014, 12:52:10 pm
Quote from:  Jeff (Developer)
Bots

    Play vs. Bots now uses the Pro rating system, which is used for fair games. If you want to play bots under the Casual system, you still can by creating a table with the casual system and inviting a bot.

This is not true, unfortunately. In the multiplayer lobby, when I create a game and choose "Add bot", I am shown an empty list, no matter whether I choose Pro, Casual or Unrated.
(It makes some sense not to allow Casual bot games since you can exploit their weaknesses on specific boards to reach the top of the Casual leaderboard, but at least Unrated bot games should be possible.)

That happens occasionally - likely the bots are offline.  It has nothing to do with the update.

This has never happened to me before; why would they be offline for hours? And I can still play bots from the starting page, but that button has now returned to Casual bot games! Is there another Update/"Downdate" going on right now?
Title: Re: Update 28 August 2014
Post by: ThaddeusB on September 10, 2014, 12:53:52 pm
Quote from:  Jeff (Developer)
Bots

    Play vs. Bots now uses the Pro rating system, which is used for fair games. If you want to play bots under the Casual system, you still can by creating a table with the casual system and inviting a bot.

This is not true, unfortunately. In the multiplayer lobby, when I create a game and choose "Add bot", I am shown an empty list, no matter whether I choose Pro, Casual or Unrated.
(It makes some sense not to allow Casual bot games since you can exploit their weaknesses on specific boards to reach the top of the Casual leaderboard, but at least Unrated bot games should be possible.)

That happens occasionally - likely the bots are offline.  It has nothing to do with the update.

This has never happened to me before; why would they be offline for hours? And I can still play bots from the starting page, but that button has now returned to Casual bot games! Is there another Update/"Downdate" going on right now?

I've seen the no bots thing many, many times before the update and have seen it work properly many times since the update.  I can't answer your why questions, as I have no inside knowledge.
Title: Re: Update 28 August 2014
Post by: ragingduckd on September 11, 2014, 12:13:17 am
Quote from:  Jeff (Developer)
Play vs. Bots now uses the Pro rating system, which is used for fair games. If you want to play bots under the Casual system, you still can by creating a table with the casual system and inviting a bot.

Hmm... so the bad news that now I have to actually play out the game when I want to test a feature.  The other bad news is that I lose 30 points even for a win:

(http://forum.dominionstrategy.com/index.php?action=dlattach;topic=11682.0;attach=1882)

I'm pretty sure this shouldn't be possible unless someone at Goko has been messing with their TS parameters.  And they'd have to be pretty insane parameter at that.  Is this just a bug?
Title: Re: Update 28 August 2014
Post by: jsh357 on September 11, 2014, 12:18:27 am
I have been seeing the same--points loss even when I beat the bots.  It's frustrating, but hard to get too invested since the whole leaderboard seems out of whack at this point anyway.
Title: Re: Update 28 August 2014
Post by: Holger on September 11, 2014, 07:11:24 am
I've seen the no bots thing many, many times before the update and have seen it work properly many times since the update.  I can't answer your why questions, as I have no inside knowledge.

Now the bots are back. Thanks anyway!

Quote from:  Jeff (Developer)
Play vs. Bots now uses the Pro rating system, which is used for fair games. If you want to play bots under the Casual system, you still can by creating a table with the casual system and inviting a bot.

Hmm... so the bad news that now I have to actually play out the game when I want to test a feature. 

Why? Can't you resign games any more?

Quote
The other bad news is that I lose 30 points even for a win:

(http://forum.dominionstrategy.com/index.php?action=dlattach;topic=11682.0;attach=1882)

I'm pretty sure this shouldn't be possible unless someone at Goko has been messing with their TS parameters.  And they'd have to be pretty insane parameter at that.  Is this just a bug?

I thought you were the one who created a rating predictor for Goko's rating system, so you should be able to check. ;)  (Well, more or less, since the bots' ratings change rapidly...)
If you played Serf Bot, it's quite likely that you'd lose points for a win due to an increased uncertainty; against other bots, it's less likely but not impossible if you have very low uncertainty.

Edit: I forgot, Goko was "perverting" the rating system to "hide" any point losses after a win. So have they just returned to unmodified TrueSkill now, or have they also changed their TrueSkill parameters for the worse?
(There's really more important things for them to do, like removing the pro rating inflations due to the recent adventure games bug. Or finally getting rid of the lobbies...  >:()
Title: Re: Update 28 August 2014
Post by: Kirian on September 11, 2014, 09:17:38 am
The better question is this:

They've already said they're essentially rewriting the whole thing from scratch.  Why even think about discussing the possibility of making these sorts of changes?
Title: Re: Update 28 August 2014
Post by: ragingduckd on September 11, 2014, 11:56:31 am
If you played Serf Bot, it's quite likely that you'd lose points for a win due to an increased uncertainty; against other bots, it's less likely but not impossible if you have very low uncertainty.

Edit: I forgot, Goko was "perverting" the rating system to "hide" any point losses after a win. So have they just returned to unmodified TrueSkill now, or have they also changed their TrueSkill parameters for the worse?

A -30 change is way too big.  Under the old system, a 7000 player might have lost something like 4 points of displayed rating by beating Serf Bot.  This is something different.

The better question is this:

They've already said they're essentially rewriting the whole thing from scratch.  Why even think about discussing the possibility of making these sorts of changes?

Ah.  Yes, it seems that they've changed the system.  I just played a test game and got some pretty crazy results:

Edit: Ack!  No, I'm completely wrong.  I wrote a bug into my code that screwed up all my numbers.  Here's my original incorrect post:

Estimating TS parameters (beta, tau, draw_prob) using Game 1...
Error minimization finished.
Residual Error: 41.8109

Initial ratings:
 A: 5995.29 +/- 279.27
 B:  963.13 +/- 1344.96
Expected post-game ratings:
 A: 5995.32 +/- 279.26
 B:  962.48 +/- 1343.81
Observed post-game ratings:
 A: 5996.90 +/- 280.42
 B:  926.24 +/- 1323.07
Rating prediction error: 1748.1527


First, my optimizer can no longer find the TS parameters (residual error should be zero).  Second, the old Goko TS parameters don't predict the rating changes correctly anymore.  Third, the new results are insane.

This was a game between a 5400 player and a 0 player and I resigned immediately as the 0 player.  Under the old Goko Pro system, the 0 player (actually 963.13 +/- 1344.96) would have lost almost nothing.  Instead, he dropped 37 points of mu.  The 5400 player (actually 5995.29 +/- 279.27) also gained a "lot" more than he would have under the old system.  Before, he only would have moved to mu = 5995.32, a gain of 0.03.  Instead he went to 5996.90 for a gain of 1.61.

In brief, both players moved by about 57x as much as they would have in the old days.  So either is WAY more volatile, or it's doing something pretty weird for players with very different ratings.  And since my optimizer no longer works, it's either it's no longer pure TrueSkill or their new parameters are REALLY out there.

So here's my pure, uninformed guess: They're still running TS, but they've tweaked it.  Those tweaks are either buggy or they've only tested them on similarly-rated players, so we're getting some crazy results when the rating differences are large.

Anybody want to collect some data for me?  I need ratings for both you and your opponent before (or during) and after the game.  Open a JS console and run this command, using your own player names instead of "guest4261":

conn.getPlayerId({developerId:'', playerName:'guest4261'}).then(function(x) { conn.getRating({
      version: 1,
      playerId: x.playerId,
      ratingSystemId: mtgRoom.options.ratingSystemPro
  }).then(function (resp) {
      console.log(resp.ratingData);
  });
})


And here's the output now that I've fixed my code.  It looks like the rating system hasn't been changed.

Code: [Select]
Estimating TS parameters (beta, tau, draw_prob) using Game 1...
Error minimization finished.
Residual Error: 0.0000

Error-minimizing Parameters:
beta:      1375.00
tau:         27.50
draw_prob:    0.05

Initial ratings:
 A: 5995.29 +/- 279.27
 B:  963.13 +/- 1344.96
Expected post-game ratings:
 A: 5996.90 +/- 280.42
 B:  926.24 +/- 1323.07
Observed post-game ratings:
 A: 5996.90 +/- 280.42
 B:  926.24 +/- 1323.07
Rating prediction error: 0.0000
Title: Re: Update 28 August 2014
Post by: rrenaud on September 11, 2014, 12:02:43 pm
Can you talk to goko directly?  Why don't this listen to you?

(Better Q, why aren't they paying you, but I think that's a different discussion :P ).
Title: Re: Update 28 August 2014
Post by: Holger on September 11, 2014, 12:04:02 pm
...

I see, that's too bad. Maybe they've increased rating changes for a while to get rid of the adventure rating bug's aftermath?
Title: Re: Update 28 August 2014
Post by: ragingduckd on September 11, 2014, 12:27:28 pm
Can you talk to goko directly?  Why don't this listen to you?

(Better Q, why aren't they paying you, but I think that's a different discussion :P ).

I've tried it on a few occasions.  I mostly just get ignored.  I sent this to a developer on MF a few days before I posted their rating algorithm.  I never got a response.

(http://forum.dominionstrategy.com/index.php?action=dlattach;topic=11682.0;attach=1885)

I've blacked out the developer's name and the quote from one of their forum posts.  No point in picking out an individual when it's the institution that's at fault.
Title: Re: Update 28 August 2014
Post by: theblankman on September 11, 2014, 01:49:23 pm
Anybody want to collect some data for me?  I need ratings for both you and your opponent before (or during) and after the game.  Open a JS console and run this command, using your own player names instead of "guest4261":

Code: [Select]
conn.getPlayerId({developerId:'', playerName:'guest4261'}).then(function(x) { conn.getRating({
      version: 1,
      playerId: x.playerId,
      ratingSystemId: mtgRoom.options.ratingSystemPro
  }).then(function (resp) {
      console.log(resp.ratingData);
  });
})
Would it be worthwhile to have Salvager do that for you, i.e. hook to the beginning and end of games to get the user's rating change and send it to a DB on gokosalvager.com (if you already maintain a DB of games, maybe just add columns for initial and final rating)?  Could you pull the opponent's starting and ending ratings with similar code?
Title: Re: Update 28 August 2014
Post by: ragingduckd on September 11, 2014, 01:58:23 pm
Guys, I'm really sorry.  My earlier post about the changes to the rating system was completely wrong.  I had a sneaky Python bug that left the code running without visible errors, but which produced all the wrong numbers.

I have no reason to think the rating system has changed.  The rating changes I'm getting still jive with Goko Pro as I described it back in April.  As for my -30 game against Serf Bot, I still have no idea how that happened.

(Better Q, why aren't they paying you, but I think that's a different discussion :P ).

Maybe this is why. :P
Title: Re: Update 28 August 2014
Post by: ThaddeusB on September 11, 2014, 06:53:56 pm
Quote from:  Jeff (Developer)
Play vs. Bots now uses the Pro rating system, which is used for fair games. If you want to play bots under the Casual system, you still can by creating a table with the casual system and inviting a bot.

Hmm... so the bad news that now I have to actually play out the game when I want to test a feature.  The other bad news is that I lose 30 points even for a win:

(http://forum.dominionstrategy.com/index.php?action=dlattach;topic=11682.0;attach=1882)

I'm pretty sure this shouldn't be possible unless someone at Goko has been messing with their TS parameters.  And they'd have to be pretty insane parameter at that.  Is this just a bug?

Normally you get 15 coins for a win, 10 for a tie.  Perhaps it was graded as a tie for some strange reason? When you tie, it still says "you're the winner".

Another possibility is the ~6 points a day you lose to inactivity are shown on the next game played.  Maybe you gained 0 for the game, but had lost 30 points since you last played.
Title: Re: Update 28 August 2014
Post by: AdamH on September 12, 2014, 09:05:38 am
Hmm... so the bad news that now I have to actually play out the game when I want to test a feature.

Presenting the quickest kingdom to play out a game (in time elapsed, not number of turns): Council Room, Smithy, Harvest, Woodcutter, Outpost, Feast, Coppersmith, Secret Chamber, Transmute, Peddler. Province/Estates.

All actions are terminals (besides Peddler which will never be bought over Gold), Council Room helps your opponent, there are no attacks, and there are few gainers or cards that require meaningful choices or take time to resolve. Contenders welcome, I refuse to test this or alternate submissions.

When I was grinding for coins to use on zaps for the adventures back in the day, I played 6P games with 5 bots that had Hermit/Market Square on them. Those games seemed to go by relatively quickly.
Title: Re: Update 28 August 2014
Post by: Holger on September 13, 2014, 04:42:32 pm
Guys, I'm really sorry.  My earlier post about the changes to the rating system was completely wrong.  I had a sneaky Python bug that left the code running without visible errors, but which produced all the wrong numbers.

I have no reason to think the rating system has changed.  The rating changes I'm getting still jive with Goko Pro as I described it back in April.  As for my -30 game against Serf Bot, I still have no idea how that happened.



(Better Q, why aren't they paying you, but I think that's a different discussion :P ).

Maybe this is why. :P

I thought it was a requirement for employment at Goko to make mistakes?  :P :P


Normally you get 15 coins for a win, 10 for a tie.  Perhaps it was graded as a tie for some strange reason? When you tie, it still says "you're the winner".

No, you only get 10 coins for a win against a bot (I think you get 15 coins only for wins against human opponents).
A tie would explain the rating loss, though I'd expect it to be even more than 30 points then.
Title: Re: Update 28 August 2014
Post by: ThaddeusB on September 15, 2014, 12:59:16 pm

Normally you get 15 coins for a win, 10 for a tie.  Perhaps it was graded as a tie for some strange reason? When you tie, it still says "you're the winner".

No, you only get 10 coins for a win against a bot (I think you get 15 coins only for wins against human opponents).
A tie would explain the rating loss, though I'd expect it to be even more than 30 points then.

I am now confident it was due to increased uncertainty after not playing for several days that carried the loss. Evidently displayed rating only changes when you play a game while true rating changes daily . Thus if you don't play for a few days a dozen or more point loss (30 here) will be incorporated into your next game.
Title: Re: Update 28 August 2014
Post by: Titandrake on September 15, 2014, 01:07:48 pm
Hmm... so the bad news that now I have to actually play out the game when I want to test a feature.

Presenting the quickest kingdom to play out a game (in time elapsed, not number of turns): Council Room, Smithy, Harvest, Woodcutter, Outpost, Feast, Coppersmith, Secret Chamber, Transmute, Peddler. Province/Estates.

All actions are terminals (besides Peddler which will never be bought over Gold), Council Room helps your opponent, there are no attacks, and there are few gainers or cards that require meaningful choices or take time to resolve. Contenders welcome, I refuse to test this or alternate submissions.

Baker > Transmute, the bot may buy them but getting to open 5/3 sounds better.

An IW/Hamlet/Workshop/Gardens deck might also be pretty fast, but your turns could take a while.
Title: Re: Update 28 August 2014
Post by: Holger on September 15, 2014, 02:41:08 pm

Normally you get 15 coins for a win, 10 for a tie.  Perhaps it was graded as a tie for some strange reason? When you tie, it still says "you're the winner".

No, you only get 10 coins for a win against a bot (I think you get 15 coins only for wins against human opponents).
A tie would explain the rating loss, though I'd expect it to be even more than 30 points then.

I am now confident it was due to increased uncertainty after not playing for several days that carried the loss. Evidently displayed rating only changes when you play a game while true rating changes daily . Thus if you don't play for a few days a dozen or more point loss (30 here) will be incorporated into your next game.

Yes, but according to Gokosalvager (http://gokosalvager.com/logsearch?p1name=Andrew+Iannaccone&p1score=any&p2name=&startdate=08%2F05%2F2012&enddate=09%2F15%2F2014&supply=&nonsupply=&rating=pro&pcount=2&colony=any&bot=any&shelters=any&guest=false&minturns=&maxturns=&quit=any&resign=any&limit=20&submitted=true&offset=20 (http://gokosalvager.com/logsearch?p1name=Andrew+Iannaccone&p1score=any&p2name=&startdate=08%2F05%2F2012&enddate=09%2F15%2F2014&supply=&nonsupply=&rating=pro&pcount=2&colony=any&bot=any&shelters=any&guest=false&minturns=&maxturns=&quit=any&resign=any&limit=20&submitted=true&offset=20)), he played another Pro game only hours before...
Title: Re: Update 28 August 2014
Post by: ragingduckd on September 15, 2014, 08:26:15 pm

Normally you get 15 coins for a win, 10 for a tie.  Perhaps it was graded as a tie for some strange reason? When you tie, it still says "you're the winner".

No, you only get 10 coins for a win against a bot (I think you get 15 coins only for wins against human opponents).
A tie would explain the rating loss, though I'd expect it to be even more than 30 points then.

I am now confident it was due to increased uncertainty after not playing for several days that carried the loss. Evidently displayed rating only changes when you play a game while true rating changes daily . Thus if you don't play for a few days a dozen or more point loss (30 here) will be incorporated into your next game.

This explanation requires the player to have initial uncertainty of ~1400.  For a regular player, uncertainty is < 300.  At +1% uncertainty per day, that's an awfully long time sitting inactive, about 5 months without a game.

Code: [Select]
Initial Uncertainty = 1400:
    AI Rating - Before:
    6500.00 +/- 1400.00
    Displayed as 3700
   
    AI Rating - After:
    6513.54 +/- 1390.26
    Displayed as 3733
   
    Displayed Change: -33

Python script to produce this output is attached, including rating info for me and Serf Bot.  The trueskill Python module is required.