Subject: Should winnable variants affect Elo?
Date: Sun Oct 30 20:54:16 2022
User: WRAC
Message:There's a lot of interesting speculation in this thread.
Even though there doesn't appear to much enthusiasm for not rating the winnables, it would still be interesting to know if players do indeed win more often when they know a game is winnable.
Hop and cellmate, I think you have the game-by-game data going back a year or so - is that right? Can either of you tell us if the winnable games are won more often when they are played as Winnable than when they are played as Streak?
Some think it makes no difference. Others disagree. My own guess is the games are won *slightly* more often when played as Winnable, but that's just a guess, and I'd love to know the truth.
Of course, there may be other factors (such as: perhaps the folks who play Winnable are more inclined to play slowly and carefully than those who play Streak), but this would give us some idea.
If you could enlighten us, that would be awesome.
Date: Sun Oct 30 22:24:36 2022
User: HopDiriDiriDattiriDittiriDom
Message:"Hop and cellmate, I think you have the game-by-game data going back a year or so - is that right?"
Right, WRAC. The data is loaded on the server actually. Anyone can access it. But I download it to my computer to create my Hop-style Elo ratings.
cellmate may even have more than that because his tournament viewer goes back to the earlier dates.
But I doubt we can see any meaningful result with one year's data. I would expect most of the games must have been played once or couple of times at most. And for most of them there wouldn't be any regular play to make a comparison.
Date: Sun Oct 30 22:35:59 2022
User: HopDiriDiriDattiriDittiriDom
Message:Hey WRAC,
Actually my Hop style Elo page has "Variants" tab and there you can find how many times a variant is played how much Elo is gained or lost.
It should give you some idea.
Link: Hop style Elo
Date: Sun Oct 30 23:09:25 2022
User: cellmate
Message:
interesting coding problem (turned out not as tricky as i thought it would be)
i ran it for 10x1 all levels for all of 2022
In both: 15097 (played in both winnable and regular)
Compared: 4214 ( with same number of plays)
maxplays: 5 ( max plays of those compared )
Regular More wins: 361 (regular play had more wins)
Winnable More wins: 1102 (winnable play had more wins)
Same: 2751 (same number of wins in both )
Date: Mon Oct 31 00:05:25 2022
User: HopDiriDiriDattiriDittiriDom
Message:13x0 has more plays. It would have more comparisons and higher maxplay.
Date: Mon Oct 31 00:29:00 2022
User: joeygray
Message:Subjectively - but I bet I get a lot of agreement here - 13x0 as with all the zeros has a way bigger chance that someone deeming it unwinnable is going to be correct. In other words, mostly immune to the effect being discussed on this thread.
Date: Mon Oct 31 00:50:16 2022
User: cellmate
Message:
Layout: 13x0
In both: 80913
Compared: 20862
maxplays: 12
Regular More wins: 1288
Winnable More wins: 1112
Same: 18462
Date: Mon Oct 31 01:14:24 2022
User: Kumquat-of-Conciliation
Message:I think joey's statement just above is likely exactly on target.
Date: Mon Oct 31 01:50:02 2022
User: HopDiriDiriDattiriDittiriDom
Message:5x8 is also interesting:
# | Variant | Games won | Games Lost | Elo Gained | Elo Lost | Diff | Diff/Plays
|
= | == | == | == | == | == | == | ==
|
1. | 5x8 | 3.218 | 2.100 | 9.150,315 | 7.920,953 | 1.229,362 | 0,231
|
2. | 5x8w | 6.709 | 1.644 | 10.118,217 | 9.583,917 | 534,301 | 0,064
|
Regular 5x8 has less plays but bigger gains.
Date: Mon Oct 31 02:01:33 2022
User: Gulbis
Message:cellmate: "i cringe whenever someone claims that knowing the game is winnable makes it easier"
First of all, I'm not saying the games are easier, I'm saying they are more likely to be won. Which is true.
Second of all, the winnable variants ARE easier than the standard variants because they don't contain the most difficult i.e. the unwinnable games. These games don't impact Elo, that's true, but you still have to play them and you don't know for sure that they are unwinnable so you may spend a lot of effort on these games. So yes, the standard variants are harder.
"No reason to discount Elo when you try harder."
Yes, there is a reason to do that. Elo predicts how likely a player with a particular rating is to win a game of a particular rating. For example, if your rating is 1800 and the game's rating is 2000 (a very difficult game), the likelihood that you will win is 24%. But let's say you encounter this game in a winnable variant and that increases the likelihood of your win to 35% percent because you will know it is winnable and you will "try harder". This is the problem. Your likelihood is either 24% or 35% based on which variant you play, and it shouldn't work like that. There should be only one likelihood, not two of them.
Tony: "Eliminating the winnable variants would be throwing out a lot of data! That is definitely a "down side", imo!"
Do you mean the fact that the games would not get rated when they're played as winnables? I see it mostly as a good thing because winnable variants tend to lower the game rating too much (because they are more likely than normal to be won). "Throwing out" bad data is not a downside.
Joey: "Gulbis, your passioned argument about how good it is to make improvements is beside the point. It is not at all clear this would be any improvement. Obviously you think it is, but a number of folks don’t."
What do you mean "beside the point"? That is the whole point of this thread!
And yes, it would definitely be an improvement. It's not my opinion, it's math. It's that whole situation I described for cellmate above. Most people will "try harder" if they know the game is winnable and that distorts Elo. ElGuapo agrees. The question is how much and whether we want to do something about it. So we are discussing it.
Date: Mon Oct 31 02:11:02 2022
User: HopDiriDiriDattiriDittiriDom
Message:cellmate, question:
Have you taken the levels into consideration or are they ignored?
Date: Mon Oct 31 02:39:49 2022
User: cellmate
Message:
Hop, yes. all levels selected. the stats are collected when cols,cells,level,gamenumber and total plays are the same in both variants
Date: Mon Oct 31 03:43:36 2022
User: cellmate
Message:
Gulbis,
Hmmm... pondering... Nope. Not convinced.
For Elo its player vs game. Player wins, player takes points from game. Player loses, game takes points from player. It is as simple as that.
But... that's just me
Date: Mon Oct 31 04:08:03 2022
User: HopDiriDiriDattiriDittiriDom
Message:"Player loses, game takes points from player. It is as simple as that. "
I like simple things. I always look for possible simplifications when attacking problems.
Date: Mon Oct 31 04:46:12 2022
User: TitanicTony
Message:No, cellmate, it is not "just you"! I agree with you. Apparently, so does Hop.
Btw, It is not true that "most players" will try harder. "Most players" are playing 8x4!
I admit that I try harder when I know (or suspect) that a game is winnable. Maybe ~19 other players also do that?
"Most players" are just playing for fun, relaxation, distraction, whatever, and don't care at all if they are in the Elo top 20, imo.
Please, Gulbis, GIVE UP (gracefully, if possible). You are not winning this argument, imo!
Date: Mon Oct 31 08:04:32 2022
User: Gulbis
Message:Tony, please don't order me around. What do you care if I give up or not? No one is making you read this thread.
For cellmate and others - I could understand if you said "Yes, I know this distorts Elo but I don't care, I want to keep receiving points for winnable games." That's a valid position.
Or if you said "Yes, I know this distorts Elo but I think it's very minimal, like 2% so this minimal impact doesn't warrant any changes." (This is something similar to what Denny said.) That's also fine and can be discussed.
But if you're arguing that it doesn't have any bad impact on Elo then it's not a valid position, you just don't understand how it works or how it is supposed to work. I have tried to explain it many times but if you don't get it then you don't get it. I must admit defeat. Sorry.
Date: Mon Oct 31 08:48:40 2022
User: Kumquat-of-Conciliation
Message:^^^^^^^^ That ^^^^^ is the gist of the matter. Whether one calls them "easier", or "more likely to be won", both affect the elo ratings. It's only a question of degree.
Irony on display............
Tony: "Eliminating the winnable variants would be throwing out a lot of data! That is definitely a "down side", imo!"
That from the one player on this site most responsible (very likely) for damaging/destroying the database.
Date: Mon Oct 31 10:18:59 2022
User: redberet
Message:I never put too much credence in single game stats, The one thing it doesn't tell you is the quality of the players. There are too many variables involved. One's perspective is the only way to judge difficulty.
Gulbis, you're stating unless one agrees with your initial premise, then a counter point is not valid. You let Flocci get to you and it looks like it still does. I'll bet most winnable players are not aware of this thread and those are the voices that should be heard.
Date: Mon Oct 31 11:01:07 2022
User: Gulbis
Message:"I never put too much credence in single game stats, The one thing it doesn't tell you is the quality of the players. There are too many variables involved. One's perspective is the only way to judge difficulty."
So you are saying that Elo is basically worthless anyway? Or what is your point?
"You let Flocci get to you and it looks like it still does."
What does it matter who I let get to me or what my motivations are? We're discussing the site, not me, and I'd like to keep it that way. And for the record - you don't know what you're talking about. Flocci got his high ranking playing the 10x0 variant, not winnables, and that problem has already been fixed (thanks, ElGuapo).
Date: Mon Oct 31 11:22:33 2022
User: redberet
Message:I don't have a point, I'm trying to understand yours. If someone can run the table on them winnables, then maybe you would have a valid point. You also can't know the motivations of those who play that variant. If it does turn out there are those who somehow manage an advantage, it's kind of late in the game to pull the rug out now. Elo is not worthless, it a gauge and nothing else.
Date: Mon Oct 31 11:59:36 2022
User: WRAC
Message:Thanks for the info, Hop and cellmate - much appreciated!
According to cellmate's comparisons, when playing Winnable, players had more success with 10x1, but less success with 13x0.
If I'm interpreting Hop's stats correctly, when playing Winnable, players had more success with 10x1, 11x1, 12x1, 13x0, 4x10 and 7x4 (6 variants), but less with 5x8, 6x5, 8x3 and 9x2 (4 variants). These numbers should be a little less accurate than cellmate's game-by-game comparisons, but the overall effect doesn't appear to be great and not something to worry about.
Date: Mon Oct 31 12:28:30 2022
User: cellmate
Message:
Here's my results for all the winnable variants...
Layout: 4x10
In both: 1020
Compared: 772
maxplays: 2
Regular More wins: 125
Winnable More wins: 104
Same: 543
Layout: 5x8
In both: 147
Compared: 132
maxplays: 1
Regular More wins: 28
Winnable More wins: 10
Same: 94
Layout: 6x5
In both: 2143
Compared: 1558
maxplays: 3
Regular More wins: 269
Winnable More wins: 232
Same: 1057
Layout: 7x4
In both: 17956
Compared: 1848
maxplays: 7
Regular More wins: 262
Winnable More wins: 376
Same: 1210
Layout: 8x3
In both: 16885
Compared: 6312
maxplays: 5
Regular More wins: 902
Winnable More wins: 887
Same: 4523
Layout: 9x2
In both: 12072
Compared: 4983
maxplays: 5
Regular More wins: 815
Winnable More wins: 354
Same: 3814
Layout: 10x1
In both: 15110
Compared: 4215
maxplays: 5
Regular More wins: 361
Winnable More wins: 1100
Same: 2754
Layout: 11x1
In both: 6984
Compared: 4876
maxplays: 3
Regular More wins: 224
Winnable More wins: 479
Same: 4173
Layout: 12x1
In both: 85613
Compared: 26355
maxplays: 8
Regular More wins: 706
Winnable More wins: 937
Same: 24712
Layout: 13x0
In both: 81003
Compared: 20857
maxplays: 12
Regular More wins: 1292
Winnable More wins: 1116
Same: 18449
Date: Mon Oct 31 12:33:41 2022
User: Kumquat-of-Conciliation
Message:cellmate, could you please lay out your methodology? Thanks. I suspect I understand most of it, but want to make sure.
Date: Mon Oct 31 12:53:59 2022
User: cellmate
Message:
Gulbis,
i don't see that your understanding of Elo is any better, or any worse, than mine
Just for you... "I don't agree that playing the winnable variant distorts Elo. I want to keep receiving points for winnable games for as long as the winnable variants exist"
some other thoughts...
your premise borders on superstition
i rarely play at my best. i might try harder starting any variant but that usually fades away into a middle of the pack effort
Elo is blind. It does a simple point exchange. it has no concept of who what when where or why
Treating player and game as equivalent doesn't result in a fair player vs player Elo
Doesn't matter how you measure it, the better players, the ones who win the most games, rise to the top
Before Elo, we had win percentage by variant... but no overall win percentage. To me, just gut feel, Elo here is analogous to what would have been the overall percentage
Date: Mon Oct 31 13:06:23 2022
User: cellmate
Message:
TN,
i gather all the winnable, and all the regular plays for the layout
where layout-level-game appears in both sets and number of plays is the same
i gather the stats
i don't try to interpret what it means when there is a different number of plays
Date: Mon Oct 31 13:19:28 2022
User: TNmountainman
Message:Let me back out a bit..............
So, going back to your post of Sun Oct 30 23:09:25 2022 above, concerning 10x1s.
There were only 15,097 games of 10x1 played for ALL of this year?
And exactly what does "the same number of plays" mean? Of those 15K+ games, for 4214 of them, you could match X number of plays? Meaning game 1234R (regular)1234W (winnable) may have been played 5 times, for example? Or does it (more likely) mean that there were 4214 instances of ANY R game being played 4 times, compared to ANY W game being played 4 times?
Thanks. Just trying to make sure I understand how you compared them.
Date: Mon Oct 31 13:38:27 2022
User: cellmate
Message:
No the 15097 is how many same games were were played. ie same game in both variants
same number of plays... eg if 1234-7 had 5 plays in winnable AND 5 plays in regular
where there's a match..,
same means count of games that had the same number of wins in both variants
similarly for regular/winnable more
Date: Mon Oct 31 13:47:40 2022
User: TNmountainman
Message:Ok, thanks. That helps.
Post follow-up