Print

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

[eHalcyon]

@ash maybe the best thing to do is to follow what Kirian said earlier -- that as far as ranking goes, it shouldn't really care whether you win, lose, or draw.  The purpose of a ranking system is to rank you, and it uses win/loss/draw info to do that.  A draw indicates something different than a win or a loss, and a ranking system should take that into account.

[ashersky]

@ash maybe the best thing to do is to follow what Kirian said earlier -- that as far as ranking goes, it shouldn't really care whether you win, lose, or draw.  The purpose of a ranking system is to rank you, and it uses win/loss/draw info to do that.  A draw indicates something different than a win or a loss, and a ranking system should take that into account.

That make sense to me.  A system can probably determine exactly how often it expects you to win, lose, or tie any given opponent, and adjust accordingly.  Then, the penalty/reward for a tie could be less/less than the penalty/reward for a loss/win.

[markusin]

Doesn't "shared" victory imply that a whole victory is divided and distributed evenly amongst the players that tied? I see a tie as being a half-victory for each player. I don't think it's outrageous for a tie to result in a loss in rank for the favoured player and a rank increase for the less favoured one.

Imagine a very simplified, hypothetical ranking system where one player essentially bets a certain amount of points that they can beat another player and the winner gets everything that was wagered. However, the less favoured player ends up wagering less than the favoured player by some sort of ratio determined by their rank difference. So like, player A puts in 10 points, and Player B puts in 20. If player B wins, they get a net point increase of 10 (-20+30) and Player A loses 10 points. If A wins, A gets 20 (-10+30) points and B loses 20 points. If they tie, they get 15 points back each, resulting in A having a net point increase of 5 (-10+15) points and B having a net point decrease of 5.

Of course, you don't have a net 0 increase in the point pool for a Goko game, and I'm sure the system I described above is totally inaccurate to what actually goes on in a rating system. Someone loses more than someone else gains or vice-versa. You could still imagine the scenario above where the pot starts off with free, magical points before each player puts in their contribution, but then there is the whole uncertainty factor and...yeah I don't study ranking systems.

[eHalcyon]

I will add that the main point about ties inflating rating is not actually about potential abuse.  No matter what the system does, there is probably a way that players can abuse it simply by rigging games.  The problem I was pointing out was that players could see their rank rise if they legitimately tied over and over again.  So, they are not trying to manipulate the system but they just happen to keep tying.

In terms of player skill, what do these results suggest?  If two players repeatedly tie, it is probable that they are extremely close in skill.  This is reasonable, right?

So if that is the case, how should rankings be adjusted?  Well, if the two players are very close in skill, then their ranking should be very close as well.  And if at first their rankings are different, then the logical conclusion is that a tie should bring their rankings closer together.  And if the rankings are supposed to get closer together, then the clear way to do that is to increase the rank of the lower ranked player decrease the rank of the higher ranked player.

Now, there are other possible ways to do this.  You could increase the rank of the lower ranked player and not change the rank of the higher ranked player.  But this does not really make sense in the context of a system with more players than just the two in that tied game.  In doing this, an assumption is made that the higher ranked player is correctly ranked and the lower ranked player is not, thus requiring the latter's rank to increase.  But that is a strange assumption to make; the natural assumption is uniform, that there is some uncertainty in EVERY rating, and therefore we see the two ranks increasing/decreasing to get closer together.

Does that make sense?

[Mic Qsenoch]

I will add that the main point about ties inflating rating is not actually about potential abuse.  No matter what the system does, there is probably a way that players can abuse it simply by rigging games.  The problem I was pointing out was that players could see their rank rise if they legitimately tied over and over again.  So, they are not trying to manipulate the system but they just happen to keep tying.

In terms of player skill, what do these results suggest?  If two players repeatedly tie, it is probable that they are extremely close in skill.  This is reasonable, right?

So if that is the case, how should rankings be adjusted?  Well, if the two players are very close in skill, then their ranking should be very close as well.  And if at first their rankings are different, then the logical conclusion is that a tie should bring their rankings closer together.  And if the rankings are supposed to get closer together, then the clear way to do that is to increase the rank of the lower ranked player decrease the rank of the higher ranked player.

Now, there are other possible ways to do this.  You could increase the rank of the lower ranked player and not change the rank of the higher ranked player.  But this does not really make sense in the context of a system with more players than just the two in that tied game.  In doing this, an assumption is made that the higher ranked player is correctly ranked and the lower ranked player is not, thus requiring the latter's rank to increase.  But that is a strange assumption to make; the natural assumption is uniform, that there is some uncertainty in EVERY rating, and therefore we see the two ranks increasing/decreasing to get closer together.

Does that make sense?

Explain it to me in terms of apples. This is a bit too straightforward.

[ashersky]

I will add that the main point about ties inflating rating is not actually about potential abuse.  No matter what the system does, there is probably a way that players can abuse it simply by rigging games.  The problem I was pointing out was that players could see their rank rise if they legitimately tied over and over again.  So, they are not trying to manipulate the system but they just happen to keep tying.

In terms of player skill, what do these results suggest?  If two players repeatedly tie, it is probable that they are extremely close in skill.  This is reasonable, right?

So if that is the case, how should rankings be adjusted?  Well, if the two players are very close in skill, then their ranking should be very close as well.  And if at first their rankings are different, then the logical conclusion is that a tie should bring their rankings closer together.  And if the rankings are supposed to get closer together, then the clear way to do that is to increase the rank of the lower ranked player decrease the rank of the higher ranked player.

Now, there are other possible ways to do this.  You could increase the rank of the lower ranked player and not change the rank of the higher ranked player.  But this does not really make sense in the context of a system with more players than just the two in that tied game.  In doing this, an assumption is made that the higher ranked player is correctly ranked and the lower ranked player is not, thus requiring the latter's rank to increase.  But that is a strange assumption to make; the natural assumption is uniform, that there is some uncertainty in EVERY rating, and therefore we see the two ranks increasing/decreasing to get closer together.

Does that make sense?

Explain it to me in terms of apples. This is a bit too straightforward.

Can you explain it in terms of Apples to Apples?

[eHalcyon]

I will add that the main point about ties inflating rating is not actually about potential abuse.  No matter what the system does, there is probably a way that players can abuse it simply by rigging games.  The problem I was pointing out was that players could see their rank rise if they legitimately tied over and over again.  So, they are not trying to manipulate the system but they just happen to keep tying.

In terms of player skill, what do these results suggest?  If two players repeatedly tie, it is probable that they are extremely close in skill.  This is reasonable, right?

So if that is the case, how should rankings be adjusted?  Well, if the two players are very close in skill, then their ranking should be very close as well.  And if at first their rankings are different, then the logical conclusion is that a tie should bring their rankings closer together.  And if the rankings are supposed to get closer together, then the clear way to do that is to increase the rank of the lower ranked player decrease the rank of the higher ranked player.

Now, there are other possible ways to do this.  You could increase the rank of the lower ranked player and not change the rank of the higher ranked player.  But this does not really make sense in the context of a system with more players than just the two in that tied game.  In doing this, an assumption is made that the higher ranked player is correctly ranked and the lower ranked player is not, thus requiring the latter's rank to increase.  But that is a strange assumption to make; the natural assumption is uniform, that there is some uncertainty in EVERY rating, and therefore we see the two ranks increasing/decreasing to get closer together.

Does that make sense?

Explain it to me in terms of apples. This is a bit too straightforward.

If two apples playing a game repeatedly tie each other, the ranking system should adjust their ranks to be closer to each other.

This is actually a pretty common result in any games allowing apples to participate, as most apples tend to be very poor players with no grasp of game mechanics and infinitely long analysis paralysis.  SERIOUSLY GRANNY SMITH, MAKE YOUR MOVE ALREADY.

Err, sorry.  I do not mean to imply that all apples are bad players.  I am sure that some are very skilled.  I am not an applist.

[Stealth Tomato]

I will add that the main point about ties inflating rating is not actually about potential abuse.  No matter what the system does, there is probably a way that players can abuse it simply by rigging games.  The problem I was pointing out was that players could see their rank rise if they legitimately tied over and over again.  So, they are not trying to manipulate the system but they just happen to keep tying.

In terms of player skill, what do these results suggest?  If two players repeatedly tie, it is probable that they are extremely close in skill.  This is reasonable, right?

So if that is the case, how should rankings be adjusted?  Well, if the two players are very close in skill, then their ranking should be very close as well.  And if at first their rankings are different, then the logical conclusion is that a tie should bring their rankings closer together.  And if the rankings are supposed to get closer together, then the clear way to do that is to increase the rank of the lower ranked player decrease the rank of the higher ranked player.

Now, there are other possible ways to do this.  You could increase the rank of the lower ranked player and not change the rank of the higher ranked player.  But this does not really make sense in the context of a system with more players than just the two in that tied game.  In doing this, an assumption is made that the higher ranked player is correctly ranked and the lower ranked player is not, thus requiring the latter's rank to increase.  But that is a strange assumption to make; the natural assumption is uniform, that there is some uncertainty in EVERY rating, and therefore we see the two ranks increasing/decreasing to get closer together.

Does that make sense?

Explain it to me in terms of apples. This is a bit too straightforward.

The problem is when you have a blue dog that doesn't like apples...

[ragingduckd]

Err, sorry.  I do not mean to imply that all apples are bad players.  I am sure that some are very skilled.  I am not an applist.

My Apple plays a very good game of chess.

[blueblimp]

It wouldn't be nonsensical to make a "shared victory" better than a "tie". If you think of a win as 1 point, a tie as 0 points, and a loss as -1 points, then a "shared victory" could be worth something in between 0 and 1 (or even equal to 1).

But if you do that, then the game becomes not zero sum, so it's not purely competitive anymore, but some mix of competitive and cooperative. I just don't see how that would be a good choice for Dominion.

[SCSN]

The problem is the implicit assumption that rating points are some sort of "reward" for winning the game, which is just utter nonsense. They reflect the system's estimation of your skill, and if the system had expected you to win on your own but instead you "share your victory", that should cost you rating points.

[Kirian]

The problem is the implicit assumption that rating points are some sort of "reward" for winning the game, which is just utter nonsense. They reflect the system's estimation of your skill, and if the system had expected you to win on your own but instead you "share your victory", that should cost you rating points.

Why is there no "+10" button?  Great job boiling this down to its essence.

[Davio]

Yeah, this way we can also see why a win could easily result in a rating drop if a sliding window of N games is used. Your win vs. a high level 50 games ago could be replaced with your last win vs. a low level.

If you always get at least X points for winning, then the rating system could be easily cheated.

I dislike using the "same" rating algorithm for pro and casual games. They could just scale casual players with 1 to 5 stars or something just to keep it simple and remove the ladder altogether for them. If you are a casual player you just wan't to make sure you're not playing against an uber-pro.

For pro players the ladder is more important, but there's no point listing a value of 5343 without actually knowing what it means.

[blueblimp]

The problem is the implicit assumption that rating points are some sort of "reward" for winning the game, which is just utter nonsense. They reflect the system's estimation of your skill, and if the system had expected you to win on your own but instead you "share your victory", that should cost you rating points.

I agree that ratings shouldn't be considered rewards, but cooperative games can have ratings too. I don't know how to make ratings for a game that's a mix of competitive and cooperative, but there must be some way to make it work.

[big layoutski]

Err, sorry.  I do not mean to imply that all apples are bad players.  I am sure that some are very skilled.  I am not an applist.

Surely you mean, "They can't all be bad apples."

[eHalcyon]

Err, sorry.  I do not mean to imply that all apples are bad players.  I am sure that some are very skilled.  I am not an applist.

Surely you mean, "They can't all be bad apples."

Yes, you are correct.

I usually prefer Oranges anyway. But again, I am not an applist. One of my favourite flavours is Green Apple!

[GendoIkari]

Ahh, that's right. It has been a while. So in any case, if you can get a tie, the higher ranked player, in this case should lose points. If your scores are tie but the game isn't it should be a full win for the victor.

this actually happens in the rating system right now. I had this happen yesterday, where i lost 25 points for a tie with a player ~1500 points lower than me.

And i certainly felt it's not right. It was a game with no +buy and we traded provinces back and forth. He broke PPR, and i couldn't do anything but tie, since i was 2nd player. Why should i get penalized for this?

I agree with you, although for different reasons.

As mentioned, there are no ties in Dominion.  Either one player wins and the other loses, or they both win.  No other options.

In the case where you forced the joint victory as 2p, you both should have been given wins (as you rejoiced in your shared victory).  Your opponent would have gotten a large boost to his rating; you would have gotten a much smaller one.  All would be right with the world.

This seems silly to me... I think it's taking the phrase "rejoice in their shared victory" far too literally. That's simply a friendly and flavorful way of stating that it's a tie. I seriously don't think that the authors of the rule book intended to distinguish that situation from a tie game! In a tournament with a cash prize, if the final round ends in a tie they aren't just going to give both players the full prize amount... they will either split the prize amount, or have another round to determine the winner.

If you tie against a player of lower rank, you should lose points and he should gain points. The point of a ranking system is to determine how good you are. If you tie against a lower ranked player, you are less likely now to better than that player, so your rank should be moved closer to that player. A ranking system shouldn't take into account things like whether the rule book uses the phrase "shared victory" or "tie"... that would defeat the purpose of a ranking system.

[GendoIkari]

The problem I have with gaining or losing points from a tie is that suddenly the higher and lower ranked players are playing games with different victory conditions. If you're lower-ranked, you're aiming for a win or a tie. If you have a higher rank, you're avoiding ties.

I don't have a problem with this... For one, it means that your ranking more accurately reflects your skill... if you tie with a lower ranked player, you are less likely to be better than him, so you should have a ranking closer to his. Second, it reflects regular unranked games, I think.... if I'm playing with someone who isn't very good at the game, and I tie, I will feel like I lost.... I should have beat him, I either screwed up or got unlucky. If I'm playing against WW or Obi, and I tie, I'll feel pretty darn good about myself.

[yudantaiteki]

Also assuming wins gain you more points than ties, the weaker player would still try for a win.

[Watno]

What if we replace the rating system with apples? After every game, 1 apple is distributed between all the winners. I don't think anyone would say that's unfair.

[Stealth Tomato]

What if we replace the rating system with apples? After every game, 1 apple is distributed between all the winners. I don't think anyone would say that's unfair.

The current rating system doesn't have a unit specified. It's possible we're all currently rated in apples.

[eHalcyon]

Someone get nutki to update his extension to make the Apples unit explicit.

[LastFootnote]

The current rating system doesn't have a unit specified. It's possible we're all currently rated in apples.

That's a lotta apples. In fact, I don't have much incentive to increase my rating anymore; I've already got enough apples to last me for several years. Apples have pretty significant diminishing marginal returns for me.

[GendoIkari]

What if we replace the rating system with apples? After every game, 1 apple is distributed between all the winners. I don't think anyone would say that's unfair.

It's already been said, but this very question seems to imply that you see rating points as a prize for winning games. That's simply not at all what the point or reason for rating points is. Rating points exist to give an estimation as to how good a player is compared to another player, NOT to reward players for victories.

Coins (on Goko) are there to reward players for victories. You get some coins when you lose; more when you win. They reward you for playing, and reward you more for playing better. But we wouldn't look at how many coins a person has to determine who is the best player... all that would tell is is who plays the most (and spends the least).

[Kirian]

The current rating system doesn't have a unit specified. It's possible we're all currently rated in apples.

That's a lotta apples. In fact, I don't have much incentive to increase my rating anymore; I've already got enough apples to last me for several years. Apples have pretty significant diminishing marginal returns for me.

Will gladly take those apples off your hands.