Consistency of APA Ratings
- Thread starter pogmothoin
- Start date
How so? If you already know a certain bent coin comes up heads 53.2% of the time, how does a measure of the exact curvature or the shape of the engraving or the wear & tear on the coin help you in predicting the next flip?
Wins and losses in pool games do not happen in a vacuum. They are made up of component parts like shotmaking ability, safety play, breaking skill, mental toughness, etc. If there were good ways to quantify and compile that data, it could go into a model that would be more predictive than simple win/loss. This seems self evident.
By contrast coin flips have no component parts. It is random chance. If wear and tear on the coin had an impact on the results, then obviously you would want to include that in any model or rating that predicts the result.
I don't think there's anything particularly wrong with your rating system - in fact I think it's a great idea. A system based on a centralized bank of thousands of thousands of games is likely to be very predictive and accurate. That doesn't mean it couldn't be better if other data sources were incorporated.
Does anyone really think the APA ratings would be BETTER than they are now if they threw out all the inning data?
Unless all those 300 matches were all played last month how does this factor in?
Your stats are the same but Rob has won the last 5, 10, 25 matches against you? So your stats may say you are 2/3 chance of winning against him and I would say based on the most recent performance you have about a 3% chance, if that, of winning against him.
It seems self evident because, I think, you are seeing a person's win record a merely on incomplete piece of information that can be supplemented by other predictive information. And yes, in some cases the win/loss information you have for the players can be used that way.
The distinction I'm trying to make is that it is in principle fundamentally different from that. Game win/loss is, in fact, exactly what you are trying to predict. And if, for example, you had reliable (as in big numbers) recent statistics on win/loss between these two players and all you know is they are about to play a game, then there is nothing you can do to improve your prediction of the outcome of the next game than to use the principle of maximum likelihood.
It is actually a fitting analogy. Many factors go into whether this bent coin comes up heads on the next flip. These include the moment of inertia tensor, laminar flow and turbulence of air around the different crevices of the spinning coin, the curvature of the bend, the elasticity of the coin-to-surface collision, details about the tosser, and so forth. And again, in the absence of knowing in advance the statistical likelihood of heads, these factors might all enter a complex model to predict the outcome.
But if you actually flipped the coin 10000 times under the circumstances you are interested in and found it cane heads 5322 times, you are justified to throw out your complex model and answer 53.2% heads.
I don't. That is a different issue.
3% ?? I must be drunk AND Rob just killed my dog...
Seriously, though, this is a separate issue. FargoRate does weigh recent games more heavily. For this argument, I should have said 300 recent games...
I do not, as I believe average innings to win is one of the biggest displays of one's true skill. It is a fact the better a player gets the less innings they will shoot when they are winning and as the inning count climbs the defensive shots for and against them do as well.
Things like me winning in X innings but my opponent scratched on the 8, made an early 8 or put the 8 in the wrong pocket should matter as it says nothing about my performance. Thus a win like that should be completely ignored by the system. I do not believe it is though as I have seen 2/3 get moved up the next week after winning a match that their opponent fouled on the 8 every time. That is silliness as you can't really rate anyone's performance at that point without seeing the match. For all they know they could have ran all the way out and legitimately missed an easy 8 they left hanging that the opponent accidentally put in or they could have broke dry and then the opponent ran all the way out to the 8 and accidentally scratched while making it with an amazing 8 rail shot they accurately calculated but missed on speed.
Both of those are the same on paper.8 on the break is another thing. It happens so infrequently and is luck. Yes I call it luck when you try it every time but only do it maybe once every 10-25 breaks, if that, of the trillion+ possibilities for the table layout. It is so infrequent counting it probably has a negligible affect on the average person's rating but would affect the skilled breaker that pretty much spots themselves a game every match if they can do it every 4 or 5 games.
OK, that makes more sense. :thumbup:
It may seem self evident, but it is wrong. *Adding* any of those factors to a simple win/loss predictor requires that they be given a weighting, and for the portion of that factor already in the win/loss system be removed. Basically, you can have a COMPLETE simulation which takes everything into consideration, each with its appropriate weighting, or failing that a simple win/loss is going to be superior.
How would you know? Studies have been done, and the number of flips required to notice a bias is VERY large. By the way, of course coin flips have component parts. Perfect coin flipping machines have been made. That means that we do have a complete simulation of coins in the air. We can replace a win/loss simulation with an accurate physics simulation, we can't join the two.
No, the reason that other data sources can't be incorporated is that they already are.
Yes. I would guess that the inning data is less correlated with win chance than a simple win/loss calculation. If win/loss is correlated to 0.75, and inning count is 0.65, why would you add it? And someone needs to guess how much to weight it by, which if they are incorrect, will make predictions even worse.
Not to mention that it would be harder to sandbag.
Thank you kindly.
Last edited:
No, you're not. You would have to compare how well the complex model predicted outcomes over the simple frequentist version. (Unless your goal in computational simplicity).
Thank you kindly.
jmarcum1984
Registered
Innings can be very misleading in that, a 3 may work very hard to make 3 balls in a row, and against another 3, a four or five inning game is the way they play. However, if the table is slightly cluttered, and a 6 does not break anything out, and waits for the opponent to move balls until the table is open, he can take five innings for the game, but make 7 balls in a row at the end to win it. On paper, the 3 and the 6 look the same because it took 5 innings, but in watching the match, they are a world apart.
It can only be considered luck if it happens with equal frequency for everybody. If you do it 1 in 10, and I do it 1 in 100, you should have a significantly higher rating than me (9%), all else being equal.
Thank you kindly.
How is this done? Do old games age (for lack of a better term)? Do they expire? Or do only the most recent games count, regardless of how old they are? Do different player's games age at different rates based on how often they play?
Thank you kindly.
Not true if you are speaking of APA, which I believe is the only league that tracks innings. If defensive shoots are being recorded as they should be for every time that 6 shoots and does not intend to make a ball that inning is a wash with the def being marked. That would be true if the 6 went 2 innings or 25 innings before making the run out.
Equal frequency for everybody is a pretty bold statement on your part as a definition of luck. If everyone did it 1 in 10, is it really luck by your view? It wouldn't be in mine, it would simply be a normal course of action, thus expected behavior, that occurs when you strike the rack 10 times. What is so "lucky" about expected behavior?
If making the 8 on the break often enough that it gives you a statistical game advantage over every opponent you play, then it should affect your rating. Being that the most games any APA player would ever have to win is 7, I would say you should have to have a better than 1 in # of games to win chance of making the 8 on break statistically for it to be a factor. Or in other leagues you play to 15 and you have a better than 1 in 15 chance of making it and your opponent does not, then it should be a factor.
To add to this, the 6 would (in The APA) only get credit for the innings that he should have actually played. I think the range of a 6 is like from 1.8 - 2.9 or something like that, I'm sure someone will correct me, anyways, the 6 might play 6 - 8 innings per game waiting for the 3 to break out the balls, and maybe he wins 5 games and the match. Lets' say he played 6 to 8 innings per game and after the match his total innings were something like ohhhh, 32 total innings. As he is a pretty good 6 he'll actually only get credit for about 13 or 14 of those innings (about 2.1 or 2.2 per game), then you'll subtract any defensive that were marked, so lets' say the scorekeeper marked 5. Now that 32 inning total just went from 32 to about 8, this will look pretty different from what the S/L 3 recorded score will be. In the end the 6 will have recorded a score of about 1.8, he may even move up a bit
Last edited:
The weighting of games in the optimization has an exponential decay factor in time that has a half-life of 3 years.
So if you beat me 40 to 20 three years ago and then I beat you 20 to 10 today, our records are considered even. We are 30 to 30 measured in "current" games.
No it is not. Every league is free to use whatever rules and/or handicapping system they wish. And there are no points or handicaps at the National tourney in Vegas.
Things like me winning in X innings but my opponent scratched on the 8, made an early 8 or put the 8 in the wrong pocket should matter as it says nothing about my performance. Thus a win like that should be completely ignored by the system. I do not believe it is though as I have seen 2/3 get moved up the next week after winning a match that their opponent fouled on the 8 every time. That is silliness as you can't really rate anyone's performance at that point without seeing the match. For all they know they could have ran all the way out and legitimately missed an easy 8 they left hanging that the opponent accidentally put in or they could have broke dry and then the opponent ran all the way out to the 8 and accidentally scratched while making it with an amazing 8 rail shot they accurately calculated but missed on speed. Both of those are the same on paper.
8 on the break is another thing. It happens so infrequently and is luck. Yes I call it luck when you try it every time but only do it maybe once every 10-25 breaks, if that, of the trillion+ possibilities for the table layout. It is so infrequent counting it probably has a negligible affect on the average person's rating but would affect the skilled breaker that pretty much spots themselves a game every match if they can do it every 4 or 5 games.[/QUOTE]
The computer program must have something like a red flag stat. I've seen a 3 player with an average record win a match in 3 innings. An 8 ball on the break and an easy run out with balls spread near pockets. The next week he went to a 4. I told the league operator to look into it and 2 weeks later he was back to a 3 and has been ever since. I think when the computer sees something unusual it automatically kicks the player up.
Both of those are the same on paper.8 on the break is another thing. It happens so infrequently and is luck. Yes I call it luck when you try it every time but only do it maybe once every 10-25 breaks, if that, of the trillion+ possibilities for the table layout. It is so infrequent counting it probably has a negligible affect on the average person's rating but would affect the skilled breaker that pretty much spots themselves a game every match if they can do it every 4 or 5 games.[/QUOTE]
The computer program must have something like a red flag stat. I've seen a 3 player with an average record win a match in 3 innings. An 8 ball on the break and an easy run out with balls spread near pockets. The next week he went to a 4. I told the league operator to look into it and 2 weeks later he was back to a 3 and has been ever since. I think when the computer sees something unusual it automatically kicks the player up.
LOs cannot lower a player, they can only lock their current rating or move them up. They can NEVER move them down so if the player was moved down there was a statistical reason for it based on their performance and how they measure it.
This kind of thing can also happen to new players or players that haven't played in a while as their rating is very volatile during the first 10 matches. When I started playing 3 years ago I went down to a 3 and then down to a 2 within a few weeks (played twice a couple nights), but then moved up to a 3 a few weeks later and then 4 a couple weeks later, eventually a 5 after about a year and then a 6 about 6 months after that (If I recall correctly). About a month after that I went down to a 5 for one week and then back to a 6 and have been there ever since. Becoming a 7 is pretty tough, but I am determined to do it.