Updated FargoRates are out

poolscholar said:
Seems like the system is too granular. What's the point of calling someone a 763 when they could be plus or minus 20 points? Maybe it needs a +/- for 1 or 2 standard deviations out next to the rating.

Then what's the use of the % chance to win race calculator, if it's highly dependent on the length of race and the possible variation in someone's rating?
Click to expand...

That probably goes without saying, you can't judge a human performance within 1% with accuracy. Or even 10%. In the US Open, Wu lost a match like 11-1 or something silly for him. Should be now be ranked a 600 because he lost to someone under him badly? Yet this is what people who say Fargo does not work do, every match is like a 90% weight of your total score to them, when it's more like 1%.

Fargo rates the population, just like any other statistic. It does not care or take into account if a guy plays bad on red cloth on Sunday, or that another guy can run 4 racks on a 7 footer but only 2 on a 9 footer. It just tracks the overall patterns of players taken together. Overall, 10,000 players will not have that much of a variance and it tracks them all in a web.
 
Last edited:
donny mills said:
This is exactly right!!! This is music to my ears! Finally I read something in this thread that makes sense! For all of you people on this forum that aren't sure what's accurate or not. This is accurate!
Click to expand...

Is there a specific cueball that is preferred by a majority of pro players?
 
donuteric said:
No, they don't. Far from it. From the website's statistics, 6 million games from 127k players means the average is far less than how the algorithm would define robustness as established. Assuming the 6 million games are total of everything, they probably have much less if they only include "recent" games.

Data collection is often far more challenging than algorithmic development and it requires partnerships from promoters, directors, and operators. My understanding is Mike et al are doing the best they can, but from a data scientist's point of view, 6 millions is a relatively low number in the big data era.
Click to expand...

so when to expect a flawless rating, one which doesnt show the speed from 3 years ago (after years of collecting already)? as i see it, fargo is way behind the reality.......
 
mikepage said:
It depends on what we mean by close. She may. If she's a little overrated and Donny's a little underrated, they of course would move toward one another.

But I think Siming plays better than Donny.

And I think the vast knowledge and experience of a handful of you who are smart and make good games and have a finely tuned intuitive sense of how people play relative to one another is a little out of its element when it comes to the top Asian women.

I see Donny as believing he is a little underrated. And I see him perusing the list for someone above him who he thinks is a little overrated and proposing a game with a modest spot. That's good and everything. I always like to see more games go down. But it is not earth shattering.

Here is I think a more interesting experiment. Take the USA top 50 list and generate a couple dozen random pairings: Oscar vs Sossei, Sky vs Josh Roberts, and on and on. And then give that list to a handful of people like you who others consider knowledgable. Ask two questions... Who will win a race to 30? How many games will the loser win? I think averaging over a number of knowledgable people, the results would be close to the expectation from FargoRate.
Click to expand...

The system is not bad .
But if she is 30 points higher then another player
You would think she would be hands down the fav .
Or any player who is ranked that many points higher then other
Player .
 
hang-the-9 said:
That probably goes without saying, you can't judge a human performance within 1% with accuracy. Or even 10%. In the US Open, Wu lost a match like 11-1 or something silly for him. Should be now be ranked a 600 because he lost to someone under him badly? Yet this is what people who say Fargo does not work do, every match is like a 90% weight of your total score to them, when it's more like 1%.

Fargo rates the population, just like any other statistic. It does not care or take into account if a guy plays bad on red cloth on Sunday, or that another guy can run 4 racks on a 7 footer but only 2 on a 9 footer. It just tracks the overall patterns of players taken together. Overall, 10,000 players will not have that much of a variance and it tracks them all in a web.
Click to expand...

I'm not suggesting it can predict a single outcome. It's all statistics like you said... the more data, the more likely a person is accurately rated. The problem is people think that a 750 is actually a 750, when its really a range of possible numbers. This not only varies with the amount of data collected (more data would usually create a smaller range) but by person. What if a one 750 get's drunk every other tournament and another 750 plays sober and consistent. They will have different distributions of data. Sober guy might be 740-760 where the other guy has a range of 720-780.

You can calculate these ranges by using various confidence intervals (ie 95% sure its in a particular range).
 
Oze147 said:
Well, that is simply not true. I don't think, that there are more than 10 different players, that have won one of the last 20 EuroTour events.
Seer.
Click to expand...

But the fact that there is no single dominant player in a short race format is what suggests the format is a crapshoot.

When you have a 100+ players in an event and claim the cream always rises to the top, your 'cream' is 1 of a group of superior players.

I would not want to bet against someone else being able to pick 1 of the top 3 players for the us open, allowing him a group of 20 players.

Same bet with group of 1? Action, jackshun.
 
mikepage said:
Two of those players are rated above Dechaine, not below him.

Regardless, they're all in the same ballpark so it doesn't much matter who is a few points above who.

Why do you think Dechaine would underperform? Is it because he hasn't been in the ring much?

Three weeks ago he played Yu Lung Chang (792) 44 games. Mike won 23 and lost 21 --about what we'd expect. And in that tournament generally he performed at 810 speed for 109 games.

I'm wondering what is the basis for your thinking about Mike.

And Dodong now has over 600 games performing at 802 speed.

It's amazing. As soon as one player spots another player a little something in a set that could go either way even with the spot in the other direction, that becomes some sort of fundamental knowledge that the spotter plays better than the spottee... You guys should be a little more careful with your conclusions and just a little more cynical about the legendary army of world-class 12-year-olds in the Philippines. I'm not saying there are not several very straight-shooting young players in the the Philippines; we actually have evidence there are. But ask yourself, if these stories that are repeated over and over again get embellished just a little bit here and there which direction are they going to morph? --like the number of pool rooms in Shanghai. I think somebody is confusing pool rooms and Starbucks.
Click to expand...


did he? cant have been much of a field while darrens tourney! so, you would bet on MD vs those 3 in a even match? or on efren in his pack? same question to anyone else.....

and to the philipinos......hmmm so by your logic it doesnt count when a unknown youngster spots and beats busti and efren (proof is out there lol) and gets spotted himself? by matchmakers who do probably nothing else than matchmaking?
oh well, you think siming deserves her rating......
 
poolscholar said:
I'm not suggesting it can predict a single outcome. It's all statistics like you said... the more data, the more likely a person is accurately rated. The problem is people think that a 750 is actually a 750, when its really a range of possible numbers. This not only varies with the amount of data collected (more data would usually create a smaller range) but by person. What if a one 750 get's drunk every other tournament and another 750 plays sober and consistent. They will have different distributions of data. Sober guy might be 740-760 where the other guy has a range of 720-780.

You can calculate these ranges by using various confidence intervals (ie 95% sure its in a particular range).
Click to expand...

The only real problem I see with FargoRate is that it is trying to use statistics (mathematical sense) to provide information to people like you who don't understand statistics.
 
real bartram said:
The system is not bad .
But if she is 30 points higher then another player
You would think she would be hands down the fav .
Or any player who is ranked that many points higher then other
Player .
Click to expand...


30 points isn’t a huge difference.
 
real bartram said:
If you are talking 100 to 130 no
But when your a 750 going to a 780
30 points there is a lot .
And she is ranked above a lot of greet players
How about the 1 she is right above thorsten
Click to expand...

You are misunderstanding one basic premise of FargoRate: Every 100pt difference means the higher player is twice as good as the lower player. I.E. a 200 vs. a 100 in a race to 10 would expect to win twice as many games. It is the same for a 800 vs. a 700.

Thus, a 30pt gap at 100-130 is the same as 750-780.

If a 700 played a 100, the 700 would be expected to win about 64 games for every one game the 100 won.
 
real bartram said:
If you are talking 100 to 130 no
But when your a 750 going to a 780
30 points there is a lot .
And she is ranked above a lot of greet players
How about the 1 she is right above thorsten
Click to expand...
In either case where the players are 30 points apart, the better player is expected to win 55.18% of the games. That is fundamental to how the system is set up.
 
Bob Jewett said:
In either case where the players are 30 points apart, the better player is expected to win 55.18% of the games. That is fundamental to how the system is set up.
Click to expand...

I'm not sure I see 1 player on that list ranked under her
That she is the favorite over .
 
real bartram said:
The system is not bad .
But if she is 30 points higher then another player
You would think she would be hands down the fav .
Or any player who is ranked that many points higher then other
Player .
Click to expand...

gxman said:
I agree.

US Open 9b race to 11.

Shane, Dennis, and Wu lost matches to players 40-50pts below then.

Thorpe beat Wu and Biado.
Click to expand...

And the undefeated Patriots lost to the Giants

It happens.
 
I dont think any of the chinese females gamble, but the Philippine girls do. Not quite as highly rated a Chen, but Chezka Centeno can play. Her Fargo is 758. She plays fast pace like Shaw.
 
poolscholar said:
enlighten me?
Click to expand...

In your example:

A 750 is a 750. His performances rate him to be a 750. He has a statistical average of 750. His performances can theoretically range all the way from 0 to infinity. The basic way this would work can be described by a basic bell curve and standard deviation can be figured, giving odds of the guy's performance based on a range.

But, the guy that you were talking about who varies widely based on whether he drinks or not, this too has to be figured into one set of statistical data. His curve is likely a multimodal distribution. The odds of him performing inside some range can also be figured, but less directly. Basically the guy has two different 'modes' affected by whether or not he is drunk.
 
PoppaSaun said:
In your example:

A 750 is a 750. His performances rate him to be a 750. He has a statistical average of 750. His performances can theoretically range all the way from 0 to infinity. The basic way this would work can be described by a basic bell curve and standard deviation can be figured, giving odds of the guy's performance based on a range.

But, the guy that you were talking about who varies widely based on whether he drinks or not, this too has to be figured into one set of statistical data. His curve is likely a multimodal distribution. The odds of him performing inside some range can also be figured, but less directly. Basically the guy has two different 'modes' affected by whether or not he is drunk.
Click to expand...

Both of these sets average to 750. Using a standard distribution with 95% confidence, the average falls in this range. Not sure how this is different from what i was stating, that rating "range" could vary a lot depending on the data set.

700, 700, 700, 800, 800, 800

706.173067641 - 793.826932359

and

740,740,740,760,760,760

741.2346135282 - 758.7653864718


(of course fargorate is using match scores, not averaging ratings like this)
 
You must log in or register to reply here.
Back
Top