Questions for Mike with FargoRate

[BasementDweller]

Or you can ask "what happens if a player ONLY plays 3 other players that are blind", and he ends up being like an 800 Fargo because he beats them 10-0 every match. It's only valid for a short time in a bubble of rating till any new data is introduced.

Mike showed that the isolated cases where players don't interact with different table sizes is pretty much at 0. Even if 20 players in Dallas only played on 7 footes, the 21st guy played on 9 footers in Houston and has a set Fargo. He starts playing with the 20 guys in Dallas and their are linked to his rating. Finding a group of players to argue validly about the 7 ft 9-foot skill level that ends up being correct statistically is like finding a tribe on an island that never had contact with modern man. Fargo does not give a crap about a single player or a group of 10 players or 20 players, it rates thousands of players over hundreds of thousands of games and presents a statistical prediction based on the average of all of that. It's like an insurance company that sets rates, they know that a person with a single accident has a greater chance of having more than someone that has not had one at all, so they automatically adjust the cost based on that. If you go argue with them because you can show them that your mom and uncle had only one accident that never had another, they will show you the 134,567 others that had one accident then had another one the same year.

Long term statistics and data trump small personal experiences and "feeling" and theoretical situations. For every Brumback that can be pointed out as an exception being amazing at one game but just OK in others there are 20 Gorsts or Fillers that are equally good at most games. Same thing for Joe Billy Bob that is a 600 at a 7 footer but can't make 3 balls on a 9 footer, that is super rare and just gets lost in the noise of statics.

I'm not confident that we would ever hear about players that have higher ratings on one table or the other, or one game or the other. We are only privy to the data that Mike gives us, so we are forced to argue in the blind.

While I'm a big fan of FargoRate I recognize it's limitations, even if they are found only in the margins.

 

[nine_ball6970]

If we ignore sandbagging.... There are two sources of error in a player's FR.

Random fluctuations can be figured out from the robustness, but how large is the expected error versus number of games recorded? Is there a secondary effect due to the lack of robustness of a player's opponents?

A second is due to the improvement (or decay) of a player's game over time. A new player who has shown a lot of improvement over the last several years is probably underrated. How large is that "lag" error for a typical new star?

Also, I was wondering if there were any observed "islands" of players -- a group that is weakly connected to the larger world of pool players. Recently a Peruvian player (Gerson Martinez) appeared from nowhere, or at least I hadn't noticed him before. Is Peru an island?

It seems like central and south america are pretty much unknown unless they come to the US. Gerson won the texas open 10 ball in feb 2022. His road partner Cristopher Tevez finished 2nd in world ten ball 2022. There was a player who came through town not long ago who is 750 with 114 games. Who knows how strong he actually plays.

 

[easy-e]

IDK.....I went 2 years without playing a Fargo match.....My Fargo went up in those two years.

The problem (as I see it) is that the TDs are using Fargo to limit the field.....If a tournament is capped at 599 and under.....How are those players going to get a Fargo above 599?

Also....playing in capped fields....limits your growth.....Way too many innings in a 599 and under match......The top guy in the field 599 gets good at playing "clean up"....and may even not care if the other guy shoots in the beginning of the rack....He/She knows they will get another shot at that rack......Then they finally step up to play a 700 player and get destroyed.....Missing a ball early may cost you multiple racks (if not the match).

EDIT:I know "why" TDs do it....still sucks for growth of the game

600-640 is no mans land around my neck of the woods.......Too good to play in the soft tournaments.....get pummeled in the open tournaments.

I believe Metro has a weekly 643 and under (how do you like that cutoff???). Once you're over 600 in AZ your options are limited. Over 643 and they're very few and far between, and even when they do let you play, most often it's handicapped against you.

 

[SEB]

I tend to agree. Certainly there are people who are more comfortable on one than the other, but if somebody is significantly better on a bar box, ultimately they're going to be better on the large table too.

600's don't start getting their asses kicked by 500's just because they move from one table to another...

You’re talking about the difference between a 600 to a 500 though.

People really dismiss it like it’s nothing but you’re talking about a skill gap of 2X!!!

A 600 is TWICE as good as a 500…so of course the table wouldn’t matter.

But a 620 vs a 635…yah man—equipment matters the smaller the gap between the players.

 

[David in FL]

You’re talking about the difference between a 600 to a 500 though.

People really dismiss it like it’s nothing but you’re talking about a skill gap of 2X!!!

A 600 is TWICE as good as a 500…so of course the table wouldn’t matter.

But a 620 vs a 635…yah man—equipment matters the smaller the gap between the players.

A 620 versus a 635 is very little difference.

Mike will tell you that any player can easily play +/- 50 points on any given day/match...

If that particular 635 beats up on the 620 at a regular basis on one table, he's going to do it on the other table.

The better player is going to be the better player regardless of equipment over the long run.

The key words there being "regular basis" and "long run"...

 

[skip100]

100 games is more than enough data to get a good estimate. I put together a Bradley-Terry model for my league and model errors were lowest when removing the impact of a starter rating at around 30 games, which is in line with the usual suggestions about sample sizes needed due to the Central Limit Theorem.

200 game robustness is all about CYA for FargoRate. It has no theoretical or practical basis from a predictive standpoint. That said, if I were Mike I probably would have done the same thing.

 

[BasementDweller]

A 620 versus a 635 is very little difference.

Mike will tell you that any player can easily play +/- 50 points on any given day/match...

If that particular 635 beats up on the 620 at a regular basis on one table, he's going to do it on the other table.

The better player is going to be the better player regardless of equipment over the long run.

The key words there being "regular basis" and "long run"...

That's the company line

And it's also what the data that we get to see will tell us. Some of us just know some stuff about pool that would "aid" in our betting choices.

In short -- a bar table can be "mastered" at a lower Fargo rating than can a 9 footer. I recently played in a bar table event where I played 3 flawless sets, where I didn't make a mistake, nor miss a ball. I've never done that on a nine footer. Don't worry -- I went on to have a bad set and paid my penance. Anyway, that's not just a matter of familiarity, as I just don't have the skill to play practically world class on a 9 footer for an extended period of time. Now, I understand a lot of this gets baked into everyone's rating, but it still doesn't tell the entire story. Some games can just veer towards mastery at a different point than others -- think playing 3 ball for example.

So what am I even saying here? There's just something counter intuitive about Fargo Ratings that we all know is true beneath the surface and it can be summed up in the fact that if our lives depended on us beating a world class player then almost all of us would want to play on a bar table and it doesn't just boil down to familiarity.

 

[tomatoshooter]

100 games is more than enough data to get a good estimate. I put together a Bradley-Terry model for my league and model errors were lowest when removing the impact of a starter rating at around 30 games, which is in line with the usual suggestions about sample sizes needed due to the Central Limit Theorem.

200 game robustness is all about CYA for FargoRate. It has no theoretical or practical basis from a predictive standpoint. That said, if I were Mike I probably would have done the same thing.

I was wondering about this. If a new player has a bad night and goes 2-5, 2-4 in a tournament, that's 13 games. I guess if a person is having multiple bad nights in a row, they might actually just be bad, or at least need to get their composure in serious competition.

 

[tomatoshooter]

if our lives depended on us beating a world class player then almost all of us would want to play on a bar table and it doesn't just boil down to familiarity

Better practice that lag, bring a comfy chair, and say your goodbyes. Pros miss one or two balls a set on the big table, on a bar box... they might never miss.

 

[Bob Jewett]

I was wondering about this. If a new player has a bad night and goes 2-5, 2-4 in a tournament, that's 13 games. I guess if a person is having multiple bad nights in a row, they might actually just be bad, or at least need to get their composure in serious competition.

A low number of games shows little. That's why FR requires 200 games for a rating to be official.

With only 30 games, the calculated FR might easily (10% of the time) be off by 100 points. With 200 games, the similar error is down around 40 points, and with 1000 games about 20 FR points.

Your hypothetical player could have lost 0-5, 0-4 just by chance -- he could have been playing his normal game. Assuming he was equal to his opponents, that's expected to happen one time in 500 outings, more or less. His one-event FR would be calculated as negative infinity.

 

[Coos Cues]

I know starter rating is supposed to not be meaningful as time goes on but I'm skeptical.

Hypothetical situation A Player is entered into the system by a league operator. Assigns a starter rating of 350. In a parallel universe the same player is entered into the system by the same league operator and assigned a starter rating of 650. From that moment forward in the parallel universes the two players who are actually the same player play the same matches against the same players and are entered into fargorate for each game.

When they reach the 200 game mark will both players have the same fargo? I can predict what Mike will say but I'm not convinced.

 

[skip100]

A low number of games shows little. That's why FR requires 200 games for a rating to be official.

With only 30 games, the calculated FR might easily (10% of the time) be off by 100 points. With 200 games, the similar error is down around 40 points, and with 1000 games about 20 FR points.

Your hypothetical player could have lost 0-5, 0-4 just by chance -- he could have been playing his normal game. Assuming he was equal to his opponents, that's expected to happen one time in 500 outings, more or less. His one-event FR would be calculated as negative infinity.

There is a good point here which is that in long races and tournament situations, results within a particular race or match can be correlated. This would decrease the number of units of measurement and suggests a larger number of games for a starter rating than the 30 shorthand. That said, we do see that long races tend to track with Fargo/Bradley-Terry predictions and lopsided matches will happen all the time due to the laws of probability.

At any rate, 200 is overkill no matter how you slice it. The purpose of 200 is optics, not predictions. From a predictive standpoint we are better off following the data from the first 50 games rather than putting 3/4 of the predictive mass on an arbitrary starter rating. If the 50-game rating is truly off by 100 points, then it will quickly adjust as more games come into the system.

If 200 really represents some kind of hard limit, then they should put their money where their mouth is and not show a rating at all.

 

[BRKNRUN]

I believe Metro has a weekly 643 and under (how do you like that cutoff???). Once you're over 600 in AZ your options are limited. Over 643 and they're very few and far between, and even when they do let you play, most often it's handicapped against you.

I'm gonna guess you are a 645+

I want to start playing in that .....My main problem for the last 5 years is I get up at 1 AM for work....I am DOT regulated....It would not be a good look for me if I was in a wreck and the lawyer asked me what I was doing from 7-1 the night before...

Gonna try and play in that 650 and under tour final.....but for that I will have to get up at midnight....race through my route....then break the speed limit getting home and/or over to the pool room for my first match......(I hardly ever get a bye....actually bye may be my middle name)

 

[skip100]

Your hypothetical player could have lost 0-5, 0-4 just by chance -- he could have been playing his normal game. Assuming he was equal to his opponents, that's expected to happen one time in 500 outings, more or less. His one-event FR would be calculated as negative infinity.

The negative infinity prediction could be avoided if they used a Bayesian approach with a prior and produced a maximum a posterior estimate rather than the pure MLE. This would be conceptually cleaner and predictively more accurate.

 

[tomatoshooter]

At any rate, 200 is overkill no matter how you slice it. The purpose of 200 is optics, not predictions.

He could make it a million games and a lot of people would still not believe his system works. As far as I can tell, it works very well but if he were to account for every variable then the ratings would only apply to a very limited set of circumstances that would rarely be appropriate for more than a handful of players.

 

[Bob Jewett]

The negative infinity prediction could be avoided

I'm sorry that I seem to have implied that Mike Page would ever publish -infinity as a legitimate FR even for a single-event performance.

 

[Cameron Smith]

That's the company line

And it's also what the data that we get to see will tell us. Some of us just know some stuff about pool that would "aid" in our betting choices.

In short -- a bar table can be "mastered" at a lower Fargo rating than can a 9 footer. I recently played in a bar table event where I played 3 flawless sets, where I didn't make a mistake, nor miss a ball. I've never done that on a nine footer. Don't worry -- I went on to have a bad set and paid my penance. Anyway, that's not just a matter of familiarity, as I just don't have the skill to play practically world class on a 9 footer for an extended period of time. Now, I understand a lot of this gets baked into everyone's rating, but it still doesn't tell the entire story. Some games can just veer towards mastery at a different point than others -- think playing 3 ball for example.

So what am I even saying here? There's just something counter intuitive about Fargo Ratings that we all know is true beneath the surface and it can be summed up in the fact that if our lives depended on us beating a world class player then almost all of us would want to play on a bar table and it doesn't just boil down to familiarity.

I wish we could get access to the data or even just data from pro tournaments. I’d love to just do my own analysis. Even if I found on my own that table size do not influence ratings on average, I’d still want to see if there is any data all that might support what people think is happening. For example, even showing that there are outliers.

I think that’s where those analysis videos can feel unsatisfying. It can easily be the case that our assumptions are based on the outliers and we’ve just disregarded everything else. But I would show that in my analysis.

 

[bpeacock]

Hey guys! Mike Page from Fargo Rate is going to join me for a podcast and I was curious if there were questions or topics you wanted discussed. I can’t promise we’ll get to everything but I just wanted a feel of what people want to dig into.

I know Mike is a member and contributor here on AZB so it’s not like you can’t ask him any question you want. But some things don’t fit well in short written responses.

Let me know and thank you!

Do you think APA leagues will be reporting to FR anytime soon? Our pool hall in central Mississippi is the only one reporting some matches (Monday night BCA league) and all of the others in central and southern Mississippi are APA.

 

[AF pool guy]

I know starter rating is supposed to not be meaningful as time goes on but I'm skeptical.

Hypothetical situation A Player is entered into the system by a league operator. Assigns a starter rating of 350. In a parallel universe the same player is entered into the system by the same league operator and assigned a starter rating of 650. From that moment forward in the parallel universes the two players who are actually the same player play the same matches against the same players and are entered into fargorate for each game.

When they reach the 200 game mark will both players have the same fargo? I can predict what Mike will say but I'm not convinced.

Yes, because the starter rating is removed entirely at 200 games. If I understand it correctly, as you progress to 200 games played the starter rating is discounted automatically. So if you have 50 games in the system then you have 150 imaginary games and it goes down for every real game entered into the system.


Sent from my iPhone using Tapatalk

 

[skip100]

I'm sorry that I seem to have implied that Mike Page would ever publish -infinity as a legitimate FR even for a single-event performance.

The iterative method doesn’t converge so I wonder how they handle these situations. Presumably some kind of early stopping?

For those wondering about the impact of the starter rating after 200 games, it is nil. In fact in the LMS you can see the player’s real rating as predicted by the games they have played. What shows up on the app and web site is the weighted average of the SR and the real rating. The SR is literally just for show.