FargoRate Blog

LMS Scoresheet Changes

Tue, 02 Mar 2021 22:38:47 Z

LMS Feature Announcement

FargoRate is excited to announce our newest enhancements to LMS. New scoresheet layout, support for scotch doubles matches, new help system. We think you will find the scoresheet changes make entering scores and accounting for substitute players much more straightforward, and everyone has been asking for scotch doubles support.

Scoresheet Changes

Until now, LMS has used horizontal scoresheets. These are scoresheets that are laid out left-to-right and scored as such. These work well when printed as it reduces the amount of writing on league night. They do not work well for operators when entering them into LMS. For example, adding subs is a particular pain.

We have now changed our scoresheets to a vertical layout. This new format makes building custom scoresheets effortless, entering subs a breeze and allows us to support scotch doubles matches.

Note: This change only affects newly created divisions. Existing divisions will continue to use the old scoresheet layout until the division is complete.

As we mentioned, scoresheets in LMS are now oriented vertically, and you fill them in from top to bottom. The first round of the match is at the top, with games in that round following directly beneath it. Each round contains its matches with the Home team on the left and the away team on the right. To enter a sub, set the sub's name in the game they first play, and the scoresheet will fill them in the remaining slots. If the original player comes back in, change that particular game, and the scoresheet will start using that player.

Scoresheet Builder

With the scoresheet builder, you can build scoresheet layouts that work the way your specific league works. You can have as many rounds, games, and players as you need, including scotch-doubles matches--and by simply dragging-and-dropping you can specify exactly how your scoresheet works. Refer to the following link for more information on the new scoresheet builder.

Please refer to our new help system for more information.

Couple Interesting stories from 2019 China Open 9-Ball

Sun, 08 Sep 2019 20:38:45 Z

First is this: both divisions were won by the highest rated player on the planet.

Jiaqing Wu (828, CHN), won the men’s division, and Siming Chen (highest-rated woman at 794, CHN) won the women’s division.

Siming Chen beat Rubilen Amit (PHI) for the women’s title.

The final four for the men came from three continents and featured a rare semifinal matchup between Wu and Shane Van Boening of the United States, who have been FargoRate numbers one and two for years. Their two previous matchups were at this year’s US Open, where Wu beat Van Boening 11-7, and at the 2014 China Open, where the score was reversed, i,e., Van Boening beat Wu 11-7. In yesterday’s match Wu gained a solid 4-game lead mid match that steadily melted away leading to a hill-hill game in which Van Boening found himself compelled to attempt and failed to pocket a high-risk rail-first shot. Wu ran out for the 11-10 win.

Anton Raga (PHI) beat Eklent Kaci [Kot Chee] (ALB) 11-8 in the other semifinal matchup.

A tale of two Filipinos

Anton (Anthony Cortes) Raga first garnered attention in the metro Manilla area of the Philippines after dominating school division events and private games in pool rooms as a teenager. At 22 now, he hasn’t travelled much. But he’s played enough that he was ranked world top 30 by FargoRate before the 2019 China Open. So to some not accustomed to hearing his name, he seems like a newcomer on the scene. But to FargoRate, he was already firmly in the company of the likes of Souquet and Oi and Kazakis and Alcaide and Woodward. FargoRate can assess this by the way be matches up with other Filipino players in, for instance, the Manny Pacquia Cup.

Raga had an amazing run here, and he lost 10-11 in the finals. So he was a mere one game away from winning the tournament. And his journey started in the qualifying rounds.

This performance skyrockets him to 4th on the FargoRate list and a rating of 820. Will he stay there? Does he belong there? We don’t know. He only has about 500 games in the system, so we will have to wait and see. Those around him have several thousand games in the system. He may very well belong there, and we are confident he belongs in the 800 club.

Jerico Banares is another Filipino player who came to the China Open. You likely haven’t seen his name, though, because he failed to advance from the preliminary rounds. You’ll see why that’s interesting in just a bit. Fans who follow professional pool tournaments likely won’t know Jerico Banares. But top players around the world who take a trip to the Philippines to test and hone their skill and maybe enjoy a beach know who Jerico is. And FargoRate knows who Jerico is. Before this tournament Mr. Banares was world #34, in the same ballpark as those listed above. And here’s the kicker. Jerico Banaras is now, following the 2019 China Open, in the 800 club. His performance at this event has raised his rating 8 points to 802 and to world #22, directly behind Yu Lung Chang and Albin Ouschan. Say what? Huh? I thought he failed to advance out of the prelims? 802??

Yes, that’s right. He performed sufficiently strongly in the preliminary matches—about 828-speed overall as a matter of fact—to raise his rating to 802. He beat some strong players by big margins and then lost a couple key gatekeeper matches 8-9 that prevented him from advancing. Too bad. Hopefully he’ll travel some more.

Earl Strickland

Thu, 05 Sep 2019 14:55:23 Z

Back in USA top FIVE ... Really?

If someone pointed out that iconic 9-Ball legend Earl Strickland

went two & out at the 2018 International 9-Ball Open last fall
went three & out at the US Open 9-Ball event this past spring
went three & out at the Predator World 10-Ball Championships this summer,

you might nod your head and think something like,

Yeah, well, OK…. Seems about right… fit as he might be, he’s 58 year’s old. I’m sure he still shoots straight, but what should we really expect from a player who was world 9-Ball champion seven years before the current world 9-Ball champion was born? He may get invited to things occasionally, but that’s more because of the attention he draws than him actually keeping up with the young guns.

If that’s where your head is at, grab a cup of coffee and be careful not to spill it.

• Earl has recorded 600 games in the last two years, and he’s performed at 790 speed for those games. There aren’t 50 players in the world at 790. There aren’t five players in the USA at 790. And that’s higher than he performed for the handful of years before that.

• Earl has played 77 games in the last year against opponents rated over 800. And he has an overall winning record for those games. Opponents include Van Boening, Shaw, Feijen, and Kaci. That’s right, he is 39 wins, 38 losses against these over-800 players in the last year in games played at Turning Stone, World Pool Masters, and International 9-Ball Open.

• Earl just last week performed at 850-speed for over 100 games en route to his second-place finish at Turning Stone, with a 9-8 win over Thorsten Hohmann, a 9-1 win over Jayson Shaw, and a 9-5 win over Shane Van Boening before losing 11-13 to Shane in the finals (aggregate 20 – 18 against Shane).

How does this stellar performance comport with the lackluster results noted above?

We should all recognize that with the deep fields being amassed for these major events, anybody can go two & out or fail to advance to an elimination phase or lose in the first round of the elimination phase without it signaling a decline in his or her game. At the US International 9-Ball Open for instance, Earl lost 11-8 to Niels Feijen and 11-9 to Alexander Kazakis. That can happen to anybody no matter how strong and no matter how in stroke.

At the US Open 9-Ball, following an 11-2 win over Pedro Botta of Florida (653), earl lost 10-11 to Dennis Hatch (775) and 10-11 to Erik Hjorliefson (748). Anytime a top player loses a match in a hill-hill situation, that player was just one roll away from prevailing in that match and then winning who knows how many more. FargoRate sees these three matches as 55 games played at 773 speed, nothing out of the ordinary for a top US player, just not distributing the wins in a way that advances in the tournament. Move along. Not much to see here. This in part is why performance ratings contain more information than do tournament finish positions. And ratings based on more games are more predictive than are tournament outcomes.

Earl Strickland’s Fargo Rating is now 784. That puts him at world number 64. And in the USA he is behind only Van Boening, Dechaine, Bergman, and Woodward.

And speaking of Dechaine, yes the window salesperson from Maine does still play and has also logged about 600 games in the last two years performing at 803 speed for those games.

So wipe up your coffee and let us know on facebook.com/fargorate what you think. Would you like to see Strickland play more? How about Dechaine? Would you like to see him play more? Is celebrating excellence a thing?

Android "Device not compatable" Message

Tue, 07 May 2019 16:20:58 Z

We've heard some users report that they're getting a "Device not compatible" error message in the Google Play Store when they try to install or upgrade the FargoRate Player app, even though their device is supposed to be compatible.

As we've investigated this app compatibility issue we found that several other major apps have run into this including Instagram (1B+ installs) and Clash of Clans (100M+ installs). It appears to be an issue with Google's Android operating system.

To fix the “your device is not compatible with this version” error message, try clearing the Google Play Store cache, and then data. Next, restart the Google Play Store and try installing the app again.

Pull down the notification bar on your Android device and hit the gear-shaped settings icon, or find “settings” in the application tray. From here navigate to Apps, or App Manager. Then scroll down and find Google Play Store. Select this, and tap Clear Cache or Data as shown below.

This should erase everything, and get rid of any corrupt files that seem to be causing this problem. Instagram and Supercell, maker of Clash of Clans, have heard from multiple users that this works.

Shakeup at Fargo Mountain Top

Mon, 29 Apr 2019 12:28:04 Z

Lots of pool in the last few weeks in Las Vegas NV USA with Matchroom Sports' inaugural go at the US Open 9-Ball Championships following the WPA Players Championship put on by WPA and CSI.

Jiaquin Wu (CHN) held the top spot on the FargoRate list from the time the ratings began in the summer of 2015 until the Fall of 2017, when he was overtaken by Shane Van Boening (USA). With his stunning performance at the US Open 9-Ball Championships last week in Las Vegas, Wu jumped 10 points from 817 to 827 and regained the top spot. Outside his 10-13 loss to Joshua Filler in the finals and another loss that will be discussed below, Wu won 8 matches with a combined score of 88 to 30.

Joshua Filler (GER) also had a stellar performance, winning the US Open event and jumping 5 points to 822 and two places to 3rd on the FargoRate list.

The 820 Club

Here is something good for pool. There are now three players rated over 820, one from Asia (APBU), one from Europe (EPBF), and one from North America (BCA).

Did you also notice?

A remarkable jump that may have fallen under many people's radar is that of Kai-Lun Hsu (TPE). Though he fell 9-11 to Haitao Liu (CHN) n the group of 16 at the US Open, Hsu jumped 20 points to 805 and 33 spots from 50 to 17. He did this by his performance in 15 matches between the WPA Players Championship events and the US Open. Strikingly, he played all three of the members of the 820 club, Wu, Van Boening, and Filler--a total of 45 games--with a combined score of 27 to 18. Hsu beat Filler 11-8 in round 3 on the Winner's side and then he beat Wu 11-3 the next round (winner's qualification).

Other notable advances are highlighted in yellow in the list below.

FargoRate Player App Release

Sat, 20 Apr 2019 14:41:28 Z

We are excited to announce the FargoRate Player App is now available in both the Apple App Store and the Google Play Store! The App--your connection to your FargoRate rating--comes in two modes: Free mode and Player mode

Linking to your game history

Your ID number to connect you to your game history is the number now found next to your name on fairmatch.fargorate.com. (Contact us if there is no number). Click "signup," enter this number and choose a password. This has no connection with any past account. You are creating a new account.

Free mode features

Free mode contains all the features that were in our FairMatch service (view your rating, search for players by name, mark players as favorites for easy following, find fair races and determine match odds) as well as a template for a curated list of upcoming tournaments and streams. Free mode is available at no cost to everyone.

Player mode features

Player mode contains more, starting with serious search capability. Search by name, city, state, country, rating range, established status, or any combination of the above. Then, sort the results however you choose. For example, want to find the best shooters in your state? Search by state and then sort by rating.

Also, with Player mode, you may view all of your matches that have been recorded for your rating (if you have a robustness of 1,000, you will see all 1,000 of those games). Drill-down to see your detailed results against any individual opponent along with an aggregate summary.

Finally Player mode connects to our League Management System (LMS) showing your upcoming league matches, team rosters, etc.

Player mode is available to two groups.

Those who pay a yearly fee of $12.99; the mode is active for a year from the time you upgrade.
Alternatively, Player mode is available at no cost to all eligible LMS users. Eligible LMS users include all BCAPL and USAPL players as well as leagues/players signed on directly through FargoRate. If your league is not currently on the list, have your league operator contact us at support@fargorate.com and we can take it from there.

Download Now

Thank You

We couldn’t continue to do what we do without all of the support we receive from you players. Thank you!

We work hard everyday to ensure we are delivering the most accurate ratings possible and would love to hear of ideas you have for future features in the app.

Stay tuned for blog posts at our blog that cover how to use all of the features. And, if you should still have questions, you can always contact us at support@fargorate.com.

Mixing games in FargoRate? A look At Corey Deuel and 8-Ball

Thu, 14 Mar 2019 19:34:53 Z

It’s cold in Fargo
Selfie with a Jaguar on the wrong side of a zoo barrier is a bad idea
Corey Deuel is good at 8-Ball

In a nobody-disagrees-with-this list, these statements rank high.

But is Corey Deuel’s Fargo Rating of 782—which is determined from 8-Ball, 9-Ball, and 10-Ball games combined—really reflective of his 8-Ball play? Aren’t some people way better at 8-Ball? And aren’t other people way better at 9-Ball? Shouldn’t there be separate 8-Ball and 9-Ball Fargo Ratings? How about a Rotation-on-a-bar-box rating? Or perhaps a template-rack-one-on-the-spot-9-Ball-on-a-blue-label-Diamond-bar-box-with-Simonis-860 rating?

That’s too many questions at once. So let’s set aside all but the first for now. Is 782 a good measure of Corey’s 8-Ball game? I mean come on, the guy’s an 8-Ball monster. He just beat Justin Bergman outright 105 to 102 games more than outrunning a few game spot. And who is stepping up to put players like Siming Chen, Boyes, Chinahkov, or Immonen, all within a few points of 782, in the box with Corey playing 8-Ball?

What I’d like to show you first is that—spoiler alert—yes, 782 is a fine measure of Corey’s stellar 8-Ball game. Then I’ll move on to the other questions, which involve some subtleties.

Corey Deuel's 8-Ball games

Corey has about 5700 games in FargoRate from July 2013 to now. The 1736 of these games that are 8-Ball were played in the following events

While these contribute to Corey’s Fargo Rating, their influence is overshadowed by the nearly 4,000 rotation games played in this time period.

Our strategy will be to divide these 1700 8-Ball games into clumps with opponents of different strength. Then we can compare how Corey actually did with how a player with a 782 Fargo Rating is expected to do. We do this kind of comparison a lot, so we’ll explain it a bit.

Clumping games by opponent strength

When you play 80 games against an opponent rated 700 and 20 games against an opponent rated 600, your expected total game wins are the same as if you played all 100 games against a hypothetical opponent rated 680. This is not the average of the ratings of your two opponents; that would be 650. Instead it is the weighted average of opponent ratings: 700 weighted by 80 games and 600 weighted by 20 games.

Suppose you won 50 games total out of that 100. There are two ways to analyze your performance that are really two sides of the same coin. The first is to note that 50 wins out of 100 games against a 680-rated opponent is the expected score if YOU were rated 680. We say you performed at 680-speed for those 100 games. Note that we can do this without even knowing what your actual rating is. The second way is to start with your actual rating and look at how many of those 100 games you were expected to win. If you are rated 663, for example, then you are expected to win 47% of your games against 680-rated opponents. We would note you were expected to win 47 games and you actually won 50.

Opponents with similar rating

The easiest place to start is with Corey’s games against opponents rated about like him. Here are Corey’s 8-Ball games against opponents that have an effective average rating of 780.3, about just over a point below Corey’s rating of 781.5.

Of the 214 games, Corey is expected to win 107, and he actually won 107. We can also say Corey is performing at 780.3 speed for these 214 games. There are swings to be sure. But the greater the number of games we look at, the more the swings wash out.

Higher-Rated opponents

Against this crowd of opponents rated 790 & up, with an effective average of 803.5, Corey is expected to win 46.2% of his games, or 232 out of 501 games. He actually won 235. This means he was performing about 785.6 speed for these games.

Lower-rated opponents

It is interesting to check out the group of opponents who average about 100 points below Corey because this crowd is expected to win about one in three games. Here is the group of opponents rated between 658 and 706 with an effective average of 680.1.

Corey was performing at 781.2 speed against this group. This sort of agreement extends down to opponents rated as low as 500.

How do we reconcile this?

These data show that across the board, Corey’s Fargo Rating of 782 is a pretty good indicator of his actual performance for just 8-Ball games. In fact we can do a separate optimization of Corey’s 9-Ball performance, and his performance in either game is within just a couple points of his overall Fargo Rating. There are three parts to reconciling this.

Part 1: Performance differences tend to be exaggerated

Overwhelmingly when we investigate perceived differences in performance

rotation games vs 8-Ball,
7-foot vs 9-foot,
different breaking/racking rules

data shows the actual differences, when they exist at all, are smaller than many people expect them to be. This is general. When you hear so-and-so plays poorly with a lead or poorly from behind or poorly against weak opponents or poorly against women or poorly on a particular table, you would get along pretty well assuming you are hearing nonsense. And when you do buy in, assume the effect is smaller than claimed.

Part 2: A player’s rating reflects what he or she does best

Your Fargo Rating is based on your own tournament history. So if you are predominantly a bar-table 8-Ball player, your record will consist primarily of bar-table 8-ball games. And if you are primarily a 9-foot-table 9-Ball player, like many of the top Europeans are, your record will primarily consist of 9-foot 9-Ball games. Because of this, all of our records tend to reflect what we do best. We think of this as a true skill determination for everyone. We each are accessing our true skill by doing what we do best.

The Europeans rated in and around Corey Deuel, players like Kazakis, Grabe, Alcaide, Makkonen, got that rating mostly playing big-table 9-Ball. It is probably true they would not keep up with Corey playing 8-Ball. But if any of them spent a summer as house pro at, for example, The Carom Room in Wisconsin, that would change. Actual skill is hard to change; game-specific skill is easier to change until it starts bumping into your actual skill. And, once again, Fargo Ratings are a measure of actual skill.

Part 3: There is no free lunch

So if Corey can outperform the Europeans in 8-Ball, wouldn’t he actually do that and therefore see his rating go up? Well he would if he actually played a lot of 8-Ball games against the Europeans. But there is another automatic balancing that goes on here. Think about who our opponents are when we play specific games. When we are at US Bar Table Championships, we tend to be playing opponents who are accessing their true skill on bar tables. When we play in a Eurotour event, we tend to be playing opponents who access their true skill playing 9-foot 9-ball. Look at who Corey’s opponents are above? They tend preferentially to be players we associate with 8-Ball. For this reason there is no easy path to a high rating.

Some players, like Corey Deuel, perform at their Fargo Rating on different size tables and playing different games. Others fall short for unfamiliar games, at least for a while.

Pool turning the corner?

Thu, 07 Mar 2019 18:28:47 Z

Unprecedented field for US Open 9-Ball

April of this year just might be the month we look back on and say, “That’s it; that’s the month that pool turned the corner.”

The evidence is strong. Did you catch that Matchroom Sport, a promoter that just doesn’t book losers, is blossoming out of its tried and true menu of few-participant exclusive events like the Mosconi Cup and World Pool Masters to bring us something BIG? The US Open reincarnation, with a $300,000 purse, will be April 21-26 in Las Vegas USA. The field is set, and as you’ll see below, expect the strongest assembly of pool players we’ve seen.

The picture is painted more clearly by the storms brewing on either side of the US Open 9-Ball. Just days before, CueSports International (CSI) teams with the World Pool-Billiard Association (WPA) to bring the $50,000 added WPA Players Championship at Griffs in Las Vegas. Did you catch that? WPA is partnering with CSI. That's a big deal and it's good for pool. And did you catch WPA President Ian Anderson’s comment that “our goal is for this event to be the first of more to follow”?

And a mere month after the US Open 9-Ball, CSI brings the US Open Bank Pool, Straight Pool, and One-Pocket all together to Griffs in Las Vegas with a projected $95,000 purse.

Finally, just two months later, league players from everywhere will share the elevators with the world’s best pool players at the Rio in Las Vegas. Predator and CSI team to bring the WORLD 10-Ball Championship to be played along with the BCAPL and USAPL International events. That’s the first 8-Ball, 9-Ball, or 10-Ball World Championship in the United States in more than two decades. And it will happen smack dab in the middle of 5,000 amateur pool players milling around. Did you catch that Predator is partnering with CSI? That’s a good thing.

That World 10-Ball is a first of three years committed, and the good news is after this coming July (2019) it moves to March (2020) along with CSI league events. So we get a second world championship just 7 months after the first. And another thing, it seems like Matchroom’s version of the US Open 9-Ball replaces the familiar Virginia event with a long history. But it really doesn’t. Something that looks an awful lot like that Virginia event in the Fall is humming along as the International 9-Ball Open, and we—members of the pool world—are the winners.

Strongest Field Assembled?

Pool has only had a rating system for a few years, so we cant very well comment on tournament fields from long ago. More recently, though, we can compare tournaments by counting how many of the world's top 100 pool players are competing in the same place at the same time. The highest we've seen in the past is 56, and we've seen that number twice. The first was at the October 2017 US Open 9-Ball Championship. The other time we saw 56 of the top 100 was at World 9-Ball in Qatar in December 2018. Other notable events in the last few years are World 9-Ball Qatar December 2017 with 54 and US International 9-Ball Open with 45 in October 2017.

It is in this context that we report an amazing 69 of the world's top 100 pool players are entered into the US Open 9-Ball event for next month. Those entered are in yellow here.

This is hands down the strongest field we've seen.

Battle of the Sexes: Rating Changes

Mon, 25 Feb 2019 18:19:14 Z

Siming won the match by winning the first two sets with scores of 21 to 19 and 21 to 20.

What does this do to the ratings?

We are often asked referring to Fargo Rating changes, what is the formula? . And somethings when people don't see or can't find a formula, they suspect we are being secretive about how rating changes are calculated. We are not. Turns out there is no formula. Computing rating changes following even a single matchup like this is a major exercise that uses a bank of dozens of computers in the cloud. The complication is that every player's rating depends on every other player's rating. So if, tentatively, my rating changes a bit. Then every other player's rating changes a bit in response. But if every else's rating changes a bit, mine is no longer right. So mine changes and then every other rating changes again. This happens until all ratings are consistent and are optimum. Still, even without a formula there are some things we can understand about rating changes.

The changes

Here are some rating changes following the 81 games played between Donny and Siming. In particular you can see Donny goes up 1.5 points (to 750) and Siming goes down 2.9 points (to 780). These results suggest several questions?

Whose rating went up?

The first thing to note may be surprising to some. Despite winning the match, Siming Chen's rating went down. And despite losing the match, Donny Mills's rating went up. The reason for this is we are always compared to an expectation based on our current rating. Because Siming was 35 or so points above Donny, she was expected not only to win but was expected to win by a larger margin than she did. She fell short of this expectation, and Donny exceeded the expectation set by his rating.

Why is Siming's rating change bigger than Donny's?

The magnitude of Siming's rating change, 2.9 points, is almost twice Donny's 1.5 point change. There are multiple reasons for this, and it is not easy to know their relative importance.

How many games?

The first thing to look at is each player's robustness--the number of games their rating is based on. A higher robustness in general goes along with a rating that is more reliable and less sensitive to new information. Here, both players have about 2300 games in the system, so that's not a big contributor.

How recent are the games?

The games in a player's record do not carry equal value. One example is more recent games carry more weight. It turns out, once again, these two players are similar on this measure. For both the most recent 750 games go back to Fall 2017, for instance.

How well known are the opponents?

Games against unknown opponents carry no information at all, and games against barely established opponents carry less information than do games against well established opponents. Here is where there is some difference. While both players have that vast majority of their games against opponents with established ratings, Siming has nearly all her games against women and many of those against Asian women. Donny's games, in contrast, are against a more diverse crowd.

Ratings of other top women?

A common question is how can a player's rating change when the player hasn't played. Here is a good example. Take Sha Sha Liu, for instance. Her rating is based upon play against other women and largely other Asian women. So when Siming goes down a couple points, that has a small effect on Sha Sha. Suddenly Sha Sha gets a bit less credit for the games she has won against Siming and is forgiven a little less for the games she lost against Siming. That by itself would be less than a tenth of a point. But it is a compounding issue. Other top women take this same hit. Then if all the top women are down a smidge, Siming's rating goes down a bit further for the same reason. The bleeding stops when everything is optimum, when an equilibrium is reached. Here you can see that Asian women dropped by about a point, and this is a notable part of the reason Siming dropped more than Donny was raised. This is basically a small tide shift for Asian women.

What about Karen Corr?

Notice that Karen Corr is largely immune to this tide shift. Her rating dropped by a small fraction of a point. This is because Karen has a much more diverse opponent pool than do the other top women players. In fact you can get a sense by looking at the other top women here which ones are coupled more diversely; they are the ones who experienced less than the full drop of 1 point or so.

Is this an "Ah ha! the beginning of the big rebalancing" situation?

No. There is no reason to think so. This kind of rebalancing happens all the time, and after the next notable piece of new information, there is just as likely to be a shift half a point down as half a point up. In fact these are the kinds of tide shifts that happen every day when players from Alaska get games against players from Arizona or players from Quebec get games against players from Florida. As more and more data couple different groups, these tide shifts get smaller and smaller.

Battle of the Sexes?

Mon, 18 Feb 2019 13:44:41 Z

There's a storm brewing in the pool world. At its eye is an upcoming match between Donny Mills of Clearwater, Florida (USA) and Siming Chen of Shanghai, China.

Who are these players and why is it a big deal?

Siming Chen

At 25-years old and at a rating of 783, Siming is the highest rated female in the world and has many women's titles under her belt, including World Games, Amway Cup, All Japan, and China Open. To put Siming's performance into perspective, she has, for the past four years, an overall winning record against each of Kelly Fischer, Allison Fisher, Jasmin Ouschan, and Karen Corr and a combined record of 112 to 91 against these opponents.

Donny Mills

Donny is a US pro-level player distinguished by (1) having a full-time job outside the pool world, and (2) being known to summon a high level game when it counts. He reached 82 in a race-to-100 9-ball match against Shane Van Boening, and at one US Open 9-Ball event he plowed through a number of champions (Boyes, Appleton, Kennedy, Morris) with lopsided scores only to lose two hill-hill heartbreakers against Morris and Immonen to reach the finals. Last year he beat Melling, Orcollo, and Bergman at Derby City Classic 9-Ball.

Women vs men

Few issues generate as much intense discussion as women vs men at pool. Opinions on the issue are strong, confident, and divergent. There is no debate there are many more high level male players than female players. The debated issue is why this is so.

To be fair, it is uncontested there is a large participation difference between men and women in pool. And given that, it is not clear it is even reasonable to seek explanations for performance differences. After all, if somebody points out the Ukraine has many more high level chess players than does Japan, nobody feels compelled to examine cognitive differences between Ukranians and Japanese. Or if someone points out Norway has many more high level hockey players than does Portugal, we don't feel compelled to examine the role of size or strength or speed differences in the populations or other biological factors.

In fact if you asked,

"why does Ukraine have better chess players than Japan?" or

"why does Norway have better hockey players than Portugal?"

you might expect a sarcastic response like,

"Duh... Is this a serious question? Maybe it is because they actually PLAY chess (hockey) in Ukraine (Norway)?"

And yet here, with pool and sex, we find despite clear participation differences a firehose of biological rationalizations.

Our view is explanations of male/female performances differences in pool are premature, that we need considerable more data to even begin to make conclusions about whether there is a performance difference to explain. Our view is also that results of this one matchup, as exciting as the matchup is, provide no answers to such questions.

But we are riveted nonetheless.

The match

Siming and Donny will play best of 3 sets 9-ball races to 21 games with the first set on Saturday February 23 and the conclusion on Sunday. The games will be winner breaks (from the box), rack your own with the 9-ball on the spot.

Fargo Ratings

One issue on some minds is whether Fargo Ratings for men and women are comparable. The concern is top women generally are rated using games played against other women, and top men generally are rated using games played against other men. We believe this is not a problem, and several tests with the data support our view. Still, the connections are indirect, and we are relying on a large number of individually weak connections. The kinds of connections that tend to match things up are, just making up an example,

Siming tends to beat Mary 7 to 3
Mary tends to play even with Victor
Victor against Donny tends to be 4 to 7

In this scenario, Siming is outperforming Donny by a bit. That is a weak thread, no doubt. But if that thread is a single strand of a rope, the connection can be strong.

Siming Chen has a Fargo rating of 783, and that puts her around world top 50, even though she essentially doesn't play opponents around her in rating. Donny Mills has a Fargo Rating of 749, and that puts him around top 20 in the United States.

It would be nice to be able to see this match as a test of Fargo Ratings. But unfortunately 100 or so games is just not enough. It requires 200 games before a player even has a Fargo Rating. It is not uncommon to see a player perform 50 points high or low for a string of 100 games. Still, these issues are on peoples's minds and that makes this match fascinating.

Prior matchups

While Siming has--at least on our radar--played little against men, she has played Donny Mills once before, a 10-ball race to 15 at Griff's in Las Vegas last summer. Siming won that match 15-13. Siming also entered the Austria Open about a year ago, where she lost 9-5 to Albin Ouschan (802) and won 9-5 against Daryl Peach (746).

The break

There is a lot of discussion about the break and break rules. When the one-ball is racked on the spot and players can make a wing-ball consistently, the objective is to have a shot on the one-ball, which banks near the side pocket and toward the head of the table. That makes 9-ball too easy at a high level. The rules here, following the International 9-Ball Open and the Derby City Classic, are to rack the 9-Ball on the spot and require three balls to pass the headstring. Here a wing-ball is tougher to make. Pocketing the the one-ball in the side is perhaps easier, but then the first shot is on the 2-ball and so control is lost.

Donny is known for figuring this kind of thing out effectively. Given that, his comment after the last match that Siming outbroke him is interesting. Is Siming good at figuring out the break? We don't know. But she has some pretty sporty hosts this week in Northern Virginia, and we might guess that besides she and her mom enjoying good hospitality and getting accustomed to the time change, there is some attention being paid to the break.

You can watch this match

This match is made possible by Roy's Basement, who have arranged to stream it for a modest fee so we can all see it. You can watch it here.

We are thankful to Roy's Basement for bringing this match to the world. Near as we can tell, Donny Mills and Siming Chen are great people, great players, and great ambassadors for our sport. Let's all watch this match!

FargoRate match odds --do they work?

Sat, 16 Feb 2019 17:56:19 Z

While rating differences give the chance each player wins an individual game, we show in a recent blog contribution how this information is parlayed into the chance each player wins a particular multi-game match. That is, rating differences lead to match odds.

How well does this work?

A look at the Eurotour

The Dynamic Billiard Eurotour is a series of 9-Ball events throughout Europe with a double-elimination qualifying phase that advances players to a final elimination round. All matches are a race to 9, so there are many matches to analyze. In particular we analyze seven events played in the last year in Italy, Austria, and Netherlands. These are good players, including 38 of the world's top 100. The established players have an average rating of 720 with 90% of them rated over 650. Readers in the US may recall those contending for the USA Mosconi-Cup played in one of these events in the Netherlands in September 2018.

That's a lot of matches

The list includes 1514 matches (over 20,000 games) between two players both with established Fargo Rating. With this number of matches, there are many in the vicinity of any particular rating difference.

Match odds

We are able, for example, to find what rating difference corresponds to match odds of 2-to-1 and then go look at all the matches that are close to that rating difference to see if they were in fact won near a 2-to-1 ratio. The results are pretty striking.

We'll start with 2-to-1. As shown here, a 2-to-1 win ratio is expected for a race to 9 match whenever the rating difference is 30 points.

There are several matches with a 30-point gap, such as Fedor Gorst (808) vs Petri Makkonen (778), Joshua Filler (816) vs David Alcaide (786), and Ruslan Chinakhov (784) vs Nick Malaj (754). But to get a big enough number, we need to go three points in either direction. What we find is 107 matches with a rating gap between 27 and 33 points. Of these, 71 were won by the higher-rated player and 35 were won by the lower-rated player. This is as close to 2-to-1 as you can get.

Next we seek the rating difference that corresponds with match odds of 4-to-1. That's 58 points. Now we have to go five points on either side to get a reasonable number of matches. This gives 97 matches of which 80 were won by the higher-rated player and 17 won by the lower-rated player. That's not bad. 78-to-19 would be as close to 4-to-1 as you can get.

We get odds of 7-to-1 with a rating gap of 80 points. Here, going five points on either side, we find 61-to-8. That's almost as close to 7-to-1 as you can get.

Finally, we note odds of 20-to-1 are expected with a 116-point rating gap. There are 41 matches with a gap that is within 6 points of 116 points, and 39 of them--all but 2--were won by the higher-rated player. That's as close to 20-to-1 as you can get.

Who played these 116-point-gap matches?

Here are some examples. There is David Alcaide (ESP, 786) vs Lukas Andersson (SWE, 673). David is a well known pro with over 5,000 games in the system. But where do Lukas's 2700 games come from? Well there are several Eurotour events as well as European Championships and many events in Upsala and Stockholm and elsewhere in Sweden. Another is Corey Deuel (USA, 781) vs Thomas Vanbroekhoven (BEL, 670). Corey has 9,500 games in the system, but where do Thomas's 577 games come from? Well Thomas has played in several Eurotour events as well as the Ardenenn Cup in Luxembourg and the BCA Pool League International event in Las Vegas.

What about bigger rating gaps?

There are 65 matches with a rating difference between 150 and 250 points. All of these were won by the higher-rated player except for one: Yvonne Ulmann-Hybler (GER, 570)'s 9-to-8 win over Benji Buckley (GBR, 724). Where do Yvonne's 1,200 games come from? Well she's played many events against Women throughout Europe including Eurotour events and German Championships. And she's also played in the women's division of the China Open. In addition, she's played a handful of events against men. This is the kind of coupling that insures Fargo Ratings are balanced between men and women and between Europe and Asia.

What about Benji Buckley? He has 4600 games in the system. Well Benji has many events from the Great Britain 9-Ball Tour. He also has played US Open and Steinway Classic in the United States as well as several regional events in Florida. And of course Benji has played in many Eurotour events.

How about odds for an individual match?

Will the odds computed for an individual match be as reliable as the ones shown here? In general no. Here when we look at many pairs of players, we benefit from some cancellation. That is, because of uncertainty in the individual ratings, some ratings gaps are over estimated and others are underestimated. And these effects tend to wash out. For an individual pair of players, it could be one player's rating is running a bit high and the other player's rating is running a bit low and these errors add in determining the rating gap. As usual, this is less of an issue as the players get more and more games in the system.

More data makes us better, always.

"So YOU’RE the Fu##er": chance-to-win-from-here explained

Thu, 17 Jan 2019 17:44:08 Z

You’ve likely seen a streamed pool match with Fargo Ratings in the bottom corners and a “chance to win from here” for each player that gets updated when the score changes. Where do these come from? Are they right? The answer to the second question is they generally are right. But there are limitations discussed toward the end of this contribution.

I’ll start with a story.

More than three years ago, when Fargo Ratings were new to the pool world, Shane Van Boening played Mike Dechaine a race-to-21 challenge match—The Tiger Challenge—at the BCAPL International event in Las Vegas. Adjacent to the playing arena was a giant screen, like the Ultrascreen at the movie theater, on which the players could not help but watch themselves live. For the first time, these things called Fargo Ratings were on the screen, and with a rating gap at the time of 39 points, poor Mr. Dechaine was forced to begin the match staring at this big 19% chance of winning the match. To make matters worse as he watched in horror, he actually had to watch himself watching. At a score of 4 to 4, Mr. Dechaine’s chances went up to 22%, and after winning the next game and leading in the match 5 to 4, Mr. Dechaine saw the high-water mark of 28%. A few games later, with a score of 7 to 8, the number was back to 19%, and then bad turned to worse. Following a 1% chance for Mike at a score of 13 to 19, Shane Van Boening finished off the match 21 to 13.
As Mr. Dechaine began to walk away from the arena with his cue case on his shoulder, he got introduced to me, Mike Page. Mike Dechaine shook my hand, mustered a big smile, and said, “So YOU’RE the fu##er.”

For the record, Mike Dechaine beat Shane Van Boening later that week in the US Open 8-Ball event. He then also beat Shane the following month at Turning Stone. And, fittingly, he beat Shane 21-to-20 the month after that in a race-to-21 challenge match at Snookers in Providence RI. And because of these wins, Mike Dechaine is now the only player on Planet Earth with an overall winning record against Shane Van Boening.

You’re welcome Mike.

The rating gap between these two is now down to 21 points, and Mike Dechaine plays more pool than most people think.

So where do these numbers come from?

I’m not going to tell you yet because that involves math, and I don’t want you to stop reading.

Analysis of Turning Stone matches

A statistical statement, like outcome Y has a 30% chance of being realized, means if you were repeat the event many times, 30% is the proportion of times you’d see outcome Y. And that is the way to test these statements, by repeating the event many times. But of course we can’t ask SVB and MD to play a large number of races to 21 with a starting score of 9 to 13.

Here is an alternative. Jayson Shaw just a few days ago won his 6th Turning Stone Classic title. It turns out 178 of the matches from that event were played between two players both with established Fargo Rating. That’s not a BIG number for statistics. But it’s not chump change either. We sorted those matches according to the rating difference between the players and put them into four groups.

Group A: 45 matches, rating dif. more than 100 points, average dif. of 156
Group B: 55 matches, rating dif. between 50 and 100 points,
average dif. of 74
Group C: 49 matches, rating dif. between 25 and 50 points, average dif. of 36
Group D: 29 matches, rating dif. between 0 and 25 points, average dif. of 16

Take group B as an example. When a race-to-9 match has a rating difference of 74 points, the chance of the lower-rated player winning is 14% as shown on the calculator here.

Note you can in the calculator always just choose 74 and 0 as the two ratings, whether you’re talking about Denis Grabe (779) vs. Tommy Tokoph (704) or John Morra (786) vs. Joey Cicero (713) or Shane Van Boening (821) vs. Jeremy Sossei (748) or Jennifer Barretta (657) vs. Steve Sutton (583). It is only the rating difference that matters. The 14% prediction suggests 8 out of 55 matches would be upsets, would be won by the lower-rated player. The actual number was 9, including Tommy’s 9-to-7 win over Denis and Steve’s 9-to-6 win over Jennifer.

Group A contains the most lopsided matches, and with a rating difference of 156 points, the chance of the lower-rated player pulling it off is 1.4%, or an expectation of about 1 match out of 45. Turns out there were 2 upsets out of 45 matches, Jed Jecen (570)’s 9-to-5 win over Matt Krah (678) and Brent Boemmel (648)’s 9-to-5 win over Mika Immonen (785).

There are expected to be more upsets for group C, where the average rating difference is just 36 points. In fact 15 out of 49 matches (30%) are expected to be won by the lower-rated player, and 15 out of 49 is the exact number won by the lower-rated player. These include, as examples, Petri Makkonen (779)’s win over Jayson Shaw (816) and Danny Hewitt (755)’s win over Alex Kazakis (792).

The expected vs actual match win percentages for the four groups are compared in the graphic below. The agreement is as strong as can be expected for the modest numbers of trials.

Does FargoRate predict a match result?

It is tempting to look at an isolated match upset like any of the ones I mentioned above and say the FargoRate prediction was wrong. The irony there is if those upsets didn’t exist, that’s when FargoRate would be wrong. Those upsets are actually predicted to occur. FargoRate predictions are really about counting over many events, not about a single event. For those who play Blackjack, FargoRate should be viewed much as “the book” is viewed in Blackjack. “The book” doesn’t say you will lose the hand if you split your 10s or double your 8 against a 9. But it does say you will be worse off than making the alternate choice if you do it a lot.

When do the chance-of-winning-from-here predictions break down?

The predictions treat the individual games as independent events. This breaks down for winner-breaks format when the players have more than a modest chance of break and run. In other words, in situations where game “packages” such as 3-packs and 4-packs are frequent, chance-to-win-from-here will be unreliable.

Where do the chance-to-win-from-here numbers come from?

If you’re still here and you are finding this blog post interesting, please share it. And if you are allergic to math, you are dismissed. If you are generally comfortable with numbers or are feeling a little brave, please stick around.

Fargo Ratings—or more correctly Fargo Rating differences—give a chance of each player winning an individual game. If the rating difference is positive number D, then the chance the better player wins a single game is 1/(1+2^(-D/100)). For a rating difference D of 0, this gives 0.5 (half). For D of 100, it gives 0.667 (two thirds). For D of 200, it gives 0.8. In equations that involve this quantity, it is usually denoted “p.” So p is the probability the higher-rated player wins a single game. That means the probability the lower-rated player wins the game is 1-p. This is usually called “q.” So p + q = 1. Every prediction using Fargo Ratings is just propagating p through various scenarios.

The chance of winning a single game, a race-to-1, is just p (or 100Xp if you like percentages).

Things already get sneakily tricky when you consider a race to 2. It seems like you can win a race to 2 two ways, by a score of 2-to-0 or by a score of 2-to-1. But there are actually three ways: (1) you win the first two games, (2) you lose the first game and win the next 2, and (3) you win the first, lose the second, and win the third. These look like

WW (probability pXp)
LWW (probability qXpXp)
WLW (probability pXqXp)

The probability of WW is pXp. This is where the assumption of games being independent events comes in. Probability of two things happening is only the product of the individual probabilities if the events don’t depend on each other. Note that the two ways of achieving a score of 2-to-1 each have the same probability, pXpXq, or p^2Xq. So when you add up the probabilities for the three scenarios, you can just consider it p^2 multiplied by 1 plus p^2Xq multiplied by 2.

Other races

For a race to 9, like at Turning Stone, a match can be won with a score of 9-to-0, 9-to-1, and so forth up to 9-to-8. The chance of the higher-rated player winning the match is the sum of the chances of winning by each individual score. And in each case, a particular specific outcome leading to 9-to-n has a probability of p^9Xq^n and this must be multiplied by the number of ways that score can be achieved. The 9-to-0 case is easy. The only way is the first player won 9 games in a row. For the 9-to-1 score, there are 9 ways to achieve it. The losing player could have won any of the first 9 games. Beyond that, the numbers get bigger fast. The number of ways to achieve 9-to-3 is 165, 9-to-5 is 1287, and 9-to-7 is 6435. This last number is combinatorial(9+7-1,7).

What about the “from here” part?

It is the same calculation. If you are down 6-to-4 in a race to 9, you must win 5 games before your opponent wins 3. This is the same as winning 5-to-0, 5-to-1, or 5-to-2 from here.

Anatomy of a Close-to-fair Tournament Tour

Sat, 05 Jan 2019 03:44:21 Z

New England 9-Ball Series

We examine here the results of over 5,000 pool games played in the New England 9-Ball Series between two players each with an established Fargo Rating (rating based on 200 or more games). These numbers are large enough to make meaningful comparisons between expected results and actual results. We’ll get to the comparison shortly. But first, some perspective.

Open tournaments – no handicaps?

There are good reasons to NOT handicap tournaments. Chief among them is the purpose of competition, outside of amusing ourselves or others, is to see who is better. There are many examples of successful non-handicapped tournament tours, like the Eurotour and The Joss Northeast 9-Ball Tour.

Modest handicaps?

Another class of tournament tour, of which the Mezz West 9-Ball Tour is a good example, reaches with an encouraging hand a little deeper into the pool of tournament players. Though developing players with a modest handicap as an incentive are pretty unlikely to finish high, a few with a little help may make it to the money rounds.

Room for something different?

The above approaches draw largely from players in the top 15% or so amongst tournament players by skill. Consider that 80% of the players who play pool competitively and who think about pool on their drive to work in the morning are rated in the 400s and 500s. So a different kind of tournament series, one that includes the top players but for which players in this latter group can be involved, can compete, and can even win, provides a nice complement to the others. It is also what makes the other tournaments become more popular and more competitive because it gets emerging players to meet and interact with better players and see higher-level play up close. These tournaments fuel the progression.

Use Fargo Ratings to handicap?

The landscape for running highly-handicapped tournaments like this has traditionally been treacherous and filled with landmines due to the need to rate players. Fargo Ratings take the burden off the promoter and provide a vehicle to mold the tournament matches according to a predetermined vision. The approach we’ll talk about here is FargoRate HOT handicaps using Chart R6. Here the match two players play depends only on the rating difference between the players.

These matches generally give a small advantage to the higher-rated player. Take as an example the 7-to-4 match. This is used when the rating difference is anywhere between 86 points and 106 points. A typical 7-to-4 match will have a rating difference in the middle of this range, i.e., 96 points. At that rating difference, the higher-rated and lower-rated players have a 55% and 45% chance respectively of winning the match. This—a modest advantage to the higher-rated player-- is typical of HOT-R6. There are also MEDIUM and MILD charts that work similarly but have less aggressive handicaps.

New England 9-Ball Series Range of Player Skill

The New England 9-Ball Series run by MD Promotions is in its 16th year and now uses Fargo Ratings to handicap matches. The tournaments typically have an upper bracket using HOT-R6, a lower bracket using HOT-R5, and then bring the top players from each group together in the end. Not all players have an established Fargo Rating. Unestablished players play by a preliminary rating that includes an approximate translation of the old subjective letter rating to a Fargo-Rating equivalent. The analysis here includes only matches for which both players have an established Fargo Rating.

The range of player skills is remarkably broad. The highest and lowest rated players in this 359-player analysis list have ratings of 800 and 254, respectively. The average rating is 524. The average rating of the 36-player top 10% group is 661. The average rating of the 36-player bottom 10% group is 366.

Analysis: What to expect.

All games are grouped together according to the rating difference between the players. So, for example a 700 playing a 600 and a 500 playing a 400 are grouped together and labeled by the rating difference, i.e., labeled 100 points in this case. Then games are added up for all matches with a rating difference of around this value. The expected result with a rating difference of 100 points is that the higher-rated player wins twice as many games as the lower-rated player. So if the ratings are good, then the actual result should reflect this, like, for example something close to 300-to-150 or 450-to-225. If pairs of players with a rating difference of 100 points played matches that were 4-to-2, or 6-to-3, or 8-to-4, or 10-to-5, then in any of these cases about half the matches would be won by the higher-rated player and half by the lower-rated player.

As an interesting aside, we are accustomed to longer races favoring the higher-rated player, that an upset is more likely in a shorter race. Note that this thinking, which many of us have internalized, applies only to lopsided matches and doesn’t apply to fair matches. So, for example, a 600 wins occasional unhandicapped races to 4 against a 700 but almost never wins a race to 10. Things change when we compare a 4-to-2 race to a 10-to-5 race. Now these players split in the long haul either way. There is no race-length effect.

The 6-to-6 match: rating differences between 0 and 27

These are the even matches, the matches with no handicap. They include, as examples, Kerry McAuliffe (651) v Rob Piersa (650) and Jenn Brown (424) v Eric Newell (422). They also include matches lopsided up to 27 points, like Eli Davenport (484) v Lida Mullendore (457). There are 225 of these matches, 1800 games, with an average rating difference of 12.9 points. When the rating difference is 12.9 points, the higher-rated player is expected to win 52.2% of the games. The total games record should be about 940 to 860. In fact it is 943 to 857. So the higher-rated player won 52.4% of the games, quite close to the 52.2% expectation. The way we will display this is by noting that for each 6.0 games won by a higher-rated player, we expect 5.5 games to be won by a lower-rated player. And the observed results match this.

The 6-to-5 match: rating differences between 28 and 51

There 124 of these matches with an average rating difference of 40.7 points. Higher-rate player is expected to win 57.0% of the games, and the actual percentage is 57.3%. Expected and actual match scores are both 6-to-4.5, as shown here.

7-to-5, 7-to-4, 8-to-4, 8-to-3, and 9-to-3 matches

This high level of agreement between expected and observed results continues in the more lopsided matches.

For 7-to-5, average rating difference of 67.0, expectation of 7-to-4.4; actual of 7-to-4.6.
For 7-to-4, average rating difference of 96.2, expectation of 7-to-3.6; actual of 7-to-3.6.
For 8-to-4, average rating difference of 126.4, expectation of 8-to-3.3; actual of 8-to-3.2.
For 8-to-3, average rating difference of 161.9 expectation of 8-to-2.6; actual of 7-to-2.7.
For 9-to-3, average rating difference of 201.9, expectation of 9-to-2.2; actual of 9-to-1.8.

Can C-Players give up 8-to-4?

One of the counterintuitive results of these analyses is that players rated under 500 can give spots like 8-to-4 and 8-to-3 and still retain the desired slight edge in the match. It is tough for some to wrap their heads around this in part because we are accustomed to the high numbers like 8 or 9 being associated with high-level play. Some imagine that winning an 8-to-3 match requires a few table runs and maybe even stringing a few racks together. Others imagine that the player going to 8 just can't make many mistakes. These thoughts are false. The player going to 8 generally can make the number of mistakes he usually makes in that number of games.

There are 10 matches in this analysis for which the higher-rated player is under 500 (average 474) and the lower-rated player is 110 or more points below (average rating of 344). For a 130-point gap, the expected game ratio is 8-to-3.2. The actual results for the 10 matches are in the ratio 8-to-2.8. The rating gap works the same in the vicinity of any skill level. That is a key feature of Fargo Ratings.

Data, Data, and more Data

The key to being able to run a high quality tournament tour like the New England 9-Ball Series using an approach like this is reliable ratings, and that comes from lots of data. While the tournaments themselves provide useful game data, relying solely on tournament data is a slow road. The real secret to getting a region firmly established and growing pool is league data. There is a tremendous benefit to the tournament scene from having reliable ratings even for league players who rarely or never enter tournaments. The reason is these players provide many strands in the web of connections between the tournament players.

FargoRate -- a look behind the curtain

Mon, 31 Dec 2018 13:49:33 Z

Fargo Ratings are ELO-like ratings for pool.

But there's more to the story. This is a description of what ELO ratings are, what FargoRate does, and how these are related. Nothing here is necessary to have a good functional understanding of Fargo Ratings and how they work. But for those who are interested, this is a peek behind the curtain.

The phrase ELO-like refers to any of a wide variety of schemes to rate competitors in activities that involve relative performance. They are named for Arpad Elo, a physics professor at Marquette University in Milwaukee who applied the ideas to chess a half century ago. So the ratings could be for pool, for chess, for arm wrestling, for tennis, for staring contests, for the game of chicken, for various online video games, for soccer, for ping pong, for football, or for any of a host of other activities.

This is to be distinguished from activities that involve absolute performance, like golf, bowling, weightlifting, running, and jumping. These don’t need an elaborate rating scheme: if Bubba in New Orleans long jumps 7.7 meters, and Vlad in Moscow jumps 7.9 meters, Vlad is better.

I describe here as clearly as I can first what is common to all the ELO-like rating approaches. Then I describe the implementation of the basic ELO scheme that FIDE (international chess organization) uses. FIDE’s approach, if not the gold standard, has history on its side. And it is the easiest to understand and implement. I follow this by a brief description of improvements to the FIDE approach (what we refer to as the Old Fargo Rating approach as well as other schemes like the Glicko approaches in chess). Finally I describe something quite different: what FargoRate actually does.

What is the same about all these ratings schemes?

The basic feature common to all ELO-like approaches is this:

A’s chance beating B depends only on the rating difference between A and B.

That indented, bold, italic statement is it. Nothing else needs to be said.

Where’s the Math?

Believe it or not, though, that simple statement says more than it seems like it says and actually has mathematics in it. The indented, bold, italic statement forces a particular kind of relation between rating difference and chance of each player winning, namely, a player’s chance of winning must follow something raised to the POWER of the rating difference.

chance of winning ~ (##)^(rating difference)

Here "^" means raiseed-to-the-power-of. Nothing else will work. No other choice is consistent with the indented, bold, italic statement.

One way to see the mathematics in this statement is to recognize the phrase “depends only on” can be turned into “=” once the particular dependence is expressed. So “chance” is on one side of the equation and “rating difference” is on the other side. The phrase “chance” refers to, for example, player A’s wins divided by the total number of games or similarly player A’s wins divided player B’s wins. Importantly it is a ratio involving players A and B. The right hand side is a difference involving players A and B. Anytime multiplication or division happens on one side of an equation and addition or subtraction happens on the other side, there is no choice but to find the above form, or, equivalently, that the rating difference is related to the logarithm of the chance

Logarithm(chance) ~ rating difference

The important point here is that any equations that relate chance or expectation to rating difference do not come from a vacuum. Other than an arbitrary choice of the size of a rating point, the relations are forced by the indented, bold, italic statement.

For Fargo Ratings, two players will win games in the ratio,

ratio = 2^(rating difference/100)

For many of the other schemes, players will win in the ratio,

ratio = 10^(rating difference/400)

The distinctions between these (note the 2 versus the 10 and the 100 versus the 400) are inconsequential and are just a matter of taste. Both these relations, once again, are just a restatement of the indented, bold, italic sentence above. They add nothing to it. What this means is if you have accurate ratings—the all elusive true ratings--for two players for any of these activities, you can calculate the chance of each player winning the next game.

Algorithm?

The Mark Zuckerberg character In the Facebook movie (Social Network) says to his friend, Eduardo, with some urgency in his voice,

“ I need the algorithm you use to rank chess players…
We’re ranking girls…
I need the algorithm
I need the algorithm…”
And Eduardo writes this on the dormitory window:

Note the similarity with the expressions above, the 10^(rating difference/400). The problem, though, is that as cool as it sounds in the movie, this relation is not an algorithm and is nothing more than a restatement of the underlined sentence.

An algorithm is a set of instructions, something you might program a computer to execute. If the subject is home-baked cookies, the algorithm is the recipe. The above, by contrast, is at best a cookie. And it only tastes good if the ratings are accurate. Calculating the chance of each player winning from the ratings—like eating the cookie-- is the easy part. Getting the ratings is the hard part. These equations say nothing about how to get the ratings.

A Sad Truth

Before discussing different rating approaches, we must acknowledge a sad truth. The ratings in Eduardo’s scribble? We don’t know them and never will. Yes, every participant has a true rating-one that we might imagine is tattooed to the inside to the participant’s chest where we can’t see it. The participant also has a tentative rating, one that perhaps is written on an erasable sign he hangs around his neck.

The usual approach to getting the ratings goes something like this. Start with a guess for everybody’s rating—maybe everybody starts at the same number or maybe you use some other criteria to make an educated guess. These initial ratings are highly tentative, and the goal is to be informed by actual game results to make them better. Even when they get better, and even when they get a lot better, our best current ratings are always tentative and are open to being informed by even more actual results. They are still the ones written on the sign around the player’s neck and not the ones tattooed to the inside of the player’s chest.

What’s different about the rating schemes?

Exploring a particular example of a rating update helps to illustrate the difference between the common approaches. In our example, two players are going to play 10 games. Their current ratings—the tentative ones we work with-- lead to an expectation of how many games each player will win. This is easy to calculate. Suppose Player A is rated 575 and B is 525. The first thing to do is compute a transformed version of the ratings to make life easier. A’s transformed rating is 2^(575/100). We’ll call that TA. B’s transformed rating, TB, is 2^(525/100). These two players are expected to win games in the ratio TA/TB.

It is that simple.

So for 10 games, player A is expected to win EA = 5.9 games, where

EA = (TA/(TA + TB)) * 10.

Player B is expected to win EB =4.1 games.

Now the match happens and the actual score is 7 to 3. So player A exceeded his or her expectation by 1.1 games (7.0 - 5.9). And player B fell short of his or her expectation by 1.1 games.

A’s rating should go up and B’s rating should go down. And it makes sense the amount they go up and down should be higher if the actual results differed more from the expectation. So we propose

Player A’s rating change = K * (1.1)

Player B’s rating change = K’ * (-1.1)

The big problem is these new factors K and K’ just appeared. So you want to increase A’s rating. But by how much? You have to make a choice for K. If you choose K too small, then ratings adjust very slowly and are too stable. If you choose it too big, then you will overshoot and the ratings will tend to bounce around too much.

There has been a lot futzing around about what to use for K. Maybe you analyze historical data to help you out; Maybe K is different for higher and lower rated players; Maybe K depends on how many previous games each player has played, and so forth. Most implementations of ELO schemes differ by how they choose K and what it depends upon. FIDE, the international chess organization, has probably the most celebrated implementation of an ELO rating scheme, and they use a rather unsophisticated choice for K.

Here is an excerpt from the FIDE website.

8.56
K is the development coefficient.
K = 40 for a player new to the rating list until he has completed events with at least 30 games
K = 20 as long as a player's rating remains under 2400.
K = 10 once a player's published rating has reached 2400 and remains at that level subsequently, even if the rating drops below 2400.
K = 40 for all players until their 18th birthday, as long as their rating remains under 2300.

So FIDE is just guessing.

Old Fargo Ratings

When we derived, back in 2002, the maximum likelihood approach described later, one of the byproducts of our efforts was a theoretical estimate for a sensible K value. In particular, if A and B play 10 games, then the K value for A is

K = 630 * (RB-10)/(RA*RB).

Here RA and RB are the robustness values (total number of games, including current and previous games) for A and B. Note some interesting things about this K value. First, suppose a player’s opponent is a complete unknown. Intuitively the first player’s rating shouldn’t be affected by winning or losing against an opponent for whom we have no information. Sure enough, in this case RB=10 and this makes K equal to zero. Likewise, if the opponent is poorly established, K will be smaller. And finally if the first player himself is well established, K will be smaller.

The simple ELO update with the above value for K is what we did for years before we began doing the ab initio global optimization (described below) in late 2014.

Mark Glickman has developed improved approaches for chess (Glicko and Glicko-2 systems) that similarly choose sensible values for K that take into account how well established are a player’s and the player’s opponent’s ratings.

Old Fargo and Glicko are improvements on the simple ELO scheme that FIDE uses, and FIDE should be using something better. Likely they are not because,

reluctance to lose the feature that players can simply compute their own rating change (transparency)
not wanting an angry mob of grand-master chess players who lost a spot or two on the ranking list
politics

FargoRate: A Different Direction

In principle, using the above schemes, player ratings will eventually get to a reasonable estimate of the true rating that forgets the particular choice of the starting guess and that is independent of the particular choice for K. This is true provided everybody plays a very large number of games and the choice for K is sensible. In practice, though, many players in the system don’t play a large number of games, and the starter-rating choice lingers. This is a particularly vexing problem with two nearly isolated groups, one of which is rated too high relative to the other.

A player’s true rating—the one we don’t have—drifts in time as the player’s skill improves or deteriorates. That player’s tentative rating—the one we do have--also changes in time due to skill changes. But the much bigger reason the tentative rating changes in time is that more data is moving the tentative rating closer to the elusive true rating.

With this in mind, imagine we either ignore the slow drifting of true skill or we limit ourselves to games played during a period of time sufficiently short that we can safely ignore actual skill changes.

Now rather than using a sequential update scheme like all of the above, we imagine we take our large number of players and assume nothing about them. Then we take our very large number of games that have been played amongst all these players and we imagine all those games are played out in front of us right now.

There exists a set of tentative ratings for every player in the system such that the matches—all the matches amongst all the players—turning out exactly like they did was most likely. This in the field of statistical inference is called the maximum likelihood approach.

Maximum Likelihood

Let’s start with a simple example of the maximum likelihood idea. Suppose a new player—Victor—moves to town and plays 20 games against Leroy, a well established player. The question is how does Victor play compared to Leroy? What is the rating difference between Victor and Leroy? Suppose each player wins 10 games. So the score after 20 games is 10 to 10. It seems reasonable to suggest they play about the same—have the same rating.

But of course this doesn’t have to be the case. What is it about that intuitive choice that makes it better than other choices? How can we justify the tentative proposal that these players have the same rating over alternative proposals?

Here are some possibilities that can be true even with the 10-to-10 game history.

A. Victor is a much better player than Leroy
B. Victor is a somewhat better player than Leroy
C. Victor and Leroy play the same
D. Victor is a somewhat weaker player than Leroy
E. Victor is a much weaker player than Leroy

If A is true, then Victor had an unusually poor day or Leroy had an unusually good day for the 20 games or both. That is, given A, the actual results are possible but pretty unlikely. Likewise, given D, the actual results are somewhat unlikely. And given C, the actual results are somewhat more likely. These likelihoods form a curve that looks something like the following:

The maximum likelihood idea is we choose as our best tentative rating difference the one for which the actual results were most likely. So we choose C, that Victor plays at the same level as Leroy.

How does the situation change if Victor and Leroy played 200 games with a score of 100 to 100 instead of 20 games with a score of 10 to 10? Now possibilities A and E are extremely unlikely, and possibilities B and D are more unlikely than before. The curve looks more like the following:

So after 200 games, the likelihood is more peaked at the maximum and we are more confident our tentative rating difference is closer to the true rating difference.

The left-right direction in the curves of the above two figures is the rating difference, and the up direction is the likelihood. If we add a third player, then there are two rating differences. One can be left-right and the other front-back. The up direction is still the likelihood.

So instead of a likelihood curve, we have a likelihood surface. Our best tentative ratings are the ones that reflect the top of the mountain.

With thousands of players there are thousands of “directions.” We can’t display it with a plot, but the surface becomes a many-dimensional hypersurface, and the concept of a likelihood mountaintop is still valid. Once again, our best tentative ratings are the ones for which the actual game results –all of them—turning out the way they did is most likely.

Locating the top of the mountain is not easy, but it is a well defined mathematical problem. Each day, as new games are added to the system, the detailed shape of the mountain shifts. As more data is added for the same group of players, the mountain generally becomes more peaked. Importantly though, the location of the peak also shifts, and this means optimum ratings for the players shift with the new data.

In general, the mountain is sharply peaked in some directions and relatively flat in others, and the curvatures in different directions give us information about variances and covariances—about how confident we are in a particular player’s rating and how sensitive is one player’s rating to another player’s rating. In fact we can compute, at the mountain peak, something called the Fisher Information Matrix that has this curvature information. But in any case the optimum ratings are what they are. They don’t depend on the details of the path or approach we take in climbing the mountain.

Every day, FargoRate constructs the mountain for all the available game data amongst all the players. Current games are given full weight, and there is an exponential decay of the weight of past games such that 3-year-old games contribute half, 6-year-old games contribute a quarter, and so forth. This time decay helps account for actual skill changes Every day the mountain is scaled to determine the optimum ratings.

We refer to this process as ab initio global optimization. The Latin phrase ab initio means from the beginning, or from first principles.

Sisyphus has it easy

As I write Sisyphus is rolling his big boulder up the hill as he will do for eternity only to have it roll down to where he must start over again. Like Sisyphus, FargoRate must scale the mountain every day. But unlike Sisyphus, FargoRate must begin by actually constructing the day’s mountain.

Getting pool on the right track

Fri, 28 Dec 2018 23:07:10 Z

On Fargo Ratings and the future of pool.

But first, a diversion.

An amazing thing happened over a few days in May of 1886. Thousands of miles of railroad tracks in southern United States got narrowed.

In preparation for the sudden change, they removed many spikes from one of the rails and designed wheels for new locomotives in the shape of a dish, so that on the big day, when the rail was shifted, the wheels could be turned inward to fit the new narrower track gauge. Now, finally, a train could bring people from Atlanta to Richmond and on to Boston, and clothing could get to Chicago.

Accomplishment or Failure?

Rather than seeing this as a post-Civil-War accomplishment, join me in seeing it a pre-Civil-War failure, a failure of the southern states to see railroads as something more than a local effort to get cotton bales a hundred miles to a waterway. Had the 10,000 miles of tracks in the south been uniform during the war, like the 20,000 miles of tracks in the north already were, troops and supplies could have traveled efficiently from Atlanta to Richmond. The outcome of the bloodiest war in US history might have been different. Or better, had the tracks been uniform before the war, the economy of the south likely would have been more joined with that of the north and that war may have been avoided altogether.

And we were slow learners.

Two decades later Baltimore residents watched more that two thousand buildings burn down when reinforcing firefighters from New York, Philadelphia, and Washington DC stood helpless with hose couplings that failed to match the fire hydrants.

The development of uniform standards—something we’ve thankfully become better at— now happens behind the scenes and before most of us realize the importance. The fact Walmart stocks products from 70 different countries and the president’s tweets are seen around the world are testament to this. Interchangeability and compatibility, cornerstones for developing and expanding markets worldwide, lead to cheaper and better products and services.

But this is about pool

Pool is just a game; civilization doesn’t depend upon its well being. Still, people reading this likely see something worth nurturing in pool, and it is worth thinking about how best to feed this passion. So what IS the best way to nurture the growth of pool?

Pool will be positioned to flourish when, and only when, it rides around on uniform tracks. Uniform tracks for pool are a system to rate player skill. Pool needs a universal rating system.

The Cocooning effect of provincial approaches to player rating

At first glance, this may seem to exaggerate the role of player ratings, that player ratings might be expected to fall below rules and below equipment on the list of items to standardize to foster growth of pool. But a deeper look tells a different story. Rules are not like train tracks; rules can be changed immediately when the will exists to do so. And there are plenty of examples in pool of successful organized play with events in different locations on similar but not identical equipment.

Ratings play a more insidious role than do rules or equipment. An organizational structure in pool, whether it is local or regional or national, needs to protect the integrity of its competition, and it usually attempts to do this by creating some sort of internal system to rate players.

And here is the problem. Generally the more effort put into a rating system, the more the organization flourishes. But the same efforts that make the organization flourish also make it more insular, firm up the walls of the cocoon it creates. League systems and tournament tours, when done well, become like the efficient 200-mile railroad run bringing cotton bails to the waterway. Look locally and narrowly, and the best are functioning fine. But growth is stunted because the tracks don’t match. The rating system is what separates the outside from the inside. Pool organizations are like countries in Africa that don’t trade with one another because they speak different languages.

Soap bubbles or cocoons?

Pool grows by interaction at the edges, by adjacent regional tours combining to create something better than the sum of their parts, by league systems merging both to enjoy an economy of scale and to increase opportunities for their players, by casual league play being seamlessly connected to more serious league play, by local tournament play being connected to regional, national, and international tournament play, and by league play being connected to tournament play.

Pool organizational structures should be like soap bubbles and not like cocoons. Some expand, some pop. Some bounce off one another. And importantly some merge to form larger bubbles in an effort to grow and reduce surface area.

Adopting a universal rating system doesn’t force the interaction. It doesn’t, in and of itself, do any of these things. But it does dissolve cocoons. By destroying a major deterrent to growth, a universal rating system encourages and facilitates growth by contrast. It allows organic growth. It allows growth to occur where growth wants to occur. And that is what has been missing in pool.

FargoRate: head in the cloud

A bank of several dozen high-speed computers in the cloud churn out, as we write, ratings for nearly 120,000 players worldwide based on a spider web of nearly ten million games played amongst them. Everything about FargoRate is increasing rapidly. Pool now has the most sophisticated rating system of any interactive sport—more so than chess, than tennis, than badminton, than table tennis. The FargoRate ab initio Global Optimization starts from scratch each day and finds the optimum ratings for all players based on all games, the ratings that best predict the actual outcome.

Because of this, a 555 league player in Arizona and a 555 tournament player in Wisconsin play at the same level. Because of this any player may track his or her progress in a meaningful way. Because of this a director of a skill-restricted tournament in one state may safely allow entry of otherwise unknown players from another state. Because of this national or international level competitions may safely segregate players by skill. Because of this promoters of invitational events may choose invitees based on skill, and—if they choose--may seed entries in the competition based on skill. Because of this, world member organizations may choose players to represent their regions in world competitions.

FargoRate now has uniform train tracks in place all around the world. As more and more data from public competitions enters the system, FargoRate gets increasingly more reliable ratings for more and more players. Our goal is to rate all pool players worldwide—that’s ALL players worldwide--on the same scale.

Last train to the future of pool.

All Aboard.

New LMS Features

Fri, 21 Dec 2018 17:09:59 Z

We work hard to ensure that LMS meets the needs league operators. We are happy to announce that we have recently added several new features to LMS that expand the types of league formats we support. These are Custom Scoresheet Layouts and Set-based division formats.

Custom Scoresheet Layouts

We recognize that not every league is the same and neither are their scoresheets. You can now use our new Scoresheet Layout Builder to setup the exact structure of your scoresheets. For example, we provide a default scoresheet layout that rotates the visiting team's players in a particular order. What if your scoresheet rotates the players in a different order? This is where the layout builder comes in. It provides the ultimate flexibility when setting your rotation. You can choose the default, reverse it, or DESIGN YOUR OWN! Farther down in this post are detailed instructions on how to use it.

Set-based Division Formats

Run a division that uses a match-play format? We now have you covered as well. You can now build divisions that are based upon sets played between players rather than rounds of games. Set-based matches often go by different names: match-play, singles, sets, etc. Whichever the case, we have you covered. You can now tell LMS whether your division is "Round"-based or "Set"-based, provide the race-lengths (whether by points or games won), and LMS will do the rest.

Instructions

Custom Scoresheet Layouts

When creating a new division, the "Players per Team", "Number of Rounds", "Games per Round" settings are now removed from the main division settings screen and are set by clicking the "SCORESHEET LAYOUT" button.

The Scoresheet Layout Builder contains the settings that influence how the scoresheet is structured. You can set the "Players per team", "Number of Rounds", and "Games per Round" and see the scoresheet layout change immediately. If the default layout isn't exactly what you need, you can customize it. If you simply want to reverse the player rotation order, you can check the "Reverse visiting order?" box and the order in which the visiting players are rotated from round to round will be reversed from the default.

If you need even more control over the layout, you can click the top and bottom boxes for each game matchup for each round. Here is an example of setting the first few games.

Step 1) Set your "Players per Team", "Number of Rounds", and "Games per Round".

Step 2) Click on the first scoring box for the HOME TEAM (top) and then click on the corresponding box you want for the VISITING TEAM (bottom).

Step 3) Continue clicking on the boxes until you have them all set.

Step 4) Click the SAVE button to save the changes.