Posts by Pat Ross

Ryder Cup: Head to Head Picks by the Golf Engine

We take a break from football to bring you a golf engine which uses machine learning to evaluate 1,500 different statistics for every golfer on the PGA Tour over each tournament since 2004. The analysis of this massive dataset provides an opportunity to predict players that are due to go low. 

The engine looks at how

We take a break from football to bring you a golf engine which uses machine learning to evaluate 1,500 different statistics for every golfer on the PGA Tour over each tournament since 2004. The analysis of this massive dataset provides an opportunity to predict players that are due to go low. 

The engine looks at how each statistical set contributes to what we can expect from players on this stage, at this tournament. It’s a complex web of information that can only be properly analyzed by a math engine, yet yields some objectively surprising results.

The United States carries a heavy advantage in the 2018 Ryder Cup as seen by the Las Vegas Odds. This is neither surprising nor out of the ordinary going into this event. The majority of the best players in the world, its professional tournaments and money in the sport come from the US.

The Europeans do however carry a lot of pride for inventing the game. And, tend to enjoy embarrassing the US on a semi-annual basis, regardless of how stacked the odds always seem against them.

We suspect this year will be no different, and in the end, it will be a very close and exciting tournament even though the US should walk-away cleanly with the Cup (spoiler alert – they won’t, drama inevitably to ensue).

Should make for a fun weekend regardless as this is so-much-more than a golf tournament and there a nearly unlimited possibilities for prop-bets out there.

Head to Head Picks

Use the drop-down menu to select the individual US and European player and see the projected head to head match-up winner.




 

How the Golf Engine makes its picks

We use machine learning to evaluate 1,500 different statistics for every golfer on the PGA Tour over each tournament since 2004. The analysis of this massive dataset allows gives us an opportunity to predict players that are due to go low.

The machine learns how these statistics can become a unique strength or glaring weakness for each golfer by comparing tens of thousands of different combinations and separating the patterns from the noise.

The resulting ‘model’ is able to ‘deep dive’ and determine when to expect low rounds from a pro, given their unique style of play. These calculations are next to impossible to do quickly and certainly without personal and subjective biases, until now.

Historical Results

View the golf engine’s picks and results for the British Open and PGA Championship or get the 2018 Ryder Cup Odds here.

 

Contributor to National Football Post & sports nut with training in statistics, machine learning, and data analysis from Galvanize – Seattle campus. Alumni of University of Colorado and University of Washington. Occasional boater, skier, and golfer.

Read More 359 Words

The Golf Engine Predicts the Best Bets for the PGA Championship

We take a break from football to bring you a  golf engine which uses machine learning to evaluate 1,500 different statistics for every golfer on the PGA Tour over each tournament since 2004. The analysis of this massive dataset provides an opportunity to predict players that are due to go low. 

The engine

We take a break from football to bring you a  golf engine which uses machine learning to evaluate 1,500 different statistics for every golfer on the PGA Tour over each tournament since 2004. The analysis of this massive dataset provides an opportunity to predict players that are due to go low. 

The engine looks at how each statistical set contributes to what we can expect from players on this stage, at this tournament. It’s a complex web of information that can only be properly analyzed by a math engine, yet yields some objectively surprising results.

This year’s Championship is no exception as the model is calling for Phil Mickelson (100/1 odds) to break into the top 5, with Kyle Stanley (80/1 odds) and Tony Finau (40/1 odds) cracking the top 10!

Some surprises:

Kyle Stanley (80/1 odds), Tony Finau (40/1) and Phil Mickelson (100/1) inside the top 10.

Zach Johnson (100/1) and Patrick Cantlay (50/1) inside the top 15.

Perhaps just as surprising are the golfers that may underperform this week. Two recent Major Championship winners were projected outside of our top 20 – Francesco Molinari (33/1 ) and Jordan Spieth (20/1).

Notable left-outs:

Tiger Woods (28/1), Henrik Stenson (50/1) and Alex Noren (50/1) finishing outside the top 25!

Rickie Fowler (22/1), Justin Rose (22/1), Patrick Reed (35/1) and Bubba Watson (50/1) all finishing outside the top 10.

A few more points of note:

Top 5

Dustin Johnson (8/1 odds). Getting the top call from both the oddsmakers and the model is not hugely surprising. The TOUR leader in Fedex Cup points and 3-time winner this season is also leading the TOUR in Strokes Gained: off-the-Tee, Strokes Gained: Tee-to-Green and Overall Strokes Gained. Oh, and he’s fresh off of a dominating win 2 weeks back and a low-round 64 at the WGC: Bridgestone this past Sunday. He is a threat to win anywhere and everywhere.

Justin Thomas (14-1 odds). No surprise here considering his dominating performance at last week’s WGC: Bridgestone Invitational.

Jason Day (20/1 odds). A two-time winner in 2018 (so far) and already a PGA Champion, putting better than 90% from inside 10 feet this season… just let that sink in.

Jon Rahm (25/1 odds). The results haven’t been there of late, but his game appears to be rounding back into form in perfect time for the final Major of the year and with the Fedex Cup playoffs just around the corner.

Phil Mickelson (100/1 odds). I’m not sure which is more surprising… seeing “Lefty” in our model’s top 5 at the age of 43 or the 5-time Major Champion at such long odds. Either way, he looks like a the value at 100-1.

Top 10

Tony Finau (40/1 odds). Finishes in Major Championships this season: the Masters (T10), US Open (Solo 5th), Open Championship (T9). At 40-1 he’d be the steal of this year’s Championship if not for Phil.

Kyle Stanley (80/1 odds). The Gig Harbor, Washington native has been playing sneaky good golf this year, including an impressive 2nd place finish last week at the star-studded WGC: Bridgestone. Can he ride the hot-hand into this week’s equally dense field and finish with another top 10?

The Field

The rest of the projections include some other surprising results like Tiger outside of the top 25 and Aaron Wise in it. Below is the entire list of all the golfers playing in the PGA Championship and their rank according to the engine. 

Projected Rank Player ODDS
1 Dustin Johnson 8
2 Justin Thomas 14
3 Jason Day 20
4 Jon Rahm 25
5 Phil Mickelson 100
6 Brooks Koepka 20
7 Tommy Fleetwood 28
8 Rory McIlroy 12
9 Tony Finau 40
10 Kyle Stanley 80
11 Rickie Fowler 22
12 Justin Rose 22
13 Patrick Reed 35
14 Zach Johnson 100
15 Patrick Cantlay 50
16 Webb Simpson 75
17 Marc Leishman 66
18 Bryson DeChambeau 80
19 Aaron Wise 125
20 Jordan Spieth 20
21 Bubba Watson 50
22 Cameron Smith 200
23 Paul Casey 50
24 Francesco Molinari 33
25 Matt Kuchar 80
26 Ian Poulter 100
27 Tiger Woods 28
28 Hideki Matsuyama 66
29 J.J. Spaun 250
30 Louis Oosthuizen 80
31 Luke List 125
32 Rafa Cabrera Bello 150
33 Kevin Na 200
34 Anirban Lahiri 200
35 Gary Woodland 100
36 Ollie Schniederjans 300
37 Henrik Stenson 50
38 Emiliano Grillo 200
39 Ross Fisher 250
40 Byeong Hun An 125
41 Pat Perez 250
42 Chesson Hadley 250
43 Brendan Steele 250
44 Kiradech Aphibarnrat 200
45 Kevin Kisner 100
46 Charl Schwartzel 150
47 Adam Scott 150
48 Xander Schauffele 50
49 Brian Harman 100
50 Daniel Berger 100
51 Kevin Chappell 125
52 Shane Lowry 150
53 Brandt Snedeker 150
54 Branden Grace 100
55 Patton Kizzire 300
56 Jason Kokrak 500
57 Ryan Moore 150
58 Billy Horschel 200
59 Jhonattan Vegas 200
60 Jamie Lovemark 200
61 Brian Gay 250
62 Charley Hoffman 100
63 Russell Henley 150
64 Jimmy Walker 200
65 Satoshi Kodaira 300
66 Russell Knox 150
67 Austin Cook 300
68 Peter Uihlein 300
69 Charles Howell III 250
70 Keegan Bradley 100
71 Andrew Landry 250
72 Chris Stroud 300
73 Sergio Garcia 100
74 Adam Hadwin 300
75 Bill Haas 500
76 Jim Furyk 250
77 Stewart Cink 200
78 Brice Garnett 300
79 Chris Kirk  
80 James Hahn 250
81 Chez Reavie 400
82 Michael Kim 250
83 Scott Piercy 300
84 J.B. Holmes 200
85 Troy Merritt 200
86 Beau Hossler 200
87 Padraig Harrington 250
88 Nick Watney 200
89 Scott Brown 300
90 Jason Dufner 200
91 Ryan Armour 250
92 Tyrrell Hatton 150
93 Paul Dunne 200
94 Matthew Fitzpatrick 150
95 Thomas Pieters 150
96 Ryan Fox 300
97 Julian Suri 200
98 Dylan Frittelli 200
99 Chris Wood 250
100 Seungsu Han 400
101 Alexander Björk 200
102 Joaquin Niemann 66
103 Brandon Stone 250
104 Ryuko Tokimatsu 400
105 Shubhankar Sharma 200
106 Jorge Campillo 250
107 Eddie Pepperell 200
108 Danny Willett 250
109 Sungjae Im 300
110 Alexander Levy 250
111 Andy Sullivan 400
112 Yusaku Miyazato 300
113 Matt Wallace 250
114 Yuta Ikeda 300
115 Martin Kaymer 300
116 Ted Potter Jr. 400
117 Andrew D. Putnam  
118 Meen-Whee Kim  
119 Jordan Smith 250
120 Thorbjorn Olesen 75
121 Hao-Tong Li  
122 Alexander Noren 50
123 Si-Woo Kim 150
124 Vijay Singh 300
125 Michael Block 750
126 John Daly 750
127 Davis Love III 500
128 Shaun Micheel 200
129 Omar Uresti 2000
130 Johan Kok  

Odds Courtesy of Bovada

How the Golf Engine makes its picks

In golf, a pro matches up as much with the golf course as another competitor. Which is why any attempt to predict the outcome of a golf tournament must account for the nuances of the course. Analyzing past and present data through the use of math can more accurately project future performance.

In this model, we use machine learning to evaluate 1,500 different statistics for every golfer on the PGA Tour over each tournament since 2004. The analysis of this massive dataset allows gives us an opportunity to predict players that are due to go low.

The machine learns how these statistics can become a unique strength or glaring weakness for each golfer by comparing tens of thousands of different combinations and separating the patterns from the noise. The resulting ‘model’ is able to ‘deep dive’ and determine when to expect low rounds from a pro, given their unique style of play. These calculations are next to impossible to do quickly and certainly without personal and subjective biases, until now.

This is the second tournament the golf engine has been used to predict on the site. Here are its predictions for the Open Championship. 

Contributor to National Football Post & sports nut with training in statistics, machine learning, and data analysis from Galvanize – Seattle campus. Alumni of University of Colorado and University of Washington. Occasional boater, skier, and golfer.

Read More 960 Words

Answering the Fantasy Football Players Injury Dilemma

All fantasy football players have had this dilemma:

Player X is my best option at WR 2 but he’s coming off of that groin injury and took limited reps in practice this week… should I start him or go with my backup who averages a couple less points?

The real question

All fantasy football players have had this dilemma:

Player X is my best option at WR 2 but he’s coming off of that groin injury and took limited reps in practice this week… should I start him or go with my backup who averages a couple less points?

The real question being asked is: what effect does being on the injury report have on average weekly performance when a player is still active for game day?

To analyze this question in an objective way, we need to look at the numbers. Specifically – we need to contrast the average or expected output of Player X when healthy (defined here as not-listed on injury report or full-participant in weekly practice) vs average or expected output of Player X when injured (defined here as listed on official injury report for the week and/or limited practice reps). But Player X is a pretty small sample size so lets aggregate the differences across the league by position and injury type. The goal is to get a better overall understanding of whether certain positions are affected more by injuries, or if certain types of injuries have a more noticeable impact on fantasy football performance.

For this information – I downloaded the full offensive stat list for each team from fantasydata.com and grabbed what I could for injury report data from nfl.com/injuries which, unfortunately, was limited to the 2017 season only. As you’ll see from the data, even an entire NFL season boils down to a pretty small sample size but there are still relevant insights to be gained.

The criteria which we used for the analysis is as follows:

Hurt: Injury listed on injury report but player active and took snaps in game

Healthy: Full practice participant with no injury specified, or not listed at all on injury report

Aggregated by player average points (standard league) when healthy vs player average points when hurt. (weeks that a player left game early and did not return the following week were removed so as not to distort the ‘healthy’ averages)

Since different players obviously average different hauls in points, we need to look at the average difference in points, isolated by player. We can then combine these across the league by position group and injury type to get a better look at the overall picture.

I’m working on creating an unbiased metric for opposing defenses so that we can more fairly drill down to how different injuries may affect specific players, however for now this will serve as a good overall indicator in the average change in production (points) between healthy vs hurt – as it would be highly improbable that all weeks when players ‘played hurt’ aligned with weaker defenses and/or all weeks when players ‘played healthy’ aligned with better ones (or vice versa).

You might have noticed that I refrained from using the word ‘drop’ or ‘decrease’ when referencing the average change week-to-week when healthy vs. hurt … and the reason for that is the available evidence is inconclusive that you should expect any change as long as said player is listed as active.

The overall distribution is pretty normal (both positive and negative) and most of the data are well within the standard margin of error week to week for each position type (about 4.18 points overall). Additionally, in most cases the higher frequency injuries (larger sample size) were closer to the healthy averages when aggregated – meaning there’s a correlation between less data and a greater deviation from the norm, and vice versa.

Avg Standard Deviation of weekly Fantasy Points by position: (Standard League, from this dataset)

QB  +-  6.49
RB  +-  4.64
TE   +-  3.08
WR +-  4.01

This further suggests that the more extreme deviations (in the data) may just be one-offs. In fact every data point outside of the 4-point margin had only 1 or 2 occurrences in 2017 to draw from and probably don’t qualify as hard evidence of any trends.

I think that may be  interesting in and of itself. This suggests that there shouldn’t be any expected change in production as long as a player is active. This also makes sense from the team’s standpoint, if you consider that by the time they allow a player to participate on gameday they expect his performance to be close-to-par. We also know that essentially every player is ‘hurting’ somewhere (from normal wear-and-tear) so ‘playing hurt’, as defined here, may be closer to the norm than most fans realize.

The net was about a 0.40 point overall reduction on average, but again with plenty of examples where players performed better than normal. It’s certainly not conclusive, however, and would be very interesting to see if this holds true over a 10 year period or more – or if the NFL has gotten better over time at diagnosing when a player is ready to come back.

Avg Change in Standard League Points (Healthy -> Inj) by Position, 2017 season

Position FantasyPoints SampleSize
RB -1.19 25
TE -0.65 18
WR -0.13 51
QB +2.74  4

Avg Change in Standard League Points (Healthy -> Inj) by Injury, 2017 season

Injury FantasyPoints SampleSize
Right Shoulder -4.66 1
Foot -4.23 1
Neck -3.97 3
Illness -3.08 2
Rib -2.92 2
Shoulder -2.80 8
Thumb -2.79 1
Groin -2.35 6
Knee -1.75  16
Toe
-1.30
3
Hamstring
-1.12
7
Abdomen
-0.95
1
Hip
-0.51
8
Calf
+0.17
5
Back
+0.43 1
Ankle
+1.15
15
Ribs
+2.54
5
Concussion
+2.71
8
Quadricep
+3.06
4
Hand
+4.82
2

*****

Stay tuned for the other side of this as we take a look at defensive players in a future post. One thing I’m curious of is the potential case that offenses attack a defensive player (or his side of the field) more often when they are known to be injured – which may lead to an increase in opportunities to make plays and accumulate points.

For reference, here is a link to the NFL’s official personnel injury report policy for the 2017 season.

Contributor to National Football Post & sports nut with training in statistics, machine learning, and data analysis from Galvanize – Seattle campus. Alumni of University of Colorado and University of Washington. Occasional boater, skier, and golfer.

Read More 911 Words

British Open Projections

The field for the 147th British Open is set at the historic Carnoustie Golf Links in Angus Scotland. We’ve modeled over 1500 statistics tracked by the PGA, for every tournament dating back to 2004 and how each stat contributes to what we can expect from players on this stage, at this tournament. It’s a complex

The field for the 147th British Open is set at the historic Carnoustie Golf Links in Angus Scotland. We’ve modeled over 1500 statistics tracked by the PGA, for every tournament dating back to 2004 and how each stat contributes to what we can expect from players on this stage, at this tournament. It’s a complex web of information that can only be properly analyzed by a machine, yet yields some objectively surprising results.

This year’s British Open is no exception as the model is calling for Webb Simpson (125/1 odds) to make a run into the Top Ten at least.

Some surprises:

Back-to-back US Open winner Brooks Koepka (22/1) inside the top 5.
Webb Simpson (125/1) and Phil Mickelson (66/1) inside the top 10.
Emiliano Grillo (100/1) inside the top 15.
Kevin Na (175/1), Luke List (125/1), and Ryan Moore (150/1) inside top the 25.

Perhaps just as surprising are golfers that may underperform this week. Rory McIlroy and Tommy Fleetwood don’t make the top 10 cutoff. Alex Noren, Francesco Molinari who finished T2 at TPC Deere Run last week, and Sergio Garcia are all projected outside of our top 25.

Notable left-outs:

Rory McIlroy (16/1) and Tommy Fleetwood (20/1) finishing outside the top 10.
Alex Noren (30/1), Francesco Molinari (33//1), and Sergio Garcia (28/1) all finishing outside the top 25.

A few more points of note:

Top 5:

It’s fascinating that Dustin Johnson gets the call for top ‘dawg from both the oddsmakers and the model. No question he is the best player in the world right now but it’s been a few years since DJ really contended (2011) in this tournament – and at a different course – Royal St George’s. He does have a pair of Top 10’s in 2012 and 2016 and has made the cut every year since 2009 (his first Open).

Justin Rose, not that his name doesn’t come up every year for this tournament – just that his style of play is generally considered to be different than the players on either side of him (Johnson and Koepka).

Speaking of Koepka, few are calling for him to win at Carnoustie though he does show up inside the top 5 here…

Jordan Spieth, although winning this tournament last year – has not put up the best numbers of his already memorable career the past few months. Frankly, I’m a little surprised the numbers bear this out…

Perhaps the third least surprising name to see on this list (aside from Johnson and Rose) is Rickie Fowler, aka Mr. Consistency, aka the Perennial Contender, aka always the Bridesmaid. Rickie almost always brings his A-game, and the data suggests it suits this course well. Curious to see if this is the year his major championship drought comes to an end.

Top 10:

Webb Simpson might be the most surprising pick on this list. Clearly, the model likes something about his game this year and the way he is set up for this tournament. A career-low 61 to open at The Greenbrier (his last start), T10 at the US Open a month ago and earning his 5th career victory at The PLAYERS were each separated by missed cuts.

Top 25:

For a guy with short odds, Rory McIlroy to be projected outside of the top 10, which really speaks to the consistency (or lack thereof) of his game this season.

Other notables:

No love from the model for Matt Kuchar, Scotsman Russel Knox, Adam Scott, Ian Poulter, or Louis Oosthuizen.

About the Author
Pat Ross – Contributor to National Football Post & sports nut with training in statistics, machine learning, and data analysis from Galvanize – Seattle campus. Alumni of University of Colorado and University of Washington. Occasional boater, skier, and golfer.

The Golf Engine Description
In golf, a pro matches up as much with the golf-course as another competitor. Which is why any attempt to predict the outcome of a golf tournament, must take into account the nuances of the course.

Beyond conjecture made by the golf pundits, analyzing past and present data through the use of math can more accurately project future performance.

In this model, we use machine learning to evaluate 1,500 different statistics for every golfer on the PGA tour over each tournament since 2004. The analysis of this massive dataset allows gives us an opportunity to predict players that are sitting on low round scores.

Projected Rank Player Odds
1 Dustin Johnson 12/1
2 Justin Rose 16/1
3 Brooks Koepka 22/1
4 Jordan Spieth 20/1
5 Rickie Fowler 16/1
6 Webb Simpson 125/1
7 Justin Thomas 22/1
8 Jason Day 33/1
9 Phil Mickelson 66/1
10 Jon Rahm 20/1
11 Henrik Stenson 28/1
12 Emiliano Grillo 100/1
13 Paul Casey 40/1
14 Patrick Reed 35/1
15 Rory McIlroy 16/1
16 Tommy Fleetwood 20/1
17 Bubba Watson 80/1
18 Tiger Woods 22/1
19 Kevin Na 175/1
20 Hideki Matsuyama 50/1
21 Bryson DeChambeau 125/1
22 Luke List 125/1
23 Ryan Moore 150/1
24 Tony Finau 100/1
25 Charles Howell III 500/1
26 Patrick Cantlay 100/1
27 Marc Leishman 45/1
28 Zach Johnson 100/1
29 Francesco Molinari 33/1
30 Brian Harman 150/1
31 Branden Grace 40/1
32 Pat Perez 250/1
33 Brandt Snedeker 150/1
34 Matt Kuchar 80/1
35 Chez Reavie 500/1
36 Jimmy Walker 250/1
37 Chesson Hadley 500/1
38 Rafa Cabrera Bello 125/1
39 Gary Woodland 250/1
40 Kevin Chappell 400/1
41 Xander Schauffele 150/1
42 Patton Kizzire 1000/1
43 Charley Hoffman 150/1
44 Brendan Steele 500/1
45 Byeong Hun An 200/1
46 Cameron Smith 200/1
47 Louis Oosthuizen 80/1
48 Ian Poulter 66/1
49 Charl Schwartzel 200/1
50 Adam Scott 125/1
51 Andrew Landry 500/1
52 Peter Uihlein 250/1
53 Kiradech Aphibarnrat 200/1
54 Tyrrell Hatton 40/1
55 Thomas Pieters 75/1
56 Bronson Burgoon 500/1
57 Satoshi Kodaira 1000/1
58 Kevin Kisner 400/1
59 Matt Jones 500/1
60 Retief Goosen 300/1
61 Russell Knox 66/1
62 Russell Henley 200/1
63 Jhonattan Vegas 750/1
64 Paul Dunne 150/1
65 Jason Dufner 250/1
66 Stewart Cink 250/1
67 Padraig Harrington 200/1
68 Anirban Lahiri 400/1
69 Jason Kokrak 1000/1
70 Austin Cook 500/1
71 Beau Hossler 250/1
72 Ernie Els 750/1
73 Keegan Bradley 300/1
74 Adam Hadwin 500/1
75 Daniel Berger 200/1
76 Abraham Ancer 1000/1
77 Matthew Fitzpatrick 60/1
78 Kyle Stanley 200/1
79 Dylan Frittelli 200/1
80 Jonas Blixt 500/1
81 Julian Suri 250/1
82 Kelly Kraft 1000/1
83 Michael Kim 300/1
84 Shane Lowry 150/1
85 Sergio Garcia 28/1
86 Ryan Armour 400/1
87 Shubhankar Sharma 500/1
88 Ryan Fox 125/1
89 Jazz Janewattananond 500/1
90 Alexander Levy 300/1
91 Matt Wallace 400/1
92 Ross Fisher 200/1
93 Matthew Southgate 125/1
94 Fabrizio Zanotti 1000/1
95 Danny Willett 150/1
96 Yuta Ikeda 1000/1
97 Martin Kaymer 200/1
98 Bernhard Langer 750/1
99 Cameron Davis 300/1
100 Lee Westwood 125/1

Data courtesy of Bovada.

Contributor to National Football Post & sports nut with training in statistics, machine learning, and data analysis from Galvanize – Seattle campus. Alumni of University of Colorado and University of Washington. Occasional boater, skier, and golfer.

Read More 856 Words