My DraftKings Project - Thought I would share

[golfnhack]

 

I have been working on this for a while, and thought I would share what I have done. There has been a tremendous amount of work that has gone into this. Since I work a day job too, it has taken a while. This was a labor of love project. If you play DraftKings golf, you know how incredibly difficult it is to come up with the winning lineup.

 

You have to:

 

1. Pick the winner
2. Have all players made the cut
3. Have at least 3 players finish in the top 5
4. Have at least 5 players finish in the top 15
5. Have your worst player finish in the top 20 (more than likely)

 

Some fields have over 150 players and the cut is 65 players. Then there is the fact that your lineup salary cannot exceed $50,000, which means you cannot just stack your lineup with all the best players. Then, throw into the equation that you can have a player like Sepp Straka miss straight 6 cuts in a row, be ranked DFL in the rankings for the Fed Ex St Jude... and lose in a playoff. So again... predicting lineups for DraftKings golf is HARD!!!!!

 

My system does a crap ton of data mining and analysis. Naturally, I am not going to share my formulas, but they are pretty advanced. I am a software engineer by trade with over 25 years of data analysis experience. The backend is in Python. The Database is MYSQL.

 

Attached is a spreadsheet that shows the rankings for all the playoff events, as well as the Wyndham Championship. I have the ability to generate a rankings sheet for every single DraftKings tournament that I have data for (which is around 200 going back to 2018).

 

I will give a brief explanation of all the columns

 

• Name
• Salary
• Recent Form Pts - Ranking points based on the player's finish position in recent tournaments.
• Trending Pts - Ranking points based on how the user has trended in recent tournaments
• Scoring Pts - Rankings points based on number of Eagles or Better, Birdies, Pars, Bogeys, Doubles or worse in recent tournaments and DraftKings scoring points for each.
• Stat Pts - Ranking points based on a player's rankings in each stat for stats that are considered significant to doing well on the course the tournament is being played on.
• Bonus Pts - Total bonus points from 3 categories: 1) Past results on the same course, 2) Starting Position (FedEx Tour Champs only), 3) Momentum carried over from last week's tournament.
• Total Pts - The total ranking points
• FPTS Value - The FPTS value for money factor, i.e. the average fpts / contest salary
• Ranking Value - The total ranking points value for money factor, i.e. the total ranking points / contest salary
• Combined Value - (FPTS Value + Ranking Value) / 2
• Average FPTS - The average FPTS earned in recent tournaments
• Risk Factors - The risk factors to consider. 2 Categories: 1) Number of recent DNF's. 2) Long period since last PGA start.
• Recent Form - A breakdown of finish positions in recent tournaments, ordered by most recent
• Trend - A visual showing  how the player is trending. The more +'s, the better they are trending. The more -'s, the worse they are trending.
• Bonuses - A breakdown of the bonuses awarded.
• Egl+,Brd,Par,Bog,Dbl- - A breakdown of the scoring stats in recent tournaments
• Rnd 1 Tee Time - Scheduled Round 1 tee time (in case of bad weather)
• Rnd 2 Tee Time - Scheduled Round 2 tee time (in case of bad weather)
• Significant stats [0-6] - The rankings in stats considered significant to playing well on the course the tournament is being played on. Anywhere from 0 to 6 stats.

The points values are normalized between 0-100. The value factors are normalized between 0-10. As mentioned, there is a lot of complex logic behind the scenes used to calculate everything that I won't go into.

 

Below is a screen shot of the rankings for The Tour Championship. Attached is an XLS of the rankings for the last 4 PGA tour events played.

 

 

I also have the ability to generate some cool heat maps. E.g. Below is a heat map showing the salary breakdown percentages per slot of the of the top 50 possible lineups from the last 4 years of The Tour Championships. PS: What I found interesting here is that building a lineup using mid-range salary players is pointless. A WHOPPING 90%!!!!! of winning lineups contained the lowest 2 salaried players. Remember that for next year. 😉

 

 

As mentioned, this is a labor of love project. I started it because I went 3 months of playing every week and not winning a cent. I am pretty knowledgeable about golf and know and follow the PGA tour and its players. So it was very frustrating. Since I have been using the rankings sheet, I have won money every single week. I only play 3 lineups. Generally 1 x $15/$20 entry and 2 x $5 entries. Given the number of possible combinations of lineups (hundreds of millions), the rankings sheets have helped a lot. The goal is to get to a point where I am comfortable raising $500 and playing 100 entries. I have the ability to generate lineups and simulate how they would perform against historical DraftKings results. There is still a lot of work to do, but it has come a long way. Everything has been written to be dynamic and tunable, so it is a matter of finding that magic set of tuning values that will result in optimal results.

 

Hope you liked this. If there is enough interest, I will try to do periodic updates of my development progress.

WRX.xlsx

[Cali_Stinger77]

Super cool!

[ChristopherMcDonald]

As a fellow data nerd, this is come good stuff! Have done similar things for my tournament poker play, logging past performances, etc. Best of luck with your wagers!

[PNWGuy]

By all means keep us updated!

[golfnhack]

Haven't been able to work on this much in the evening/weekends lately because we just finaled our game (video game programmer)... but hoping to get back to it this weekend. Have made some changes since the original post.

 

Salary heat table has been changed to use an aggregate of percentages for each tournament rather than a sum of all instances of the salary. This makes it so that if in 1 tournament a salary appears hundreds of time, it does not skew the results heavily in that salary's favor.

 

I also tweaked the significant stats code to now add strokes gained stats. However, if a sg stat is chosen as a significant stat together with a related stat, it ignores the related stat (and vice-versa). This will prevent stat skewing.

 

Eg

 

if order of significant stats is

 

sg_putting   

gir_pct

putting_5_to_10_feet

 

significant stats become

 

sg_putting

gir_pct

 

if order of significant stats is

 

putting_5_to_10_feet

gir_pct

sg_putting   

 

significant stats become

 

putting_5_to_10_feet

gir_pct

 

What is going to make first tournament of the year tough is the fact that there will be many rookies. I am contemplating mining KFT stats and results, and weighting the accordingly. However, not sure if this will be worth it since it only really impacts the first 6 tournaments. VERY rarely does a rook win one of the first 6. I actually can't remember when the last time was that this happened. In any case.. will work on that this weekend (assuming work does not take precedence)

Edited September 9, 2022 by golfnhack

[AC168]

This is very interesting!  Any updates?