Pheonix Open Picks, look back at the Farmers
It’s Superbowl week – we got the Superbowl 16th hole par 3 coming up at the Pheonix (check this out for the best golf-watching / NFL / college football parody ever). Sports are amazing. I’d love to rant, post awesome videos and detailed analysis about this weeks contests… but nobody should care about my opinion. The only thing they should care about is what my computer says is going to happen this weekend, and wether or not these figures can make them any cash.
Last week I built my golf squads purely on these projections and cashed in two of the three sites big-field tournies I was in, as well as winning the majority of my head to head matchups. I actually had two teams cash in DraftKings amazing golf kickoff event – which was an awesome start to what I hope is a long exciting fantasy golf season on that site.
And this performance was, by all estimations, only an average week as far as the projections are concerned. What really helped was dodging a few bullets – and while I had Tiger almost everywhere that wasn’t DraftStreet (and so did everyone else), and while he missed the secondary cut, I didn’t have Phil who submarined many fantasy rosters. I also had no part of Snedeker who had a sub-par projection considering his pedigree and course history. In the world of daily fantasy sports, dodging two land mines might be all it takes to give you an edge over the field.
Add this to the rather unbelievable projections I made at the end of last week’s post:
“A top 10 for K.J.? Pretty bold… Lee Westwood AND Bubba favored over Phil? Seems crazy…”
Turns out all three of these machine-made predictions came true. The last two probably only due to Phil dropping out – but hey, you take a win wherever you can get it.
Let’s see exactly how the algorithm did last week, keeping in mind that the analysis is only performed for those golfers whose data was pulled for this coming week.
Look back at The Farmers Fantasy Projection Performance
The same caveats as always apply to this analysis. Perhaps mid-way through the season – when almost all of th early round data has been captured because almost all the golfers have played in a at least two tournaments – I will post a more comprehensive look-back at performance. For now, this is all I have time for – and the proof is in the pudding. I’m winning – so retroactive analysis is less important then pressing ahead with the projections.
As usual the comparison picks are from Notorious at Rotogrinders.com, a very well respected name in the daily fantasy sports industry. If you don’t know what these graphs are showing (and you care) then just go to first fantasy golf post that’s hyperlinked above – it contains all the explanations of the relevant concepts used in this analysis.
Below is a grid showing Notorious’ draft street points projections (DSP), the converted mean field draft street points projections (%MFDSP), the machine’s %MFDSP and the actual %MFDSP’s:
Last week was interesting in that there was something inherently difficult about the tournament to predict. Maybe it was the two very different courses? Maybe the poa grass that caused the greens to be so bumpy literally made the outcome more variable and subject to chance? Whatever the reason – both the machine and Notorious had a harder time predicting last week accurately than any week so far this season. Still, as has been the case each week so far, the machine had a slightly lower RMSE – and blending the two models produced an optimal RMSE that was lower than either method taken alone, but this time the machine’s projections were damn-near optimal:
Here is the output from the blending procedure to find the optimal blending alpha between the two:
The farther left you are on this line, the more of the machine’s projections you are using, the farther right here represents Notorious. So the optimal blending parameter for Farmers turned out to be about 17% Notorious and 83% Machine. I may bump the split I use personally up from 10% to 15% of Notorious’ projections for this coming week.
Lastly – here’s the cut percentage for both of us:
This week wasn’t really that great for either of us. In fact, both one of my teams that cashed on DraftKings had two people miss the cut (including Tiger), and the other team had three! I think this cut% number meant less last week: the US Open like conditions caused so many bogies, which can cause a player’s round to be worth less than zero, that there was a much lower premium on having players that simply made the cut than there will be in this coming week’s tournament where the scouring is typically much lower. Last week you needed to have the winner, and to cut your losses quickly on any golfer who was going to stink up over the weekend. This week you need six golfers that make the cut – if you do that you’ll probably make money.
Like Humana, the Farmers posed interesting computational issues – there is a two-course rotation with North Torrey Pines being played once, and the 4-5 stroke more difficult South Course being played three times over the course of the tournament – so course-to-course or tournament-to-tournament comparables are less, well, comparable since it’s hard to properly capture this nuance in the tournament structure. The projections did seem to suffer a bit, but my intuition is that this drop off in performance reflects two rather unrelated things:
1) low-fantasy-scoring tournaments have a higher standard deviation in their percent mean field scoring as a mathematical artifact. Winning the tournament represents a much higher portion of the fantasy points a golfer earns than does playing skilled golf. It’s not that these two things aren’t correlated to one another – but only one golfer wins the tournament, while at Humana you had plenty of the field scoring 15 under par or better. To the extent that these outcomes can be predicted, the algorithm takes its cues from things that affect real-world golfer performance (course history, recent history, driving length / accuracy, etc.) – but in a ‘winner-take-all’ fantasy-point structure, the ‘above-average’ golfer who doesn’t win, could turn in a below average day for his fantasy owners.
2) Weird stuff went down. Phil bowed out due to back pain. Tiger bowed out due to unexpected crappy golf. Snedeker – who the algorithm projected low compared to what Notorious and other serious fantasy golfers were discussing, but who was still projected in the top 25 – absolutely stunk it up, not even coming close to making the cut. It wasn’t that the machine was off, or that Notorious was off – it was that this year Farmers really was quite unpredictable. It happens all the time when you’re trying to predict uncertain outcomes – best to recognize it and (cautiously) move on without too much to-do. The method is sound – it’s the world that’s crazy.
In this context I’m very excited about this coming week’s contest. There’s no whacky pro-am thing happening. There isn’t a third-round cut, or some good-course bad-course act messing up the balance of the universe. It’s just four days, on the same course, with one hole that has stadium seating and a crowd of 20,000 strong, many of them alcohol-fueled fraternity brothers with BUBBA spelled out in big blue letters on their shirtless beer bellies, singing the ‘Olay! Olay Olay Olay!’ as Hideki Matsuyama tees it up on the 16th hole and wonders what alternative universe he’s been transported to. Outside of that – everything’s normal.
Picks for the Pheonix Open
Here they are:
Keep in mind that Phil is not going to be in the field as he recovers from the back issue which forced his withdrawal last week. That bums me out because predicting a Hunter Mahan win with Phil in the field is much more impressive without him. Still – Mahan hasn’t been on his A-game, so when I saw his number come out I had two thoughts – one was that most fantasy golfers will be off of Mahan due to his recent lackluster performance, and two was that the algorithm must think it knows quite a bit to make such a bold statement. It’s the kind of prediction that’s really fun to make. Mahan beating Simpson and Bradley just doesn’t carry the same punch.
At any rate – outside of Mahan as a favorite I think it’s interesting to note how low a guy like Graham Delaet (Idaho local, mad props), Bubba Watson and Brandt Snedeker have come in – all outside the top 20. And again – Choi now with a freaking top 5?!? It’s possible I’ve built in an Asian bias; I do appreciate their calm humility and dedication to all things mathematical.
It’s going to be a blast – good luck everyone in your contests. Talk to you next week.