Armed with the knowledge that Our Greyness is not infallible (Grey’s ‘Pertise), I was determined to prove that the computer could, well, out-predict him. After all, the man makes split-second calls on “please put these in order,” “who is best for the rest of the year?”, “in a twelve-team, keeper, with SLG and OBP in place of AVG and HR, place in order …,” and, simply, “of these pitchers who should I start/sit tomorrow?” Surely, given all of the data available and the crunching ability of The Whopper, the machine could out-perform him on, say, at least the “who should start/sit tomorrow issue.” Without a doubt. It would just take time and patience. That I had.
Step one: Go to The Master himself. “Grey, what factors do you consider in deciding who to start/sit?”
Answer: “Last few starts, match up history, ballpark… Opposing pitcher doesn’t matter usually.”
Sounds good. Now I know he has a pretty good memory for all of that stuff, but he can’t possibly be as good as having every bit of detail categorized and calculated. I embarked on, what would turn out to be, roughly 40 hours of data gathering and model building.
The goal was to predict where a pitcher’s performance from his next start would place him in a 12X12 RCL league. This would give us a relative number to use in deciding who to start/sit.
Key factors were:
1. The last two starts (home or away, depending on the upcoming start).
2. Overall Home, or Overall Away, for the 2009 season.
3. Heads up (regardless of home or away) for 2009.
4. Historical match ups: specifically home, or away, for 08/07.
5. Ballpark: per team (not park) from 2009 data. I.e. If a pitcher’s last two starts were against light-hitting teams SF and SD, the computer would adjust the K/9, ERA, and WHIP from those games to account for the next start, say, a heavy-hitting Texas. It would project a comparatively less expected pitcher-performance.
The test set of data for model-building was from all RCL pitchers for the week of July 16-20, immediately following the All-Star break.
Finally, the model was adjusted to “weight” factors according to their value in predicting the results. In this, the computer got to “cheat.” It actually knew the outcomes and was solving to meet those. Grey went in blind.
When the model completed its work, it had projected K/9, ERA, and WHIP, 12X12 rankings for each, as well as an overall projected RCL rank for the projected performance.
All that was left was to prove that Grey’s split-second calls couldn’t come close to matching the 40 hours of work and number-crunching ability of the computer. Trust me, the computer and I are darn good at number crunching.
This was going to be the equivalent of: “With a dramatic victory in Game 6, Deep Blue won its six-game rematch with (chess) Champion Garry Kasparov.”
“Grey hits the Disgraceful List when the Fumethrower falls to Deep Fred.”
I culled twenty (repeats were not counted) start/sit questions that were put to Grey for the test week. Here are the results: (If one side came out ahead, it got the count.)
Grey: 7*
The Model: 4
Matching: 9
(*This includes 3 for newer pitchers where the model did not have enough info to project; advantage Grey)
The model is good. When/if Grey isn’t available, I will use it.
Now, some of you may be making similar calculations, but I would wager none as much as the computer crunched. Still, Grey was clearly superior. Obviously, there are other factors that Grey weighs subjectively when he makes a call. It might be possible to compile an even more sophisticated model, but there would be no guarantee, the work would be overwhelming considering the benefit, and, who needs it—we can just post the question, and count on the answer. I know he won’t nail every one, but I am also confident that there is no living human being on the face of the planet that can match him.
I would love to see a competition amongst leading pundits, all given the same set of questions, and then measured against results.
Folks, all of this energy and calculating was for just the simple, “who to start/sit.” It doesn’t even begin to tackle the larger issues that Grey deals with.
“Grey Topples Deep Fred”
