from Amazon
Tuesday, February 20, 2007
More Pitch-by-Pitch
Schmanzy-fancy acronym aside, this is the way to look at it. And, if you include the count, this is exactly how I look at it, which alllows you to generate a Markov for each pitcher and hitter. That is, rather than relying on the 2000 historical PA for a pitcher or batter, you rely on their 7500 pitches to create a much stronger profile as to exactly how they hit and pitch. You remove tons of luck this way. You can further add parameters to your Markov model by also considering base/out and inning/score, as well as two-pitch or three-pitch sequence state, rather than the current count. And if you want to get even more creative, you include the profile, or identity, of the opponent.
The only thing missing from the article was the actual data itself. All we see was one example.
"The only thing missing from the article was the actual data itself. All we see was one example.”
Thanks for the link Tango. We are working right now on getting the data incorporated into our stats suite on the website.