from Amazon
Wednesday, February 04, 2009
Neural networks for the Hall of Fame
Phil breaks it down for us.
Let’s see… the data used by the black box looks at Wetteland’s 2.93 ERA and Nen’s 2.98 ERA, Wetteland’s 330 saves to Nen’s 314 saves, Wetteland’s 48 wins to Nen’s 45, .516 win% to .517 win%, both had 3 All-Star games, and Wetteland’s 127 win Shares to Nen’s 105, looks at no other piece of data, and concludes that Wetteland has a 51% chance of being a HoF and Nen at 32%? And Tom Henke, comparable to both, but with a much lower ERA at 19%?
That is one highly sensitive black box. Indeed, we have no idea if the ERA is always considered a plus, that perhaps some odd combination of low ERA, high wins and low saves is enough to look like a negative. We really don’t know.
What I would like instead is for the system to be fed a set of controlled data, so that we can see the impact of HoF probability. This way, we can say “assuming you have 50 wins, 300 saves, 100 win shares, .500 win% and 3 All-star appearances, then having an ERA of 3.50 gives you X probability, 3.00 gives you Y, and 2.50 gives you Z”.
Clearly, there are ridiculous results that no SME (subject matter expert) would make. I love the basic idea behind the process, but if you are going to present a black box like this, at least show us the impact of each parameter, to try to explain why Robin Yount is not a HoF.
Recent comments
Older comments
Page 1 of 344 pages 1 2 3 > Last »Complete Archive – By Category
Complete Archive – By Date