The Jeopardy! Tournament of Champions begins today. Here’s a quick roundup of some Jeopardy!-related links I’ve come across recently:
From Gawker, How a geek cracked the Jeopardy! code. The “geek” in question is Roger Craig, who built a tool to help him study for the show; it seems to have helped, as he won six games, and set the one-day record of $77,000.
Slate had a roundup of some of the most common categories which is pretty interesting. This came out around the time of Watson.
In fact, IBM’s Journal of Research and Development had a whole issue on Watson.
And there’s Eric Feder’s model for the probability of winning at a game of Jeopardy! (based on the state within a single game) and a forum thread discussing it.
The data from the Jeopardy archive has been scraped as well if you want to do some of your own analysis.
The Final Wager analyzes the game theory of Jeopardy! wagers.
What I’m curious about right now are the odds of winning today, given that you won yesterday – and more generally, building a model of player strength, in order to predict things like the odds of winning a tournament. In this year’s tournament there are two that stand out above the rest: Julia Collins (who won 20 games) and Arthur Chu (who won 11). They’re very different players – here’s a profile in the Wall Street Journal. SB Nation wants them to have an Ultimate Jeopardy Battle of Fire and Lasers. Can I get odds on that battle happening, or on its winner? It seems quite difficult to do this accurately, though – most ranking problems depend on the fact that players will play each other repeatedly, but the Jeopardy! format, in which once you’re gone, you never come back (except for possibly in a tournament) make the analysis more uncertain.
(Bayes should help.)
God plays dice 