Aug 14 2009

My Talk at the 2009 Southeastern Ranking and Clustering Workshop

My attempts to record the talk failed due to what appears to be a broken Flip Mino. (Note to self: don’t upgrade the software on a working product the night before you plan on using it.)

All is not lost, as you can download the slides for my talk:

In addition, you can download the code and data used in the creation of the presentation by downloading:

  • rw09.zip: View the README file for information on where everything is and what it does

Since some parts of the talk are better explained, I’d be happy to answer any questions you have. So just leave a comment if you have a question!

If you enjoyed this post, use RSS to get notified of new posts.

8 Comments on this post

Trackbacks


  1. edkupfer said:

    Page 9, drawing of a halfcourt (http://i26.tinypic.com/2ihmii0.png): where did you get that? Did you draw it in R? I ask cuz I have code to draw something remarkably similar, but without all those hash marks.

    Also, i was a little confused by the terminology, especially “rating”, which I assume has some kind of generalisable definition. Can you expand on that?

    Finally, can you point me to a decent resource on Bradley-Terry models, classical and fancy schmancy? An R-centric resource would be ideal.

    Also finally, is there a way to enable a “preview comment” feature?

    August 14th, 2009 at 10:16 pm
  2. Ryan said:

    That’s actually an image I created for obtaining shot locations while tracking data, so it’s not exactly R friendly. I do, however, have some R code somewhere from the guy that created these http://www.stat.columbia.edu/~cook/movabletype/archives/2008/10/phoenix_suns_sh.html if you’re interested in that.

    The best way to define a rating is that it is simply a quantitative value we assign to something to determine the ranking of players or teams or some other items of interest. In my case, I’m most interested in ratings that we can use to make predictions, like the odds of an event, or magnitude of a win, etc. Not all ratings are setup for that sort of thing.

    A quick search found this result titled “Bradley-Terry Models in R”: http://www.jstatsoft.org/v12/i01/paper I haven’t seen this before, and it looks like a good reference. My first introduction to this was from Agresti’s Intro to Categorical Data Analysis.

    Good idea on the preview comments. I myself like that sort of thing, but since I’m an admin I can easily edit my mistakes. Everyone else, however, isn’t so lucky. 🙂

    August 14th, 2009 at 11:24 pm
  3. Ryan said:

    I added a basic preview. Let me know how it works. Not the prettiest thing ever, but it’s functional.

    August 14th, 2009 at 11:47 pm
  4. edkupfer said:

    I am interested in whatever R code you have for generating court “drawings” in R. I found my old clunky code, which just draws this. (Hey, I can use html in the comments! That’s why I needed to preview the comments, to see what kinds of formatting thingys were allowed.)

    I’m most interested in ratings that we can use to make predictions, like the odds of an event

    See, that’s where my confusion enters. To me, a rating is simply a dimensionless value with no reference to the real world, and the only thing it is good for is to rank order stuff. I can’t wrap my head around the idea that a rating can refer to odds. I am being confused by mere terminology!

    And it turns out I have that Agresti text, so thanks for the reference. And thanks for the preview look.

    August 14th, 2009 at 11:58 pm
  5. Ryan said:

    I’ll try to boil down the court drawing code and post it here tomorrow.

    In a general sense, ratings are sort of abstract in nature that we can use for ordering, so I can see where the difficulty can come in here. Not all methods care about translating these ratings in terms of predicting stuff.

    That said, when we derive these ratings in specific model formulations, like the ones in the presentation, we can then use these ratings to estimate the odds a player makes a free throw, 3pt shot, etc. For example, on slide 5 we show that a player’s log-odds of making a free throw is betai, and we can call betai our ratings. Slide 6 shows the distribution where we expect these ratings to come from.

    So with these ratings, we just use inverse logit to estimate the probability of a player making a free throw: Pr(Make) = ebetai / [1 + ebetai]

    August 15th, 2009 at 12:16 am
  6. Ryan said:

    So that court stuff isn’t as straight forward as I’d hoped. Some comp issues I’m having will make this take a little longer.

    August 15th, 2009 at 12:58 pm
  7. edkupfer said:

    Don’t go bananas, it’s just that I thought it would be neat to have that code in the public domain. It’s not that important.

    August 15th, 2009 at 4:04 pm
  8. Ryan said:

    Yeah I agree. I thought it’d be a quick snap to do, but unfortunately that’s not the case. It will come soon enough.

    August 15th, 2009 at 4:07 pm