APBRmetrics

I'm going to this (since I'm living in the Bay Area now), thought I'd post the announcement.

http://analytics.theiegroup.com/sports

Kirk Lacob giving a talk:

The Golden State Warriors Silicon Valley Startup: A New Era
Perspective on Technology & Data Analysis

Ben Alamar (OKC):

Treating Analytics as Innovation - Getting Good Tools Used

Kirk Goldsberry:

Visual Analytics for the NBA

Also looking forward to a football talk from Keith Goldner who does some cool Markov models:

Situational Analysis in the NFL & Fantasy Analytics

I'd be interested in hearing your report on the NBA topics after the fact.

Crow wrote:I'd be interested in hearing your report on the NBA topics after the fact.

By Twitter, of course.

If you want to post 300 tweets, that is fine. I'll read that.

By the way, when does your new metric get unveiled? Any blending involved?

I hope when you met one or both of the Lacobs they indicate they have read your stuff.

Crow wrote:
By the way, when does your new metric get unveiled? Any blending involved?

I'll tell you what I did, but I want to work on it more before I post anything. The weights for A4PM were originally based on a regression on RAPM. I wanted to see what would happen if added interaction terms to the regression. So, for example, OREB*ORR*OFTR, etc. The rankings that currently are in the lead on my site ("C" - 89% of the vote currently) are the ones done with those interaction terms included. "A" is RAPM (mine, not Jerry's). "B" is A4PM without the interaction terms.

Before I write more about it, though, I'd like to do more validation. Specifically, I want to calculate the predictive errors of each system. Adding all those interaction terms might give a better fit to the data, but may result in overfitting and worse prediction. (And I would be the first to admit that people may not like the idea of adding those terms, because the model becomes more of a black box.)

Hoping to get something done by the end of September, though.

Thanks for the explanation.

Would you have any interest in blending those 3 as another mark? From a technical perspective what would be the pluses and minuses of doing so? Could machine learning or some other technique be used to fine-tune the mix? Is the analogy of a multi-element lens applicable and useful or not?

Awesome, an analytics conference in SF. But damn is that registration price steep. Evan, any chance you want to meet in person after the conference?

For sure, send me a pm and some contact info. You live in the Bay Area?

Evan, sent you a PM! I'll be lurking around on Thursday.

The non-twitter update.
http://www.goldenstateofmind.com/2012/9 ... ion-summit
Would have taken 51 tweets.

Doesn't sound like it was that thrilling.