Bill Barnwell is up to his usual tricks at Grantland. This time, he’s tired of hearing that Flacco is an elite quarterback and wants a new measure of quarterback value. Flacco gets credit for piling up wins, which Barnwell thinks is unfair:

For whatever good or bad Flacco provides, he has spent his entire career as the starting quarterback of the Baltimore Ravens, who perennially possess one of the league’s best defenses. He also has Ray Rice and a solid running game to go alongside him on offense. It’s safe to say that a win by, say, Cam Newton usually requires more work from the quarterback than one by Flacco.

I agree with this wholeheartedly. In response, Barnwell tries to capture quarterback value by creating an “expected wins” measure based on points allowed by the defense and comparing this to actual wins. He argues that a quarterback with more…

Edit: Please see my later post as well, which corrects an omission here.

Miguel Cabrera has a shot at the Triple Crown this year. No one has done it since Carl Yastrzemski. Is it really possible that he could win the Triple Crown and not win the MVP? Well, yes. Every advanced stats guy out there is trumpeting Mike Trout for MVP, with his “wins above replacement” (WAR) above 10 (next best in the majors is 6.8) and his 13 “total zone total fielding runs above average” (basically, this is the number of runs he has saved with his fielding, compared to an average fielder).

The discussion is eerily similar to the AL Cy Young conversation in 2010. Felix Hernandez won because he led the AL in innings pitched, ERA, and, most importantly, WAR,  even though his win-loss record was a mediocre 13-12.

The 2010 Cy Young was a victory…

If you haven’t yet discovered the competitive machine learning site kaggle.com, please do so now. I’ll wait.

Great – so, you checked it out, fell in love and have made it back. I recently downloaded the data for the getting started competition. It consists of 42000 labelled images (28×28) of hand written digits 0-9. The competition is a straight forward supervised learning problem of OCR (Optical Character Recognition). There are two sample R scripts on the site to get you started. They implement the k-nearest neighbours and Random Forest algorithms.

I wanted to get  started by visualizing all of the training data by rendering some sort of an average of each character. Visualizing the data is a great first step to developing a model. Here’s how I did it:

Which gives you:
Notice the wobbly looking ‘1’. You can see that there is some variance in the angle of…