posted 2009-10-19T21:04:38Z (apropos 1990-02-28).
when fitting a model, often one can trade off bias and variance in order to get better predictions. "shrinkage" is a classic phenomenon where one finds that biasing a fitted model coefficient towards zero actually improves the model's performance. stephen m. stigler gives a good introduction in his 1988 neyman memorial lecture. stigler.pdf.
---> back to algomagic.com