Title: Practical Least Angle Regression Authors: Tim Hesterberg, Insightful Corporation Chris Fraley, Insightful Corporation Abstract: Least Angle Regression (Efron et al., 2004, Annals) is a possible major advance in regression and classification modeling with a large number of potential explanatory variables, producing more stable and accurate predictions than methods such as stepwise regression. The idea unifies and provides a fast implementation for a number of modern regression techniques. There are R and S-PLUS versions available from http://www-stat.stanford.edu/~hastie/Papers/LARS. We describe work to generalize that basic software by improving the numerical accuracy, supporting explanatory variables with multiple degrees of freedom (factor variables and spline bases) and interactions, and providing nonlinear regression models. Our ultimate goal is to produce robust and easy-to-use software for a wide audience.