|
Abstract:
|
Classification and regression problems with large numbers of candidate predictor variables occur in a variety of scientific fields, increasingly so with improvements in data collection technologies. Goals in model selection include accurate predictions, interpretable models, stability, and avoiding bias. Methods related to least angle regression (LAR) are among the most promising approaches to these issues. We describe our efforts to develop open-source Splus/R software for LAR, lasso and associated methods, in which we aim to establish a collaborative framework for future development and enhancement.
|
