search

sigvaldm / localreg

Multivariate Local Polynomial Regression and Radial Basis Function Regression
GNU Lesser General Public License v3.0
46 stars 4 forks source link

LOESS/LOWESS procedure for enhanced robustness #4

Open sigvaldm opened 3 years ago

sigvaldm commented 3 years ago

One of the original papers about the LOESS/LOWESS method added an iterative method for giving less weight to outliers.

danoffenbacker commented 2 years ago

Seconding this post. Would truly be enhanced with the addition of an iterative method.

sigvaldm commented 2 years ago

Thanks for your input :) Unfortunately, I don't think I have time to implement it in the near future. When I do find time for localreg, though, I now know this is a requested feature.

tbpassin commented 1 year ago

I implemented this once, a long time ago, in a TurboPascal program. It would calculate the LOWESS smoothed curve. Then when a function key was pressed, it would de-weight all data points in the smoothing window outside of the 2-sigma standard error limits and re-calculate. It had the nice feature that showed the disqualified points with a different symbol (open instead of closed squares) so it was easy to see what the routine had done. You could keep repeating the procedure, but that usually wasn't very helpful because eventually you would usually eliminate most of the data points.