localreg
localreg copied to clipboard
sigvaldm
Multivariate Local Polynomial Regression and Radial Basis Function Regression
It would be useful to localreg to return the standard deviation or preferably the standard error of the fit. My own antique hand-coded LOWESS 1-D routine returns the standard error...
One of the original papers about the LOESS/LOWESS method added an iterative method for giving less weight to outliers.
Basis functions with local support has two problems: 1) you get the "bed-of-nails" when the distance between them is much larger than their radius, and 2) they have to decay...
Today the centers are obtained by K-means clustering. This leaves no basis functions on the edges of the training data domain, which either lead to large radius (if automatically tuned)...
One could have a function or class before using e.g. multivariate local regression to change the axes of the data using PCA, and then scale them to have unit standard...
Today `polyfit()` doesn't return fitting coefficient like NumPy's does, but NumPy's is only 1D. Now that localreg is multivariate, it would be nice to make it more similar to NumPy's...
Local regression and RBF are examples of kernel methods. Other kernel methods such as kernel density estimation (KDE) would fit nicely in this package.
localreg fails to perform calculations that call `np.product()` because that method has been renamed to `np.prod()`. `np.product()` was deprecated many versions ago and has been removed in recent versions. Change...