forum.alglib.net

Dear Dr. Borchkanov,

I have a question about Shepar modified method in Alglib.
I have a really very slow dll function and I'm working on creating a nx = 5 dimension data set in order to interpolate data with Shepard method.

I'm trying to understand if it si better to create data on a regular grid or to create random points in the domain (using the grid and adding random number in the process in order to explore all the domain).

1) In the case of the regular grid, is it always better to use ShepardR specifying a r radius which is consistent with all directions of the grid (this can be done normalizing the different dimensions)? Or can I use the modified Shepard (without R!) if the grid is very "dense"?

2) In the case of random distributed points, I should use modified Shepard (without R!), is this right?

Could you please help me?

Best regards
Luca Cecchinato

PS
In the meantime, I created two vb.net sub which write and read an *.xml file in order to save/load the interpolant object once it is created. This should be faster than building the interpolant object each time I want to use the model.