Incorrect computation of descriptive statistics for GWR about tinfour HOT 3 CLOSED

I am currently revamping the SurfaceGWR class to address the issues described above. I have confirmed that the parameter estimates are correct, but that estimates for standard deviation, confidence and prediction intervals, and AICc tests -- all the things that tell you about the quality of the regression -- are just plain wrong. Expect to see an update in a couple of weeks. I will fixing correctness of implementation issues, making SurfaceGWR more useful for use in stand-alone applications (separate from the main body of Tinfour), and adding more documentation about the appropriate use of the class (and when you should just go with standard tools like ArcGIS, R or GWR4).

from tinfour.

I extensively revised the SurfaceGWR class. I have verified the regression calculation as well as the standard deviation, AICc, and other descriptive statistics with the well-known GWR4 program.

Unfortunately, the automatic bandwidth selection method is much slower than it used to be. Apparently, getting the right answer takes more time. I am working on ways to improve this.

I am also looking at adding a cross-validation algorithm for automatic bandwidth selection.

I am not quite ready to close out this issue since I want to do more testing and clearing up some code. I also need to add more documentation to the Javadoc and code comments for future developers.

from tinfour.

Issue closed.

from tinfour.