[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Test data for (m-code) semiovariogram?

From: Jose
Subject: Re: Test data for (m-code) semiovariogram?
Date: Mon, 7 Oct 2013 21:03:32 +0300
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:24.0) Gecko/20100101 Thunderbird/24.0


I am also playing with spatial statistics.

On 06/22/2013 12:32 AM, fork wrote:

It yields a similar but not exact empirical semiovariogram to the SAS example
linked by UCLA, at least by looking at the graph (see below the coal data).

Not sure how to interpret it with my real data (a home-rolled migration index
by county), but I am working on it.

Short points from your code
- I could not find the "distmat" in my octave installation (3.6.4), neither with a quick search on Internet. - Although it does not have significant importance, note that the data that you use in your test is not exactly the one from the UCLA web (check line 114).

I have also created my own script to calculate the semi-variogram.

The main motivation was to learn, and to vectorize some parts to make it faster (I have large datasets).

Regarding the testing, I have the same problems than you, I am not completely sure about how to test it properly. I have added some simple tests, as you can see in the code (If you run the tests, a file with data will be downloaded automatically to your working directory and deleted after).

In particular I was also trying to reproduce as exactly as possible the figure from
Note that it seems to be using a robust method (which is not specified in the page), not the classical one by Matheron. So in my search for a perfect match I also implemented some robust methods that I found in the literature, which still fail to reproduce the figure exactly. However, the figure from
is well reproduced by my and your code (I used my own version of distmat).

If somebody has reference datasets (and results) with which I could test Cressie, N. and Hawkins, Genton and Dowd's robust methods, I would appreciate some links.


reply via email to

[Prev in Thread] Current Thread [Next in Thread]