Hi Mike,
Thanks for pushing gstat to it's limits. The trouble is that the
number of point pairs you have for a semivariogram estimate
exceeds the maximum an unsigned long integer can take in C on a
32 bits machine, which is 4294967295: you can see that the
third column of output increases to a value lower, and then
changes to negative values. The obvious solution is: use gstat
on a 64 bits machine. If you don't have one, I can think of
two solutions:
1. build in a check that stops adding point pairs when this
maximum is reached. I guess that this would slow down compu-
tations pretty seriously, just think about how often we should
do this check
2. don't use _all_ your data. Usually, the variogram doesn't
change much when instead of 5x10^5 you use 5x10^4 data or so.
Use a sampling plan that is representative (in some sense
random) and that has enough information on short distances.
The DINFINITY things you noted while compiling on the hp are
in the meschach matrix package -- I didn't write it, but will
have a look into it.
Mike Sukop wrote:
> Sorry if this is addressed in the Computers and Geosciences article. Our
> library cancelled its subscription in 1996. If you have a reprint, I would
> love to have a copy.
Please send me your mail address.
Good luck with all the super computing,
--
Edzer
