by Didier Renard » Mon Sep 16, 2013 7:19 pm
Dear Nicolas
This is the answer to your question.
If you run a first try of model.auto using the argument verbose=T (as you did) but without bothering about specifying bounds, you will immediately see that the number of parameters to be fitted are 9:
1) sill of the nugget effect
2 & 3) range and sill of the gaussian
4 & 5) range and sill of the cubic
6 & 7) range and sill of the exponential
8 & 9) range and sill of the spherical
Now let's come to the upper or lower arguments. The module is expecting a COMPLETE vector of bounds. If the dimension of the lower or upper vector is not equal to 9, the input arguments are ignored (without message).
Instead, using an input vector as follows:
upper = c(1,NA,NA,NA,NA,NA,NA,NA,NA)
will produce the following printout (with verbose option ON):
Model
-----
Structure : Nugget Effect
1 - Sill : 0.860091 [0.000000,1.000000]
Structure : Gaussian
2 - Sill : 0.339909 [0.000000,NA[
3 - Range U : 0.001070 [0.000010,NA[
Structure : Cubic
4 - Sill : -0.000000 [0.000000,NA[
5 - Range U : 4.526838 [0.000010,NA[
Structure : Exponential
6 - Sill : 0.000000 [0.000000,NA[
7 - Range U : 6.048644 [0.000010,NA[
Structure : Spherical
8 - Sill : 2.400000 [0.000000,NA[
9 - Range U : 5.200000 [0.000010,NA[
We clearly see that the upper bound of the sill of the nugget effect has been taken into account correctly.