May 2003 This directory contains the code for the normal likelihood nonstationary GP model (with a stationary GP model as a special case). The code is currently implemented only for 1-3 covariates. All files necessary for compiling the code are here. To compile a binary Unix/linux executable, just run 'make' To run, see the file example.csh init{1,2,3}d.dat are example initial value files for 1,2, and 3 dimensional covariates for the approach of using the same correlation structure for all the eigenprocesses. Make sure to rename your initfile as init.dat. parm{1,2,3}d.dat are example parameter files. Again, rename as parm.dat. The prior values in the parm files should be reasonable provided your function values aren't extremely large (one easy solution is to standardize your observations to have mean 0 and sd of 1). You will need to adjust the proposal standard deviations to achieve good acceptance rates (should be around .44 for scalar parameters and .23 for vectors). The initial values are meant to be reasonable for function values that have been standardized. Note that the covariates are always scaled to lie in [0,1] so the priors and initial values for the eigenprocesses should be reasonable as given in these files.