Skip to content

the libgp file format

Manuel Blum edited this page May 21, 2014 · 1 revision

The libgp file format

Gaussian process models and the training data can be easily written to text files using the //write// command. The syntax of the model files is very simple.

  • Lines beginning with # are interpreted as comments and are ignored.
  • The first non-comment and non-empty line contains the input dimensionality.
  • The second line describes the covariance function. Composite covariance functions are specified by their name, followed by two covariance functions (composite or simple) enclosed in parenthesis and separated by comma.
  • The next line contains the log-hyperparameters of the covariance function.
  • The remaining lines are filled with the training data. The first number is the target value, the following values represent the input vector.

Example

# Tue May 22 15:22:15 2012

# input dimensionality
2

# covariance function
CovSum(CovSEiso, CovNoise)

# log-hyperparameter
0 0 -2 

# data (target value in first column)
-0.2418600751 -0.414140905 1.361941478 
-0.3589076369 -0.7252289108 1.545713733 
-0.01130130786 -1.362525494 -0.4651365008 
-0.2589561064 1.599417225 -1.345816197 
1.031256984 0.4165767588 0.3307960848 
-0.4366041211 -0.8263977195 0.9695096241 
-0.2039803933 -0.3800694657 1.429511771 
-2.104040652 1.38590312 -1.988979674 
1.423510208 1.151506484 -0.9375510601 
1.036225178 0.403420546 0.4348626134 
-0.2472420233 -0.781371593 -1.392560544 
0.6467016903 0.5744393156 1.014213103 
0.3509130214 -0.1625587347 0.6099537879 
0.1124851074 -0.5278405303 1.775561357 
-1.398084471 -0.908916189 -1.902803377 
-1.460878427 1.5107734 -1.762525226 

Clone this wiki locally