Skip to content

Latest commit

 

History

History
50 lines (45 loc) · 2.42 KB

changelog.md

File metadata and controls

50 lines (45 loc) · 2.42 KB
Raw

Version 0.0.0.9003

  • GP training now uses a dedicated genetic algorithm
  • Bug fixes for Cauchy covariance
  • Covariance matrix calculations implemented as proper internal methods (no more covfun slots)
  • Removed weights argument in GP and SPGP
  • Added data as default value for pseudo_inputs in SPGP
  • Constant and linear covariances now supported
  • Testing a more stable method for matrix inversions and Cholesky decomposition

Version 0.0.0.9002

  • Corrected a bug in SPGP object initialization the prevented the use of points3DDataFrame objects as pseudo-inputs
  • Corrected a bug that caused the GetContacts() method to convert the labels to factors
  • Corrected a bug in the Predict() method that caused an error for SPGP_geomod objects
  • Added a default for the midrange parameter in anisotropy3d() method
  • Corrected a bug in the Simulate() method for the standard GP
  • Added regularization to the smoothing covariance matrix for SPGP simulations
  • Corrected a bug in the Simulate() method for the SPGP, where the mean was not added back when making smoothed simulations
  • Corrected a bug in the Simulate() method for the SPGP, where the mean was added improperly during the simulation, generating inflated values
  • SPGP simulation now uses only the approximated variance as prior to avoid value inflation
  • Creation of the covarianceModel3D class to handle the covariances and the nugget effect on the same object
  • points3DDataFrame can now be initialized with no arguments
  • Corrected the computations of the log-likelihood
  • GP_geomod can now handle missing data labels
  • SPGP method Predict now outputs a value that measures the quality of the sparse approximation
  • GP_geomod now uses the same covariance model for all classes
  • Improved documentation

Version 0.0.0.9001

  • Formalization of structural data as a directions3DDataFrame object
  • Support for block models as blocks3DDataFrame objects
  • Inclusion of Sparse Gaussian Processes for regression and classification (implicit modeling)
  • Support for simulations
  • Removed support for covariance matrices in variogram form
  • Support of cross-validation for Sparse Gaussian Processes
  • Implementation of as.data.frame() method
  • Unification of covariance methods for all kinds of spatial objects
  • Added a parameter for nugget effect of structural data in the GP object
  • Calculation of the implicit model's probabilities is now done through sampling