## DiceKriging 1.5.4
## change / DiceKriging 1.5.3
* Implementation of a multistart option for the BFGS optimizer, allowing to perform several optimizations from different initial starting points.
* Bug corrected in 'drop.response', that prevented users from removing intercept in trend formulas.
* Bug corrected in method 'simulate': 'newdata=NULL' is now also possible when 'cond=TRUE'
* Bug corrected in method 'update' about power-exponential covariance function.
* Undesirable printing removed in kmEstimate
* Minor improvement in speed about initialization for the noise-free case, when an initial point is provided.
* test added about prediction: comparison with Ornstein-Uhlenbeck in 1D and Kriging with 'exp' covariance.
## change / DiceKriging 1.5.2
* Non-ASCII parameters removed in two help files
* Help file of scalingFun improved
## change / DiceKriging 1.5.1
* creation of a coef method
* trend.deltax: Default value for h is now fixed to sqrt(.Machine$double.eps) instead of 1e-8.
* Internal: The cases for optimisation now explicitely depend on the way likelihood is concentrated. Expert users: Mind that the 3 character strings for the slot 'case' have changed.
* Internal: Reorganisation of methods, according to covariance classes (One file per covariance, containing all the related methods)
## change / DiceKriging 1.5
* update method: It is now possible to reestimate the trend and/or the nugget effect when adding new points
* checkNames.R: correction of a bug in function checkNames (mistake in columns permutations) + addtion of a test function
## change / DiceKriging 1.4.0 & 1.4.1
* covMatrix: Now returns C and the vector vn, instead of the two matrices C0 and C.
* update.km : new function/method.
* Leave-One-Out: implementation of the Dubrule's formula, see 'leaveOneOut.km'.
* Estimation:
- in addition to MLE, cross-validation (leave-one-out) is now available when there is no nugget effect nor noisy observations.
- it is now possible to estimate the trend only (the covariance parameters being fixed)
* Prediction: The trend has been added to the output list.
* Implementation of the derivative of the trend f(x) with respect to x, see 'trend.deltax'.
* Minor: modification of 'computeAuxVariables' to include the case where 'model@trend.coef' has not been computed yet (then only T, M are updated, and z is set to numeric(0)).
## change / DiceKriging 1.3.3
* addition of the class "covUser", which allows the user to provide its
own covariance kernel (with known parameters).
* the user is now able to tune the optimization parameters of 'genoud' and
'BFGS' by the argument 'control' in function 'km'. Compatibility with the
previous 'control' option is ensured.
* modification of the 'predict' method:
- Creation of the argument 'bias.correct': Contrarily to
DiceKriging<=1.3.2, the estimated (UK) variance and covariances
are NOT multiplied by n/(n-p) by default.
- Creation of the argument 'light.return', that allows to save
memory (expert usage)
* merging of 'kmNoNugget', 'km1Nugget' and 'kmNuggets' in a single file,
named 'kmEstimate'
* the auxiliary functions 'covparam2vect' and 'vect2covparam' are now defined as methods.
## change / DiceKriging 1.3.2
* the checkNames option is now TRUE by default, and produces a warning instead of an error.
In the future, it will produce an error.
## change / DiceKriging 1.3.
* extension of methods "simulate.km" and "plot.km" to noisy observations.
* addition of a consistency check between the input names in the design and in
newdata. It can be used for prediction and simulation by turning the argument
"checkNames" to TRUE. The checking stage is done in the new function "checkNames".
* creation of the function "kmData", which allows the user to give the response and
the design in a single "data" argument (as for lm).
* modifications in the documentation of km (examples), predict.km and simulate.km
(section warning + examples).
## change / DiceKriging 1.2
* addition of the class "covScaling", corresponding to the general case of scaling
(varying number of knots per dimension). Implementation of scalingFun and scalingGrad
in C