c : comment
* : bug fixed
+ : new feature/function
x : changed functionality
--------------------------------------------------------------------
Version 2.5, September 4, 2002
* REGPOLY2 did not work properly for dimension greater than 2
Bug fixed.
--------------------------------------------------------------------
Version 2.4, September 3, 2002
x DACEFIT gives an error return in case of a rank deficient matrix F.
May arise with a poor combination of choice of regression model
and distribution of design sites.
--------------------------------------------------------------------
Version 2.3, August 30, 2002
x DACEFIT gives an error return in case of zero columns in matrix F.
May arise with a poor combination of choice of regression model
and distribution of design sites.
--------------------------------------------------------------------
Version 2.2, August 27, 2002
x DACEFIT allows design data with standard deviation zero
in one or more directions.
--------------------------------------------------------------------
Version 2.1, August 26, 2002
* PREDICTOR returned erroneous mse-values in case of single
response, multiple entry. This bug is fixed.
--------------------------------------------------------------------
Version 2.0, August 5, 2002
x The element dmodel.regress is renamed to dmodel.regr
in DACEFIT, lines 19, 106
in PREDICTOR, lines 58, 90, 107
c Documentation is adapted to the new version. A separate report
documenting the algorithm and other special parts of the toolbox
is available.
--------------------------------------------------------------------
Version 1.3, July 3, 2002
* A developer debug statement in PREDICTOR line 52:
save NX x
is removed.
+ DSMERGE is a new function for merging data sets with multiple
design sites (i.e., sites sampled more than one time).
See 'help dsmerge' for more information.
+ PREDICTOR now return Jacobian matrices for model and mean
squared error also for cases with multiple response functions.
c We would like to acknowledge the comments and suggestions from
Thierry Dalon, from Siemens VDO Automotive AG, which is a great
help in improving this software.
--------------------------------------------------------------------
Version 1.2, June 25, 2002
+ DACEFIT now has a second output parameter
[dmodel, perf] = dacefit(S, Y, regr, corr, theta0)
or
[dmodel, perf] = dacefit(S, Y, regr, corr, theta0, lob, upb)
The second output parameter gives the convergence history of the
optimization - see 'help dacefit' for more information.
+ DACEFIT now allows the second input argument to be an m*q matrix,
i.e., allows q response functions to be modelled, the elements of
dmodel: beta, gamma, sigma2, Ysc are changed accordingly.
c The optimization method in DACEFIT is changed. A separate report
documenting the algorithm and other special parts of the toolbox
will be available in August 2002.
+ PREDICTOR is adapted to the changes in DACEFIT, though in case
of multiple response functions it does not yet return the
Jacobian matrices for the predicted models and their respective
mean squared errors.
+ CORRCUBIC is the new name of the function named CORRSPLINE
in version 1.1. See 'help corrcubic' for more information.
x CORRSPLINE gives a correlation function based on a cubic
spline. See 'help corrspline' for more information.
The above mentioned report will also discuss this new
correlation model.
x The parameter vector q in GRIDSAMP is changed so that now it
is the number of equidistant points in each coordinate direction.
This is in accordance with the similar parameter in LHSAMP
and LINSPACE.
--------------------------------------------------------------------
Version 1.1, June 7, 2002
* A sign error in the gradient calculation in CORRGAUSS is
corrected: line 33 in the file is changed from
dr = repmat(-2*theta(:).',m,1) .* d .* repmat(r,1,n);
to
dr = -2*repmat(-theta(:).',m,1) .* d .* repmat(r,1,n);
--------------------------------------------------------------------
Version 1.0, April 16, 2002
c First version released