Resolve -Wsometimes-uninitialized, -Wformat-zero-length and -Wformat-security compiler warnings.
Resolve "lost braces" issue in a documentation page.
Much faster init_grid() (and likely more minor gains in other functions too).
New init_grid(., maxmin) argument.
New dependency 'matrixStats'.
Checks of invalid input (and resulting screen output) adapted to non-Migraine usage.
'lbfgsb3' is replaced by 'lbfgsb3c' as a suggested package.
Requests a more recent (>= 3.1.0) version of 'spaMM'.
Long-standing glitch that substantially slowed blackbox down (though with correct results).
(Migraine-specific) New composite parameters and profile computations for migration parameters.
Dependency on 'ff' package removed, assuming a recent version of spaMM (here >= 2.4.8).
Dependencies changed for parallel computations. See help("blackbox.options") for details.
(Migraine-specific) writeFinalInfo() now correctly writes the results when sampleByResp() has not been run.
(Migraine-specific) CKrigcoefs() could fail when minSmoothness=maxSmoothness.
(Migraine-specific) user-given 'CovFnParam' were ignored.
(Migraine-specific) now computes Dgmu parameter only when Dg/2N is a parameter of the model (e.g. not a for a discrete change in pop size).
(Migraine-specific) resetCHull() handles input points in linear space of lower dimension than the number of fitted parameters, now avoiding a bug in the rare case where there are many profile points, all of them in such a space.
No user-level changes.
Changes for compatibility wth Rcpp::compileAttributes() in Rcpp >= 0.12.12
Other internal changes for compatibility with planned changes in other packages.
(Migraine-specific) Parallel computation is implemented for profile point computations.
Less primitive convergence diagnostics are implemented in bboptim().
(Migraine-specific) New mesures of ”goodness of prediction” implemented.
(Migraine-specific) add logscale for 1D profile plots
blackbox.options(<new values>) did not return old values.
(Migraine-specific) Return code modified for consistency with documentation (but this does not really affect the analyses).
(Migraine-specific) variables names could be lost in generateNewPoints().
(Migraine-specific) Invalid requests by users for CIs for some composite variables could lead to an execution error.
Added native routine registration.
More C++ portability issues fixed.
(Migraine-specific, without any impact on inferences) optimization convergence diagnostic (actually nloptr()'s return status) was missing from the output_...txt file, resulting in lines with different number of values.
(Migraine-specific) toCanonical() corrected to handle single-candidate input matrix.
(Migraine-specific) fixed parameter space issues in generateNewPoints() and canonize().
Modified init values of some smoothing parameters may change bboptim() results for rough response surfaces.
Some function definitions has been moved from the spaMM package to blackbox.
(Migraine-specific) Wrong variable spaces in generateNewPoints() and toCanonical() corrected.
(Migraine-specific) Calls to Rf_error() removed from Rcpp code.
(Migraine-specific) An invalid cbind() that affected the generation of 'nextpoints' has been corrected.
(Migraine-specific) Within subHullWrapper(), the scdd() results with double precision are now checked and replaced by a computation with rational arithmetic if they are inaccurate.
(Migraine-specific) "samplingSpace=,Nb," has been broken for a long time. This is corrected.
(Migraine-specific) sampleNextPars() could stop when called with non-default 'samplingSpace'. This is corrected.