NEWS

serrsBayes 0.6-1

Fixed residual resampling in fitSpectraSMC: parents were not kept in place, so the in-place (OpenMP-parallel) copy could overwrite particles that were still being copied from. This caused biased resampling and, with multiple threads, a data race (CRAN check failure on r-devel-linux-x86_64-fedora-gcc)
resampleParticles now copies the full spectral signature of each particle, not only the first npeaks wavenumbers
residualResampling no longer modifies the caller's log-weights
Added unit tests for residualResampling and resampleParticles

Added SHLIB_OPENMP_CXXFLAGS to Makevars for parallel computation on supported platforms (e.g. Apple clang 21 and libomp 17.0.6)
mhUpdateVoigt no longer copies the spline matrices for each particle, using pass-by-reference instead.

Merged PR #4 from Dirk Eddelbuettel for compatibility with Eigen 5.0
Replaced deprecated Matrix coercions as(., "dgCMatrix") with coercions via virtual classes
Fixed NOTE for using old-style personList() or as.personList() in CITATION
Added citation to Moores et al. (2025 MATRIX Annals Part II) instead of arXiv preprint
Also cited Gracie et al. (Anal. Chem. 2016) and its associated SERS dataset (Gracie 2015)
Replaced Travis with GitHub Actions for CI and codecov for code coverage
Enabled Rhub 2 GitHub Action
Added new unit tests for SMC functions

Zero MCMC iterations will be performed if the number of unique particles is already >= minPart (adaptive number of MCMC steps)
Replaced the default uniform prior distribution for the peak amplitudes with an exponential prior

Merged pull request from Rcpp team to use constant M_PI instead of PI to support STRICT_R_HEADERS
Fixed issue with plots in Introduction vignette
Fixed broken link to EPSRC grants website (now uses UKRI Gateway to Research)
Added package Hmisc to suggests to fix Undeclared package in Rd xrefs NOTE on CRAN
Try-catch when computing Cholesky decomposition of the Gaussian random walk variance-covariance matrix (fallback to diagonal matrix by default)
Removed Travis CI from README, since builds have ceased on June 15, 2021

MCMC iterations will stop when temp ESS reaches the total number of particles (adaptive number of MCMC steps)
Added DOI from Zenodo to CITATION file.

Fixed issue with digits = 0 in vignettes (CRAN WARN)
Removed obsolete call to Rcpp::LdFlags() from Makevars.win
Added fallback to boost MCMC acceptance rate if it falls below 15% of proposals
Regenerated the datasets result.rda & result2.rda to adjust for the new way of calculating FWHM

Added example of fitSpectraSMC with informative priors to the methanol vignette.

Removed the R package hyperSpec from Suggests because it is no longer available on CRAN.
Fixed dependency on hyperSpec in Introduction vignette due to new CRAN check. This also affects the TAMRA dataset.
Fixed a rare bug in reweighting that could potentially result in an infinite loop.
Peak locations are now constrained to lie within the range of observed wavelengths.
Locations are sorted to preserve identifiability.
Fixed error in equations for FWHM in mixedVoigt & getVoigtParam as well as in both vignettes.

OpenMP implementation of mhUpdateVoigt provides around 3x improvement in elapsed runtime
New vignette illustrates when to use the 3 functions fitSpectraMCMC, fitSpectraSMC, and fitVoigtPeaksSMC
New datasets methanol and TAMRA for use in the vignettes
Debug output prints the means of the amplitudes of the peaks at each SMC iteration

Fixed bug in fitVoigtPeaksSMC where it was using the upper-triangular instead of lower-triangular Cholesky factorisation to generate random-walk Metropolis proposals
Introduction vignette no longer attempts to download a .zip file over HTTPS
Deleted unused Rcpp function randomWalkVoigt