- More memory management improvements to the C++ pipeline to speed up processing. Thread-local buffers, fewer heap allocations.
- Patched a bug wherein residual scoring for AF candidate selection would inflate massively due to improperly normalized AF spectra.
- Safe solve to prevent crashes if singular values obtained.
- Reproducible random number generation for
fix.my.unmix().
- Hand-rolled Huber slope robust linear modelling to support adaptive unmixing.
- Faster FCS reading with reduced memory usage thanks to tips from Paul Heisig.
- Faster unmixing with reduced memory usage thanks to tips from Paul Heisig.
- Set "-O3" flag for C++ compiler to speed up processing.
- Fixed issue causing divergence in some events when running
unmix.poisson.fast(). Briefly, the weights could collapse to infinity, and are now clamped to the approximate noise floor. Additional issues with tracking convergence and fallback are fixed.
- Fix issues causing C++ crash when reading or writing large FCS files
- Fast kernel density estimation for gating and plotting in AutoSpectral.
- Parallelized C++ assignment and unmixing of per-cell autofluorescence
- Faster (~10x) per-cell optimization using residual-alignment pre-screening.
- Import functions from AutoSpectral rather than duplicating.
- Added fast Poisson–IRLS unmixing with incremental updates.
- Added OpenMP support with optimized C++ kernels.
- Implemented new SSM calculation pipeline.
- Faster Poisson–IRLS unmixing with fast QR decomposition
- Better handling of convergence with step halving, deviance monitoring
- Allow early exit if convergence reached
- unmix.wls updated to match AutoSpectral, ensuring non-negative weighting and a more numerically stable solve.
- Hopefully faster compiler flags.
- Initial estimates for IRLS are no longer clamped to non-negative values
- Indentation error in optimize_unmix_rcpp_woodbury