bigPLSR 0.8.0

Fixed RKHS score prediction to center cross-kernels with the fitted training statistics. A row’s latent score is now invariant to the other rows included in the same prediction batch, including when using the streamed backend.
Added an optional filematrix row-block provider for future streaming backends. It exposes sequential row/column/block reads as dense R matrices without changing existing bigmemory code paths.
Added experimental backend = "filematrix" support for NIPALS PLS1/PLS2. The backend uses block-wise file reads and avoids memory-mapped bigmemory access for the predictor/response inputs. Other PLS algorithms remain outside the filematrix backend scope.

bigPLSR 0.7.3

Big-memory NIPALS PLS1 now uses the same row/chunk streaming backend as PLS2, avoiding the legacy XtX materialization path for very wide predictors.

New tuning option: options(bigPLSR.stream.block_align = 8192L). All streamed backends (bigmem SIMPLS, streamed scores, RKHS/klogitpls Gram passes, and bigmem predict) round their chunk_size up to a multiple of this alignment, then clamp to the available number of rows. Typical sweet spots are 4096–16384 on modern CPUs.
If you always need scores on disk, prefer scores = "big" to avoid large R dense allocations; it streams directly into a big.matrix.
Added benchmarks results and analysis as two vignettes.

Added plot_pls_bootstrap_scores() and group-aware ellipses for plot_pls_biplot() to visualise latent structures.
Exposed bigPLSR_stream_kstats() for streamed RKHS centering statistics and corrected the bigmemory RKHS interface to accept dense response blocks.

Stabilised kernel logistic PLS class weighting, reinstated IRLS fallbacks and improved dense/big-memory parity.
Reworked the Kalman-filter state helper to reuse the SIMPLS backend, ensuring identical coefficients/intercepts to batch fits.
Added dedicated RKHS/RKHS-XY and plotting vignettes, and refreshed the PLS1/PLS2 benchmarking guides with notes on the new algorithms and parallel helpers.

Added optional future-powered parallel execution to pls_cross_validate() and pls_bootstrap().
Extended pls_bootstrap() with (X, Y) and (X, T) strategies, percentile and BCa confidence intervals, numerical summaries, and coefficient boxplots.
Added group-aware score plotting with confidence ellipses in plot_pls_individuals().
Added vignettes covering cross-validation/information-criteria workflows and bootstrap diagnostics.

kernelpls on backend=‘bigmem’ now uses streaming XXᵗ/column paths; the previous dense fallback was removed. Control with options(bigPLSR.kpls_gram = ‘rows’|‘cols’|‘auto’) and bigPLSR.chunk_rows, bigPLSR.chunk_cols.

Algorithm auto-selection: new internal heuristic chooses among
- XtX SIMPLS (standard cross-product SIMPLS),
- XXt (“widekernelpls”) for n << p,
- NIPALS when memory is tight or rank is low. Tuned by options(bigPLSR.mem_budget_gb = 8). Users can override with algorithm=.
Kernel-style PLS routes: algorithm = "kernelpls" and algorithm = "widekernelpls" implementing Dayal & MacGregor–style (1997) kernel PLS in X-space and wide-X (XXᵗ) space.
Implemented high-performance kernel and wide-kernel PLS algorithms in pls_fit() for both dense and bigmemory backends using RcppArmadillo.
Introduced optional coefficient thresholding.
Added fast-running examples to all exported functions to improve documentation usability on CRAN.

Added kernel PLS and wide-kernel PLS algorithms to pls_fit() for both dense and bigmemory backends.
Refreshed plotting helpers with variable plots, arrow-based loadings and a dedicated VIP bar plot.
Introduced convenience prediction wrappers, information-criteria helpers, and expanded cross-validation/bootstrapping utilities to support the new algorithms.
Improved summaries with explained-variance reporting and updated package documentation.

Added unified path pls_fit() for plsR regression that features : dense and bigmemory, simpls and nipals.