news.md

In zdebruine/RcppML: Rcpp Machine Learning Library

RcppML 0.5.0

Major changes

• Launch pkgdown site
• Added the nmf S3 class to the result of nmf function
• Introduce S3 methods for NMF ([, align, biplot, dim, dimnames, head, mse, predict, print, prod, sort, sparsity, summary, t)
• New plotting methods for NMF (biplot.nmf, plot.nmfSummary, plot.nmfCrossValidation)
• mse is now an S3 method for nmf objects
• project now handles only projections of w, for simplicity
• New vignette on Getting Started with NMF!

Minor changes

• Support for specific sample and feature selections for NMF removed to increase performance on C++ end
• Removed updateInPlace advanced parameter for nmf because advantages were not convincing
• mask_zeros implementation is now specific to sparse matrices, multi-thread parallelization, and projections with transposition
• Added cosine function for fast cosine distance calculations
• Condensed and pared down documentation throughout. Advanced usage discussion will be moved to future vignettes.

RcppML 0.5.1

Major changes

• three new datasets (hawaiibirds, aml, and movielens)
• Move NMF models and methods from S3 to S4 for stability
• Better random initializations (now using both rnorm and runif with multiple ranges/shapes, when multiple seeds are specified)
• added L2 regularization to NMF
• Support for masking values
• add impute and perturb methods to crossValidate

Minor changes

• better random initializations (now using both rnorm and runif with multiple ranges/shapes)
• New vignette on random restarts
• better "head" and "show" methods
• return "w_init" with model

RcppML 0.5.2

Major changes

• add linked NMF
• update all documentation

Minor changes

• clean up C++ API
• C++ API gets meta-templating

RcppML 1.0

Major changes:

• better cross-validation, now exclusively using the mean squared error of missing value imputation (random speckled patterns of missing values)
• complete migration to the S4 system, with backwards compatibility for CRAN version 0.5.0
• new vignettes and built-in datasets

Minor changes:

• compatibility with latest version of the Matrix package

zdebruine/RcppML documentation built on Sept. 13, 2023, 11:44 p.m.