enhancement - added support for bpcells, better handling of seurat v3, v4, v5 objects
performance - eliminated R-side dense matrix conversion and list creation overhead by passing sparse matrix pointers directly to Rust
bug fix - fixed probability extraction from Rust that caused query inference to fail
bug fix - fixed viewmastR_infer to apply same normalization as training (log-normalization) for consistent results
viewmastR v0.4
new feature - added new bulk deconvolution suite of tools, see vignette
viewmastR v0.3.1
bug - recursion bug seen only with Rust 1.90
viewmastR v0.3.0
bug - fixed a bug that failed to correctly perform ReLu on all NN
enhancement - upgrade to burn 0.17.1 and rust 2024 version
enhancement - store feature names alongside model.mpk in file meta.mpk
viewmastR v0.2.4
bug - incorrectly printed training accuracy in specific situations
bug - fixed an error in probability calculations
performance improvement - enabled train_only mode so that inference can be done on large datasets separately; embedded minor performance enhancements throughout - See BigQuery Vignette
viewmastR v0.2.3
new feature - naive bayes implementation
new feature - make3file function for easy exporting Seurat objects
performance improvement - improved sparse variance function efficiency
minor bug - fixed return type ‘matrix’ in training function
minor bug - fixed tf_idf function
enhancement - added debug function that enables tracking dimensions across learning steps
enhancement - made a number of enhancements in rust code