CRAN v0.2.10.0

Updates:

• fixes bug that led to non-reproducible results in certain situations (#20)

CRAN v0.2.9.0

Updates:

• compatibility with Matrix 1.4-2
• change to minimal version of R to 3.6; this is related to compatibility with Matrix and continuous integration tests
• adjustments to internal code style

CRAN v0.2.8.0

Updates:

• support for umap-learn 0.5 (#14)
• rudimentary support for sparse matrices (#15). With method="naive", sparse matrices are converted to non-sparse matrices; with method="umap-learn", sparse matrices are passed as-is to the python implementation.
• adjustments to package description to address R CMD check

CRAN v0.2.7.0

Updates since v0.2.5.0:

• support for umap-learn 0.4
• more visible support for precomputed knn (this was possible since early versions, but is now simplified with object constructors and with documentation)
• extended vignette with new sections, some vignette edits

(This version has been on CRAN since Nov 4, this is a late github release)

CRAN v0.2.5.0

Updates:

• fixes a bug that prevented spectral initialization of layouts (RSpectra was not loading correctly)
• small improvements in loading times

CRAN v0.2.3

Updates:

• improves stability of output from predict() when performed in batch and one item at a time. This improves consistency and usability, at a slight performance cost. Thanks to @JenniferSLyon for raising this issue.
• shift more of the layout optimization procedure to Rcpp
• adjustments to vignette to explain when to expect stable results
• slight edits to tests

CRAN v0.2.2

Updates:

• fixes bug interfacing with umap-learn. Thanks to @lgaborini for raising this in an issue.

CRAN v0.2.1

Updates:

• fixes for some documentation typos
• changes in initial layout for multi-component graphs

CRAN v0.2.0

New features:

• function predict() projects new data onto a layout defined by an object created via umap(). Works for method="naive" and method="umap-learn"

Updates:

• performance imprevements for default implementation of umap with method="naive"
• interface to python implementation moves from method="python" to method="umap-learn"
• interface to python implementation is consistent with umap-learn versions 0.2.x and 0.3.0

first release on CRAN

Includes "naive" implementation (from scratch) and a "python" implementation (calling python package umap-learn via reticulate).