Dear R users,
I’m please to announce the available on CRAN of new package largeVis.(*)
largeVis offers three major features:
- A fast implementation of the LargeVis algorithm. LargeVis is for
visualizing high-dimensional datasets, similar to (and of similar quality to)
t-SNE. But, LargeVis runs in O(n) time, which makes it feasible to use on
datasets with millions of rows and thousands of columns. LargeVis is also
insensitive to hyperparameter changes, which is important when running on large
datasets that take time to compute.
- Very fast approximate nearest neighbor search. I believe it to be the
fastest nearest neighbor search available for R.
- A fast implementation of the HDBSCAN clustering algorithm. HDBSCAN is
a density-based clustering similar to DBSCAN and OPTICS (which are also
implemented), but HDBSCAN allows the density threshold for clusters to vary.
This makes it insensitive to hyperparameter changes and more flexible than
either DBSCAN or OPTICS.
There are other features as well, such as functions to visualize image
embeddings using largeVis.
Some examples are available here: https://github.com/elbamos/largevis
Benchmarks comparing the speed of the nearest neighbor search to RcppAnnoy are
here: https://github.com/elbamos/largeVis/blob/master/benchmarks.md
Examples of HDBSCAN are here:
https://cran.r-project.org/web/packages/largeVis/vignettes/momentumandusedata.html
The package is available here:
https://cran.r-project.org/web/packages/largeVis/index.html and for best
results, to take advantage of 64-bit machines and multiple cores, should be
installed from source.
Thank you!
(*) A prior version was available on CRAN but not announced.
[[alternative HTML version deleted]]
_______________________________________________
R-packages mailing list
r-packa...@r-project.org
https://stat.ethz.ch/mailman/listinfo/r-packages
______________________________________________
R-help@r-project.org mailing list -- To UNSUBSCRIBE and more, see
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
