largeVis
would fail if saveNeighbors was set to FALSE.edgematrix
objects. This allows the as_dist function to be an S3 method for as.dist applied to class edgematrix.projectKNNs
projectKNNs
now supports training with momentum, which offers an additional 20-50% performance improvement. See the vignette for details. projectKNNs
useDegree
parameter allows the user to select the negative weighting method. largeVis
now has an additional parameter, save_edges
, to control whether the edge matrix is preserved. This is to simplify using the
clustering functions. hdbscan
, lv_dbscan
and lv_optics
now all accept a largeVis
object as the first parameter.hdbscan
. hdbscanToDendrogram
function to make hdbscan objects compatible with other R hierarchical clustering implementations.sgdBatches
helps estimate training time for datasets. sgd_batches
where 10x to many batches would be used for dataset < 10000 nodes. buildEdgeMatrix
and distance
now store the distance_method
in attribute method
of the returned object. useDegree
parameter, and clustering. Fix to hdbscan selecting wrong K, and gplot failing.
Hotfix for a bug in the neighbor search when max iterations was 0.
The OPTICS implementation has been temporarily removed. This reason is that the code was based on the code in the dbscan
package, and the CRAN
administrators objected to the inclusion of a separate copyright notice. OPTICS will be restored once the code is sufficiently re-written.
buildEdgeMatrix()
and buildWijMatrix()
, which are simpler. Rewrote neighbor-search code to improve memory performance.
Performance of wji calculation improved by an order of magnitude.
Many changes because of OpenMP compatibility issues.
Moved from Rcpp to RcppArmadillo objects, etc.
Initial development releases. Focused on correctness, performance, testing against larger datasets.
Added a NEWS.md
file to track changes to the package.
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.