Remove error on filtered data point names not matching; if names are
not given, then mappeR will assume the output of the filter
is in the same order as the original data points.
The generic create_mapper_object function should
correctly handle non 1D lens functions.
Fixed a bug where sometimes empty bins could get through to the
clusterers and mess them up.
mappeR 2.3.0
Add more input sanitization for distance matrices
Add more cluster metrics to output, namely medoid name, maximum
distance to medoid, and sum of squared distances to medoid
Examples now use actual noisy circle data; better examples coming
soon
mappeR 2.2.1
Relax some tests as to avert floating point shenanigans
mappeR 2.2.0
Force user to have the names of the filtered data match the names of
the original data set. You may need to call
names(filtered_data) = row.names(data) or similar before
making a call to generate a mapper object. The Ball Mapper flavored
versions should still work fine.
Add the option to specify a global cut height when using the global
hierarchical clusterer.
Update existing documentation and remove .Rd pages written about
unexported functions (documentation still exists in source).
mappeR 2.1.0
Add better hierarchical clusterers; there is one which cuts each
dendrogram uniformly, and one which cuts per patch. The cut height is at
the beginning of the longest unbroken branch in the dendrogram (note
this is not always the “correct” number of clusters!).
Remove igraph dependency; see shinymappeR
for an example of how to use igraph with
mappeR.
Change output vertex dataframe column names from
tightness to mean_dist_to_medoid and
bin to patch.
Change edge weight to use Jaccard index.
mappeR 2.0.2
Really, for real this time, make the default clusterer cut
correctly.
mappeR 2.0.1
Default hierarchical clusterer now correctly cuts dendrograms at the
midpoint of their tallest branches.
mappeR 2.0.0
Clustering is now handled by a clusterer, which is a
function that can handle a list of distance matrices (one for each
bin/level set) and output clustering results for each one. The
hierarchical clustering included previously is now available as a
clusterer called hierarchical_clusterer because I am very
creative.
User-defined clusterers are possible as long as they can handle
inputs/outputs correctly. The basic idea is that the output of a
clusterer should look like a list of calls to
cutree from the hclust package. Look to the
clusterer farm for more examples in the future.
mappeR 1.3.0
Adjust compute_tightness to no longer normalize by the
maximum distance from the medoid (easier to see behavior in a single
mapper graph, may add options in future)
Add option to consider each level set locally when clustering;
default is still to do it globally
mappeR 1.2.0
Add other hierarchical clustering methods (those available from
fastcluster)
Fix issues with 100 percent overlap situations caused by
mapply simplifications
mappeR 1.1.0
Added safety checks for mapper input functions (no NA
inputs, etc)
Removed Cytoscape interfacing for the release version; will remain
in a separate branch
mappeR 1.0.0
Balls/bins with equal amounts of data no longer cause wonkiness
Zero overlap situations should work correctly with
igraph
Added eccentricity lens/filter fuction
mappeR 0.1.5
1D tests will not fail due to a one interval cover
mappeR 0.1.4
Reduced edge widths for Cytoscape visualization.
Rescaled tightness calculation down a bit.
mappeR 0.1.3
Edge overlap data is now present in output mapper objects.
Cytoscape edges are styled by width instead of transparency.
mappeR 0.1.2
The algorithm now still runs even if there is only one level set to
be considered.