Matrix
.compare_clusterings
to allow complex parameters in lists (#59).compare_clusterings
, fixed a bug that caused the order of columns "k" and "method" to be wrong in the data frame with results for hierarchical configurations.sdtw_cent
now uses stats::optim
by default.bigmemory
in case it is archived. It can still be used if available.array
.dtw_basic
gained parameter sqrt.dist
.compare_clusterings
handled preprocessing of series.RcppThread
.hclust
and proc_time
with setOldClass
now leaves them as virtual classes.pam_cent
.NULL
windows.dtw2
wrapper for multivariate series.compare_clusterings
: cvi_evaluators
.pick.clus
in compare_clusterings
now also contain the configurations' data instead of just the scores. This will require adjustment of existing pick.clus
functions.centroids
in the objects returned by tsclust
now have an attribute series_id
if said centroids were extracted from the provided series without modification. It is an integer vector indicating which series were chosen as centroids.compare_clusterings
.no.expand
to compare_clusterings_configs
.dtwclust-package
.interactive_clustering
and ssdtwclust
.compare_clusterings
: repeat_clustering
.distmat
to fuzzy_control
for when centroid = "fcmdd"
.rngtools
. Changing the random number generator after loading dtwclust
should not be a problem now.tsclust
handled the seed
when called from compare_clusterings
. This will affect compare_clusterings
results of hierarchical or TADPole clusterings made therein if a centroid subject to randomness was used (e.g. dba
).RcppParallel
for parallelization, refer to the new parallelization vignette for more information.sdtw_cent
is now actually passed to nloptr
.tsclust
for dtw_lb
distance and pam.precompute = FALSE
(it was always using dtw_basic
, so the results should not change).clusinfo
slot for TSClusters
class for clusterings that ended with empty clusters.sigma
estimation in the non-proxy version. The estimation routine in the proxy version also changed slightly, and the estimates will be different due to randomness.tslist
to hopefully make it faster.tsclust-controls
) now specify version 2 of the main loop by default.clue
and ggplot2
are no longer automatically attached by default.plyr
dependency for dplyr
.cvi
.TADPole
clustering through tsclust
, custom parameters can now be passed to a custom centroid function if provided.pam.sparse
to TRUE
if you want this functionality.zscore
for data frame input (it was still coercing column-wise instead of row-wise).proxy
versions of lb_keogh
, lb_improved
, SBD
, GAK
and dtw_basic
to C++
and improved them by using the bigmemory
package.TADPole
to C++
.C++
.tsclustFamily
's dist
function for matrix or data frame input.dtw_lb
distance and pam.precompute = FALSE
(#16).dtw_lb
to C++
when using dtw_basic
.TADPole
for multiple k
and dc
values.pam.precompute = FALSE
by using sparse matrices from the Matrix
package.shape_extraction
by using the eigs_sym
function from the RSpectra
package.C++
.C++
.symmetric
control for fuzzy clustering.compare_clusterings
now take into account the symmetric
control parameter.pick.clus
in compare_clusterings
changed depending on the value of return.objects
.tsclust
to have duplicated elements in the args
slot.update(old_TSClusters_obj)
.dtwclust
function is now deprecated. Try using as(dtwclust_class_obj, "TSClusters")
for old objects.compute_envelop
to compute_envelope
(old one still available but deprecated).compare_clusterings
and helpers to compare many clustering configurations, possibly in parallel.tsclust
that prevented CVIs to be calculated for hierarchical/TADPole cases if a custom centroid function was used.seed
to the objects returned by tsclust
.tsclust
's ellipsis are now passed to all preprocessing, centroid and distance functions.tsclust
when using DTW.tsclust
in the examples.dtwclust
with a parallel backend and multiple values of k
, in case the centroid function has randomness associated.tsclust
that should be functionally equivalent to dtwclust
but is hopefully more coherent in general.dtw_basic
when backtrack = FALSE
.force.symmetry
argument in dtw_lb
was removed since it didn't serve any real purpose.DTW/SBD
algorithms (and hence the functions that depend on them) might give different results in installations with 32-bit architectures.dtwclust
objects saved, try updating them with attr(dtwclust_object, "centers") <- NULL
if you run into compatibility problems.DBA
arguments changed order.dba.alignment
argument from DBA
since other step.pattern
s don't really work.GAK
.clue
.CharTrajMV
.dtw2
function.zscore
and reinterpolate
functions.proxy
.shape_extraction
. Reminder: multivariate shape extraction might not be a good idea.center(s)
arguments/slots will be removed and replaced with centroid(s)
.dtw_basic
function, which should be faster due to its limited functionality. It can
also be used with dtw_lb
and DBA
. It will now be used by default.SBD
and shape_extraction
functions used the crossprod
function internally, which returned a 1x1 matrix by default, causing some dimension inconsistencies. In the future, R
will give an error about the inconsistency, so the function has been changed in dtwclust
, but it resulted in very small numerical differences, which may be enough to alter some clustering results.cvi
functionforce.pairwise
used in some of the proxy
distances is not necessary anymoreNULL
dimensions for each series (in case multivariate series are to be used), use with cautionk = 2
predict
generic.fcluster
slot needed one final update during clustering. It should be correct now, but it will vary slightly with respect to what was previously given in v2.1.0.k
values for multiple runsdtwclust
function were dropped and implemented in the formal class dtwclustControl
. For now, they will still be supported through ...
with a message.flexclust
's kccasimple
hclust
class now, and all its associated methodsshape_extraction
now accepts series with different lengths!proxy
distances directly (except DTW2
)distmat
slot and update
generic to save some time if possible. See examples of dtwclust
.dtw_lb
function now correctly warns about pam.precompute
being TRUE
.DTW_LB
, SBD
, LB_Keogh
and LB_Improved
. See their respective notes. The distance function created for the dtwclustFamily
slot also supports this.doRNG
packageTADPole
and dtw_lb
too, probably negligible (maybe even detrimental) for small datasets.SBD
function registered with proxy
, it's a lot faster nowDBA
and shape_extraction
so that centers are only recomputed if necessaryTADPole
DBA
, as well as window constraintDBA
calculationsdtwclust
classDBA
and SBD
)kccaFamilies
, supporting any distance registered in proxy
insteaddtw
or dtw2
were being used, in which case they may have been erroneously computed beforeAdd the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.