Nothing
seqdef
added as an exact copy of TraMineR::seqdef
, to enable experienced
users of MEDseq
& TraMineR
to use the former without needing to explicitly load the latter.MEDseq_clustnames
gains the arg. weighted=FALSE
for use when size=TRUE
:
this is now respected by the weighted
arg. to plot.MEDseq
where relevant.dist_freqwH
added for calculating pairwise dissimilarity matrix associated with
wKModes(..., freq.weighted=TRUE)
for subsequent use (e.g. silhouettes).plot.MEDseq
function's type
arg. gains the option "dH"
,
provided version 2.2-4
or later of the TraMineR
package is installed.plot.MEDseq
also gains the "similarity"
option for its type
argument.MEDseq_AvePP
added.wKModes
now also returns x$tot.withindiff
(i.e. sum(x$withindiff)
).wKModes
when freq.weighted=TRUE
.type="dbsvals"
& type="aswvals"
in plot.MEDseq
.plot.MEDseq
related to its seriated
arg. in G=1
settings.MEDseq_fit
& wKModes
.G>1
.G>1
.MEDseq_meantime
gains the map.size
arg. and a related print
method.summary
(and related print
) methods for MEDCriterion
objects.MEDseq_entropy
added.TraMineR
release, w.r.t. "mt"
and "ms"
plots.WKModes
(& thus related MEDseq_control
init.z
options "kmodes"
/"kmodes2"
), by further altering klaR::kmodes
: wKModes
arg. random
(defaults to TRUE
).modes
is supplied as a number with aggregated data, e.g. "kmodes2"
.MEDseq_fit
& other functions now work for sequence alphabets of any size;
previously, only sequences with fewer than 10 states/categories were accommodated.dbs
function when supplying clusters
with a noise component.sapply
replaced with vapply
, with other negligible speed-ups.init.z
options "kmodes"
& "kmodes2"
in MEDseq_control
, with new function wKModes
provided for running the k-modes algorithm on weighted data: previously, k-modes initialisation
was only available for unweighted sequences via the now-replaced klaR::kmodes
function
(consequently, the klaR
package has been removed from the DESCRIPTION
Suggests:
field).plot.MEDseq
gains the arg. subset
, for use with the TraMineR
type
plots:
allows plotting some but not all components, e.g. only the noise component (see documentation).MEDseq_fit
to crash when weights
are supplied and unique=FALSE
.unique=TRUE
, the default).type="ms"
plots for models with a noise component when SPS=TRUE
.noise.gate
in MEDseq_compare
for G=2
models w/ noise & gating covariates.G
in MEDseq_fit
.plot.MEDseq
gains a number of new arguments: soft
allows soft cluster membership probabilities to be used for the "d"
, "f"
, "Ht"
, "ms"
,
& "mt"
type
plots (default: soft=TRUE
) + the "i"
& "I"
plots (default: soft=FALSE
), in a
manner akin to WeightedCluster::fuzzyseqplot()
: previously, all but the "ms"
plot used the
hard MAP partition and discarded the soft assignment information (i.e. soft=FALSE
, implicitly).sortv
allows overriding the smeth
arg. to instead order observations in certain plots
(where seriated
is one of "observations"
or "both"
) by the "dbs"
or "asw"
values;
additionally, and for consistency with WeightedCluster::fuzzyseqplot()
,
sortv="membership"
is provided for soft=TRUE
type="I"
plots.weighted
(TRUE
, by default) allows control over whether the weights (if any) are used;
relevant only for "d"
, "f"
, "Ht"
, "i"
, "I"
, "ms"
, & "mt"
type
plots.MEDseq_clustnames
& MEDseq_nameclusts
functions and added SPS
arg. to plot.MEDseq
,
MEDseq_meantime
, MEDseq_stderr
, & various/more print
/summary
methods: now certain plots &
outputs can be (or are by default) labelled with the central sequences in SPS format, as per the paper.seriated
options "observations"
& "both"
can now be used for "i"
type plots,
with related minor fixes for "i"
& "I"
type plots for weighted data with seriated observations.predict
, fitted
, & residuals
methods for "MEDgating"
objects, i.e. x$gating
.MEDseq_meantime
gains the arg. wt.size
(defaults to FALSE
).modtype="CU"
.itmax
arg. to MEDseq_control
: the 2nd element of this arg. governs the maximum number of
MLR iterations --- consequently, its default has been modified from 100
to 1000
, which is liable to slow
down internal calls to nnet::multinom
, but generally reduces the required number of EM iterations. Suggests:
package viridisLite
now only invoked if available.x$gating
object, especially for equalPro
models
with a noise component and weighted models without any gating covariates at all.weights
arg. is explicitly supplied to MEDseq_fit
in cases where the "stslist"
object passed via seqs
has the "weights"
attribute.MEDseq_fit
when the number of states exceeds 9,
to better inform of this bug which will be rectified in future updates.gating
formulas when there are duplicates.get_MEDseq_results
and how its optional args. are internally handled by plot.MEDseq
.gating
formula which are not found in covars
.type="mean"
option renamed to type="central"
in plot.MEDseq
.type="ms"
plots now work properly when seriated="clusters"
or seriated="both"
."mt"
TraMineR
type
plots.MEDseq_meantime
when MAP=FALSE
.print.MEDseq
for models where DBS &/or ASW statistics weren't computed."d"
, "f"
, "Ht"
, "i"
, & "I"
plot types now properly account for sampling weights.TraMineR
further, plot.MEDseq
also gains the type
options "ms"
& "mt"
.opti="medoid"
setting.criterion="bic"
is now the default for MEDseq_control
, MEDseq_compare
, and
get_MEDseq_results
(previously "dbs"
), with modifications to print
& summary
functions.print.MEDseqtheta
) & plotted (plot.MEDseq(..., type="mean")
) always:
the preczero
argument has thus been removed from both functions.MEDseq_meantime
gains two new arguments (see documentation for more details): weighted
(default: TRUE
, old: FALSE
) allows the sampling weights to be used,
with or without the cluster assignment probabilities, in the computation of the weighted averages.prop
(default: FALSE
) divides the output when norm=TRUE
by the sequence length.MEDseq_control
gains the arg. random=TRUE
, governing tie-breaking of estimated central sequence
positions: old behaviour (always choosing the first candidate state) recoverable via random=FALSE
.plot.MEDseq
arg. quant.scale=FALSE
replaces old arg. log.scale
: quantiles now used
to determine non-linear colour breakpoints when invoked with type="precision"
.init.z="kmedoids"
initialisation via pam
for unweighted sequences, by using the
highest available value for the pamonce
option, based on the cluster
package's version number.init.z
gains the options "kmodes"
& "kmodes2"
, though only for unweighted sequences:
both require the newly suggested klaR (>= 0.6-13)
package.plot.MEDseq
gains the arg. smeth
, governing the seriation method to be used ("TSP"
, by default).init.z="kmedoids"
is now itself initialised by Ward's hierarchical clustering.opti
settings (esp. "mode"
).SPS
arg. (default=FALSE
) to print.MEDtheta
& summary.MEDseq
.dbs
gains the optional/experimental arg. clusters
- no change to default.seriated
arg. to plot.MEDseq
: seriate
to avoid conflict with function seriation::seriate
. seriated
options "clusters"
/"both"
for models with no noise component. seriated="observations"
(the default) now also works for type="I"
plots. seriated="clusters"
now also works for type="dbsvals"
& type="aswvals"
plots. MEDseq_fit
now always internally normalises the weights
to sum to the sample size.noise.gate=FALSE
.noise.gate=FALSE
when G=2
.MEDseq_stderr
now respects the algo
, opti
, & noise.gate
settings of the original model.MEDseq_compare
now returns & prints opti
info where relevant.print
& summary
methods for MEDgating
objects if equalPro=TRUE
.MEDseq_fit
now coerces "character"
covariates to "factor"
.print
method for MEDlambda
objects also.plot.MEDseq(..., type="gating")
.print.MEDseqCompare
gains the args. maxi
& rerank=FALSE
.G=1
models.viridisLite (>= 0.2.0)
to Suggests:
(for plot.MEDseq(..., type="precision")
).matrixStats (>= 0.53.1)
and TraMineR (>= 1.6)
in Imports:
.summary.MEDseq
gains the printing-related arguments
classification=TRUE
, parameters=FALSE
, and gating=FALSE
.x$params$lambda
now inherits the MEDlambda
class,
with its own print
method as per x$params$theta
.x$params$tau
now has informative dimnames
.x.axis
to plot.MEDseq(..., type="gating")
.rmarkdown
to Suggests:
.MEDseq_stderr
is provided for computing the standard errors of the
coefficients for the covariates in the gating network via either the
weighted likelihood bootstrap or jackknife methods.get_MEDseq_results
when what="MAP"
and non-noise models are chosen.summary
on x$gating
.plot.MEDseq
when type="clusters"
for small sample sizes.donttest
examples.Any scripts or data that you put into this service are public.
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.