miclust 1.3.0
Minor changes:
- Dependences:
- The package is now completely independent of the
irr
package, which was previously used to internally calculate Cohen’s Kappa
(irr::kappa2). Now, Cohen’s Kappa is calculated using a
native base implementation to improve stability and maintainability. In
addition, the Cohen’s Kappa calculation is now faster.
- Documentation:
- Updated source code documentation to comply with the latest
roxygen2 (v8.0.0) standards.
- Improved
README.Rmd plot rendering.
- Integrated documentation for
miclust() and its related
methods print.miclust() and plot.miclust()
into a single help page for a more fluid usage.
- Integrated documentation for methods
summary.miclust()
and print.summary.miclust().
miclust:
- The width of the verbose message has been reduced (now, breaks line
each 20 imputations).
print.miclust:
- Fix erratum in labels of the output of
print.miclust.
plot.miclust:
- New parameter
metric, to select which metric is
displayed. Defaults to "all", maintaining compatibility
with the output from versions 1.2.8 and earlier. Other options are:
"nclfreq" (percentage of times each number of clusters has
been selected); "critcf" (critCF distribution for each
number of clusters); "nvarfreq" (distribution of the number
of selected variables), and "varsel" (percentage of
appearance of the variables that remained in the final set of selected
variables).
- New parameters
col.nclfreq, col.critcf,
col.nvarfreq, and col.varsel, to customize the
color used in the plots for metrics, "nclfreq",
"critcf", "nvarfreq", and
"varsel", respectively. Default values are consistent with
results under versions 1.2.8 and earlier.
- New parameter
col.all. When specified, it overrides
col.nclfreq, col.critcf,
col.nvarfreq, and col.varsel, and applies that
color across the entire panel. Default (NULL) is consistent
with results under versions 1.2.8 and earlier.
summary.miclust:
- Fix output errors related to within-cluster missingness of the
selected variables.
- Fix an internal error in
summary.miclust() (originally
triggered during kcca() execution inside
assignprobandkappas()) that occurred when a specific
imputation or the final assigned clustering result contained an empty
cluster. Specifically, add a validation check in
summary.miclust() to detect empty clusters across all
imputations and the assigned cluster for the given k. The
function now prints the affected cases and halts execution with a
detailed error message.
- An internal error in
summary.miclust() (originally
triggered during kcca() execution inside
assignprobandkappas()) occurs when the final clustering for
a given k and the set of selected variables results in an
empty cluster within specific imputed datasets. This indicates that such
a combination of k and selected variables is unable to find
as many clusters as requested. A validation check has been added in
summary.miclust() to detect these scenarios and halt
execution with a detailed error message.
print.summary.miclust:
- The
Within-cluster summary section of the output has
changed. To reduce the output width, the columns for missingness, mean,
and standard deviation have been collapsed into a single column. The
section label has been changed to
Within-cluster summary [%miss.;mean;sd]. Separated columns
(version <= 1.2.8 style) are still available in the
summarybycluster element of the object.
miclust 1.2.8
Minor changes:
- Update deprecated code in
CITATION.
miclust 1.2.7
Minor changes:
- Fix two missed figures in
README.
miclust 1.2.6