ggInterval_ecdf() displays exact endpoint ECDF envelopes
and an optional midpoint ECDF; ggInterval_CRcor() compares
pooled and optional group-demeaned Pearson associations among centers
and full ranges; ggInterval_deviation() displays
endpoint-deviation contributions; and ggInterval_overlap()
displays geometric overlap between interval boxes.interval_diagnostics
attribute. They accept symbolic tables, ggInterval objects containing
intervalData, and paired lower/upper matrices. They do not
change existing plotting interfaces or algorithms.Installed the existing aggregation-example CSV under
extdata, so examples can locate it with
system.file() after package installation. The file is
unchanged; the documented cleanup removes its empty trailing
column.
Added stock.data.LH, a 62-row symbolic table of
low-high intervals and accompanying date/price columns for the financial
line-plot example. It is derived from the randomly generated
TTR::ttrc data (TTR 0.24.4), not an actual security. The
stored table supports head() and the existing
interval-valued line-plot interface without a run-time TTR
dependency.
Corrected the INCOMELEVEL category labels of
Environment to the order of the SODAS source files (25, 75,
50, 90, 100). Earlier versions listed the labels as 25, 50, 75, 90, 100
while storing the proportions in the source order, so the labels 50 and
75 were interchanged. Compatibility note: only these two labels change;
all proportions and interval endpoints, and hence the radar bar
positions and heights, are unchanged.
ggInterval_PCA(poly = TRUE) now stops with an
informative message when duplicated vertices were removed (zero-width
intervals or vertices shared between observations); it previously failed
with subscript out of bounds, for example for
mushroom, blood.i and
Cardiological. The polytope edge table is now built in one
vectorised step: results are identical and the facedata
example runs in a fraction of the previous time.
Clarified the scatterplot example’s transparency settings without changing the function’s existing mapping precedence or rendered example.
Clarified that quantile-box widths are heuristic interval-mass display weights, and that unseeded K-means examples show one illustrative run.
Radar layers now map columns from their own layer data. This removes discouraged external-data aesthetic mappings and corrects the coordinates of two invisible guide paths that previously reused the final observation in a loop. The guides remain invisible; the accepted manuscript radar displays retain their visible geometry.
Radar interval-label opacity keys now use an explicit manual
scale with the existing palette and ordering. The polygon
alpha argument is unchanged; interval labels do not acquire
a quantitative interpretation.
Scatter mappings no longer retain duplicate aesthetics. Existing first-mapping precedence is preserved, including the default mapped alpha. The three-dimensional scatterplot’s fixed coordinate axes no longer inherit every observation as data for constant aesthetics.
The documented RSDA PCA example converts the score-table container to a row-name-compatible symbolic data frame before assigning observation names. Interval columns, PCA results and plotted labels are retained.
The ungrouped, single-variable equal-bin histogram now draws its
precomputed frequencies with geom_bar(stat = "identity").
This removes warnings about unused binwidth and
bins parameters while retaining the existing bin
construction, frequency tables and bar geometry.
Named-column aggregation in classic2sym() now passes
the grouping column through dplyr::all_of() to RSDA. This
removes the external-vector selection deprecation warnings without
changing the grouped summaries; a missing grouping column still produces
an error.
ggInterval_hist(method = "equal-bin"),
ggInterval_2Dhist() and
ggInterval_2DhistMatrix() now assign a point interval
(min == max) unit mass to exactly one bin. Bins are
left-closed and right-open, and the final right endpoint belongs to the
final bin. Previously a point produced NaN frequencies in
the 1D histogram and was counted in every touching bin in the 2D
histograms (a point on an internal break received mass 4 instead of 1).
Positive-width intervals keep the existing overlap-length
weighting.
Constant variables (all intervals share one value) no longer
fail. In the mass-based modes (1D equal-bin and 2D) their support is
padded by max(abs(value) * 0.05, 0.5) on each side (the
rule already used by the 2D helpers). In the 1D unequal-bin
coverage-count display a constant variable, including a single
observation, is represented by its coverage count at the exact value as
a zero-width rectangle (vertical spike); no 0.01 grid is built for it,
so the work does not grow with the magnitude of the value.
Equal-bin breaks are generated with exactly bins + 1
values from the pooled minimum and maximum, so the last bin can no
longer be lost to floating-point error.
bins, xBins and yBins are
validated as finite positive integer scalars before use in equal-bin
mode; the existing bin-count limits are unchanged and unequal-bin mode
still ignores these arguments.
The 1D unequal-bin display initialised its first rectangle from
the first coverage count instead of the first grid index; the first run
is now encoded correctly when several intervals cover the minimum.
Compatibility exception: this changes an existing positive-width output.
For facedata AD the first unequal-bin table row changes
from [149.35:150.32] 1 (0.2.5) to
[149.34:150.32] 2 (KHA1 and KHA2 both start at 149.34); the
other facedata variables and all equal-bin/2D outputs are
unchanged.
Non-finite interval endpoints are rejected with an explicit error
instead of propagating NA.
Updated the introductory vignette with direct loading and preview
of the synthetic stock.data.LH table, an optional TTR
reconstruction, and executable examples of the four diagnostics using
the manuscript’s unchanged simulation mechanism and seed
20260709.
The vignette now distinguishes histogram mass allocation from coverage counts, documents point/constant intervals, explains BG formula provenance and normalization, and states the limits of the face and Environment data. It also clarifies PCA and radar interpretations and adds layers to the returned PCA plot component.
Histogram help now states the break construction, the overlap weighting, the point and endpoint convention, the constant-range padding, and the actual returned components; the 1D unequal-bin display is described as a 0.01-grid coverage count (“Cumulative Concepts”), which is not a probability.
ggInterval_PCA() documents its actual return value
(an augmented princomp object with
$scores_interval and $ggplotPCA).
scale() documents the class-specific behaviour of
the symbolic methods.
facedata documentation describes nine subjects with
three image sequences per subject (Arce Garro and Rodriguez Rojas,
2019).
Environment and facedata documentation
describes the verified structure and source chain, including the
corrected INCOMELEVEL label order.
cran-comments.md
history).