chac {adjclust}  R Documentation 
S3 class for Constrained Hierarchical Agglomerative Clustering results
## S3 method for class 'chac' as.hclust(x, ...) ## S3 method for class 'chac' print(x, ...) ## S3 method for class 'chac' head(x, ...) ## S3 method for class 'chac' summary(object, ...) ## S3 method for class 'chac' plot( x, y, ..., mode = c("standard", "corrected", "totaldisp", "withindisp", "averagedisp"), nodeLabel = FALSE ) diagnose(x, graph = TRUE, verbose = TRUE) correct(x) cutree_chac(tree, k = NULL, h = NULL)
x, object, tree 
an object of class 'chac' 
... 
for 
y 
not used 
mode 
type of dendrogram to plot (see Details). Default to

nodeLabel 
(logical) whether the order of merging has to be displayed
or not. 
graph 
(logical) whether the diagnostic plot has to be displayed or
not. Default to 
verbose 
(logical) whether to print a summary of the result or not.
Default to 
k 
an integer scalar or vector with the desired number of groups 
h 
numeric scalar or vector with heights where the tree should be cut. Only available when the heights are increasing 
Methods for class 'chac'
When plot.chac
is called with
mode = "standard"
, the standard dendrogram is plotted, even though,
due to contingency constrains, some branches are reversed (decreasing
merges). When plot.chac
is called with
mode = "corrected"
, a correction is applied to original heights so as
to have only non decreasing merges). It does not change the result of the
clustering, only the look of the dendrogram for easier interpretation.
Other modes are provided that correspond to different alternatives
described in Grimm (1987):
in mode = "withindisp"
, heights correspond to withincluster
dispersion, i.e., for a corresponding cluster, its height is
I(C) = ∑_{i \in C} d(i,g_C)
where d is the dissimilarity used to cluster objects and g_C is the center of gravity of cluster C. In this case, heights are always non decreasing;
in mode = "totaldisp"
, heights correspond to the total
withincluster dispersion. It is obtained from mode = "standard"
by
the cumulative sum of its heights. In this case, heights are always
non decreasing;
in mode = "averagedisp"
, heights correspond to the
withincluster dispersion divided by the cluster size. In this case, there
is no guaranty that the heights are non decreasing. When reversals are
detected, a warning is printed to advice the user to change the mode of the
representation.
Grimm (1987) indicates that heights as provided by
mode = "withindisp"
are highly dependent on cluster sizes and that
the most advisable representation is the one provided by
mode = "totaldisp"
. Further details are provided in the vignette
"Notes on CHAC implementation in adjclust".
The function plot.chac
displays the dendrogram and
additionally invisibly returns an object of class
dendrogram
with heights as specified by the user through
the option mode
.
diagnose
invisibly exports a data frame with the
numbers of decreasing merges described by the labels of the clusters being
merged at this step and at the previous one, as well as the corresponding
merge heights.
The function correct
returns a chac
objects with
modified heights so as they are increasing. The new heights are calculated in
an way identical to the option mode = "corrected"
of the function
plot.chac
(see Details). In addition, the chac
object has its
field method
modified from adjClust
to
adjClustmodified
.
The function cutree_chac
returns the clustering with
k
groups or with the groups obtained by cutting the tree at height
h
. If the heights are not increasing, the cutting of the tree is based
on the corrected heights as provided by the function correct
.
Grimm, E.C. (1987) CONISS: a fortran 77 program for stratigraphically constrained analysis by the method of incremental sum of squares. Computer & Geosciences, 13(1), 1335.