attimp: Compute and plot relevance of the attributes for selected...
In mkampert/rCOSA: Clustering Objects on Subsets of Attributes (COSA)

Description Usage Arguments Value Note See Also Examples

This function computes and displays the revelance (importance) of each attribute to the clustering of a selected group of objects. The relevance values are displayed in order of descending value. Optionally, on the same plot, are displayed corresponding ordered importances of randomly selected groups of objects of the same size.

attimp(
  X,
  group,
  range = 1:min(20, ncol(X)),
  times = 0,
  lX = if (class(X) == "data.frame") {     lX <- NULL } else {     lX <- 1 },
  donames = FALSE,
  fast = FALSE,
  trans = 1,
  maximp = 20,
  xmiss = NULL,
  horiz = FALSE,
  main = "Cluster",
  att.names = 1:ncol(X),
  ord.names = FALSE,
  names.size = 1,
  ylim = c(0, wtave[o[1]]),
  ntick = 20,
  rlines = TRUE,
  targ = NULL,
  lwd = 1,
  cache = FALSE
)

`X`	data.frame or an input data matrix object of class matrix and mode numeric. Rows are objects and columns are attributes (variables).
`group`	vector of identities of objects in the group
`range`	range of ordered importance values plotted (all are computed)
`times`	number of randomly sampled groups for which ordered importances are also plotted
`lX`	either an integer or a vector of attribute flags with as much elements as columns in `X` (see the help file of `cosa2`). This argument is ignored if `X` is a data.frame '
`donames`	`TRUE` / `FALSE` => do/don't show names of attributes on plot.
`fast`	`TRUE` / `FALSE` => do/don't use fast approximate calculation (less stable estimate). Should only be done with very large groups (when regular calculation takes way too long).
`trans`	power used to transform important scale.
`maximp`	maximum obtainable importance value
`xmiss`	missing value flag (see the help file of `cosa2`).
`horiz`	`TRUE`, `FALSE` => do/don't plot importances horizontally (ignored if time `> 0`).
`main`	plot title
`att.names`	vector of names for each attribute (ignored if donames = `FALSE`).
`ord.names`	a logical value that indicates whether the attributes labels should be ordered from high to low attribute importance (ignored if donames = `FALSE`).
`names.size`	expansion factor for names.
`ylim`	vertical range of plot.
`ntick`	number of tick marks on importance axis.
`rlines`	a logical value that indicates whether the replicate lines should show in the plot as well.
`targ`	target values for computing targeted dissimilarities. The `targ[k]` is the target value for `X[,k]`. The value is ignored if `lX[k] = 0`, `1`, or `4`. If `lX[k]=2` or `5`, then `targ[k]` contains the single target value. If `lX[k]=3` or `6`, then `targ[k]` contains one of the two target values. Special values: `targ = "low"` `targ = "high"` `targ = "high/low"`
`lwd`	line width (see par).
`cache`	a logical value indicating whether the data should show as a variable (object) in the output as well (NB. NOT IN USE YET).

`$imp`	vector of all attribute importances ordered in descending value.
`$att`	vector of attribute identities in descending order of importance.

the lX vector and/or the value of xmiss can be specified as attributes to the input data matrix before invoking attimp (see cosa2)

hclust, getclust, and attvalues.

# make sure X and group are defined... For that, see getclust.
# at <- attimp(X, group[[2]])
# at <- attimp(X, group[[1]], 1:100, trans = 0.5, att.names = names, horiz = TRUE, main= 'Group 1')
# at <- attimp(X, group[[3]], 1:ncol(X), 10, main =' Group 3')

mkampert/rCOSA documentation built on Dec. 23, 2019, 8:21 p.m.