Description Usage Arguments Details Author(s) References See Also Examples
Plot a som object, several plot can be drawn, see below.
1 |
object |
a som object |
type |
character specifying the type of the graph. Possible values are : [distance | mean.ss | effectif | meta | variable | energy | dxdy] (see details) |
cex |
character expansion of titles and legend. |
nom.variable |
It has no effect if type is different from 'variable'. Name of the variable of weights to plot. |
precision |
Precision of the displayed label if it is numeric |
cex.label |
character expansion of the displayed info in each polygon of the network. |
distanceeuclidean distances between a class and all the other classes. Change the reference class with <s-expression>set.current.case</s-expression>. \itemmean.ssmean sums of squares for each prototype. \itemeffectifnumber of rows classified in each prototype. \itemmetameta-groups (see <s-expression>plot.clust</s-expression> for precisions) \itemvariableweights values for a variable in each prototype. It allows to see if a variable if used by the network or not (if not, values are quite equals). \itemenergyplot the energy function during the learning process. This represent the intra-inertia extended to the neighborhood at a stage of the learning process. This is the objective function to be minimised during learning \itemdxdyThe quality of a projection can be evaluated by the 'dy-dx' representation. It is a plot of all the possible distances in the input space, dx's, versus their respective distances in the output space, dy's. For a linear projection the 'dy-dx' plot should be linear.
David Gohel
Demartines, P. and J. Herault. Curvilinear component analysis: A self-organizing neural network for nonlinear mapping of data sets. Kohonen, T. (1995). Self-Organizing Maps
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 | library(MASS)
lcrabs <- log(crabs[, 4:8])
lcrabs.som <- som ( formula = ~ . , data = lcrabs
, neighborhood = "uniform"
, grid = grid ( xdim = 10 , ydim = 10 , type = "hexagonal" )
, weights.min = min (lcrabs), weights.max = max (lcrabs)
)
lcrabs.som <- learn( lcrabs.som , number.iter = 500 , max.alpha = 0.5, min.alpha = .001, max.rayon = 3 , step.eval.si = 50)
plot(lcrabs.som, "energy")
#--- quality of learn, points must be higly correlated and in a line from c(0, 0) -> c(1, 1)
plot ( lcrabs.som , "dxdy" )
#--- plot number of cases per neuron/class
plot(lcrabs.som, "effectif", cex.label = 0)
#--- plot means sums of squares
plot(lcrabs.som, "mean.ss", cex.label = 0.3)
#--- change current case
lcrabs.som <- set.current.case(lcrabs.som, 4)
plot(lcrabs.som, "distance", cex = .75, cex.label = 0)
#--- plot values for a variable... useful for selecting active varaible
plot(lcrabs.som, "variable")
plot(lcrabs.som, "variable", nom.variable = "FL")
## Not run:
#--- construct meta class , click on the graph, the y will be used to collapse classes in meta-classes ---#
lcrabs.som <- plot.clust(lcrabs.som, interactive = T)
## End(Not run)
#--- plot meta class ---#
plot(lcrabs.som, "meta" , cex.label = .5 )
|
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.