BoltzmannMachinesPlots.BoltzmannMachinesPlots
— ModuleContains all plotting functions for displaying information collected in module BoltzmannMachines
. Most important function is plotevaluation
.
BoltzmannMachinesPlots.crossvalidationcurve
— Functioncrossvalidationcurve(monitor)
crossvalidationcurve(monitor, evaluationkey)
Plots the results of cross-validation experiments conducted over the course of the training. For each training epoch, the results of the evaluations for the different data sets used in the cross-validation are displayed in form of box plots. The mean values of the evaluations are also displayed. Such a plot be used to determine the optimal number of training epochs if the other parameters are given.
The monitor
argument contains the monitoring results of an evaluation criterion. Similar to plotevaluation
, an evaluationkey
can be specified if the monitor
object contains multiple evaluations.
See also: BoltzmannMachines.crossvalidation
, BoltzmannMachines.monitored_fitrbm
, BoltzmannMachines.monitored_fitdbm
.
BoltzmannMachinesPlots.plotestimatedprob
— MethodPlots the information about the estimated lower bound of the log probability that has been gathered while training a BMs.
BoltzmannMachinesPlots.plotevaluation
— Functionplotevaluation(monitor; ...)
plotevaluation(monitor, evaluationkey; ...)
Plots a curve that shows the values of the evaluation contained in the monitor
and specified by the evaluationkey
over the course of the training epochs. If no evaluationkey
is specified, the evaluation type of the first monitor element is used.
Optional keyword argument sdrange
: For evaluations with keys BoltzmannMachines.monitorloglikelihood
and BoltzmannMachines.monitorlogproblowerbound
, there is additional information about the standard deviation of the estimator. With the parameter sdrange
, it is possible to display this information as a ribbon around the curve. The ribbon indicates the area around the curve that contains the values that deviate at maximum sdrange
times the standard deviation from the estimator. Default value for sdrange
is 2.0.
BoltzmannMachinesPlots.scatter
— Methodscatter(xy; ...)
Creates a two-dimensional scatter plot from the first two columns of the matrix xy
. Each of the rows in the matrix is displayed as one point. The two columns of the matrix contain the x- and y-values of the points.
Optional named arguments:
labels
: a vector of labels of the same length as the columns in the matrixxy
. Each entry contains a label for a row ofxy
.opacity
: the opacity of the dotsxlabel
: a label for the x-axisylabel
: a label for the y-axis
scatterhidden(bm, x; ...)
scatterhidden(h; ...)
Creates a scatter plot of the logarithmized activation potential of hidden nodes. The activation is either induced by the dataset x
in the Boltzmann machine bm
or it is directly specified as matrix h
. This function can be used to inspect pairs of hidden nodes. For getting a reduced view on a larger number of hidden nodes, consider employing BoltzmannMachines.top2latentdims
and scatter
.
Optional keyword arguments:
hiddennodes
: Tuple of integers, default(1,2)
, selecting the first two nodes of the (last) hidden layer.labels
: a vector containing string labels for each of the data points