4.1.4 NetPlot

NetPlot can be used to illustrate a perceptron, its classification, and its training. Depending on how the option DataFormat is set you can illustrate the classification and the training in different ways.

Display of perceptrons and training of perceptrons.

If the output data y is submitted then the correct class and the class according to the perceptron model can be indicated. For two-dimensional problems the default is to plot the data and the border line(s) of the classifier. For higher-dimensional data the result is represented with bar charts.

NetPlot takes the following options when it is applied to a perceptron network.

Options of NetPlot.

The option DataFormat takes the following values:

DataMap is the default for two-dimensional problems, producing a plot of data along with the class discriminant. If a training record is submitted, then the class discriminant for the intermediate classifiers during training are also plotted. The option Intervals can be used to set the frequency of the intermediate classifiers.

BarChart illustrates the classification result with a bar chart.

ClassPerformance is the default if a training record is supplied instead of a perceptron, and when the input dimension is larger than two. The classification performance is plotted against training iterations with one plot for each class. Each plot indicates the classification performance versus training iterations for one specific class. The solid line indicates the number of correctly classified data samples to that class, and the dashed line indicates the number of incorrectly classified data samples to that class.

DataMapArray gives a graphics array of the progress of the classification during training. This applies only to two-dimensional problems and when a training record is submitted.

The last two possibilities are encountered only when NetPlot is applied to a training record. The option BoundaryDash is active only when DataFormat->DataMap and when NetPlot is applied to a training record.

Depending on the value of DataFormat, NetPlot uses the Mathematica command MultipleListPlot, BarChart, or BarChart3D, and you can modify the plot by submitting the related options.

With DataFormatBarChart the classification result is illustrated with a three-dimensional bar chart indicating class according to output data and class according to the perceptron model. The height of the bar at position {n,m} illustrates the number of data objects belonging to class n according to the supplied output, and classified to class m by the perceptron. Therefore, the bars on the diagonal correspond to correctly classified data.

If no output is submitted, then only the classification according to the model can be given, and the plot illustrates how the input data is distributed over the classes according to the classification of the perceptron.