8.1.5 NetPlot

The command NetPlot can be used to illustrate the performance of a model, and to evaluate the training of it. The command can be used in the same way as for FF and RBF networks. Depending on how the option DataFormat is set, the command can be used in very different ways. Here, the possibilities that are interesting in connection with dynamic models are presented. Please see Section 5.1.4, NetPlot, for other possibilities.

The NetPlot function.

When NetPlot is applied to training records from NeuralAR(X)Fit it takes the same options as when it is applied to training records from NeuralFit. The following two options have different default values:

Options of NetPlot.

You can also submit options to modify the graphical output. Depending on the chosen option for DataFormat the graphic is created by BarChart, MultipleListPlot, ListPlot, Plot3D, or Histogram.

If a dynamic neural network model is submitted then the option DataFormat can be given any of the following values.

HiddenNeurons: gives the output values of the hidden neurons as a function of time when the model is used for prediction on the data.

LinearParameters: linearizes the model at each time instant at the value of the regressor vector and displays the linear parameters versus data.

FunctionPlot: plots the mapping using the range of the supplied data. This option can only be used if the model has a one- or two-dimensional regressor.

ErrorDistribution: gives a bar chart of the prediction errors. You can modify the presentation of the result using any of the options applicable to Histogram.

If you submit a training record to NetPlot, then the default value of DataFormat is ParameterValues. This gives a plot of the parameters versus training iterations.

Except for the default value, ParameterValues, you can also give DataFormat any of the possible values. You then obtain animations of the corresponding results as a function of the number of training iterations. The frequency of the plotting can be set with the option Intervals, which indicates the number of iterations between each plot.