Daniel@0: function demhint(nin, nhidden, nout)
Daniel@0: %DEMHINT Demonstration of Hinton diagram for 2-layer feed-forward network.
Daniel@0: %
Daniel@0: %	Description
Daniel@0: %
Daniel@0: %	DEMHINT plots a Hinton diagram for a 2-layer feedforward network with
Daniel@0: %	5 inputs, 4 hidden units and 3 outputs. The weight vector is chosen
Daniel@0: %	from a Gaussian distribution as described under MLP.
Daniel@0: %
Daniel@0: %	DEMHINT(NIN, NHIDDEN, NOUT) allows the user to specify the number of
Daniel@0: %	inputs, hidden units and outputs.
Daniel@0: %
Daniel@0: %	See also
Daniel@0: %	HINTON, HINTMAT, MLP, MLPPAK, MLPUNPAK
Daniel@0: %
Daniel@0: 
Daniel@0: %	Copyright (c) Ian T Nabney (1996-2001)
Daniel@0: 
Daniel@0: if nargin < 1 nin = 5; end
Daniel@0: if nargin < 2 nhidden = 7; end
Daniel@0: if nargin < 3 nout = 3; end
Daniel@0: 
Daniel@0: % Fix the seed for reproducible results
Daniel@0: randn('state', 42);
Daniel@0: clc
Daniel@0: disp('This demonstration illustrates the plotting of Hinton diagrams')
Daniel@0: disp('for Multi-Layer Perceptron networks.')
Daniel@0: disp(' ')
Daniel@0: disp('Press any key to continue.')
Daniel@0: pause
Daniel@0: net = mlp(nin, nhidden, nout, 'linear');
Daniel@0: 
Daniel@0: [h1, h2] = mlphint(net);
Daniel@0: clc
Daniel@0: disp('The MLP has been created with')
Daniel@0: disp(['    ' int2str(nin) ' inputs'])
Daniel@0: disp(['    ' int2str(nhidden) ' hidden units'])
Daniel@0: disp(['    ' int2str(nout) ' outputs'])
Daniel@0: disp(' ')
Daniel@0: disp('One figure is produced for each layer of weights.')
Daniel@0: disp('For each layer the fan-in weights are arranged in rows for each unit.')
Daniel@0: disp('The bias weight is separated from the rest by a red vertical line.')
Daniel@0: disp('The area of each box is proportional to the weight value: positive')
Daniel@0: disp('values are white, and negative are black.')
Daniel@0: disp(' ')
Daniel@0: disp('Press any key to exit.'); 
Daniel@0: pause; 
Daniel@0: delete(h1);
Daniel@0: delete(h2);