function s = getNeuralNetExpression(net, outputIndex)
% getNeuralNetExpression (SUMO)
% Originally part of the Surrogate Modeling Toolbox ("SUMO Toolbox")
% Contact : sumo@intec.ugent.be - www.sumo.intec.ugent.be
% This code is released under the GPL version 2 or later
%
% Description:
% Return the symbolic expression for the given Matlab Neural Network
% Note: only standard feed forward networks (created through newff) are supported
% so no delays or combination functions other than netsum are supported.
if(~exist('outputIndex'))
outputIndex = 1;
end
% Get the expression for the given output neuron
% The final expression is recursively built up
s = expressionForNeuron(net,net.numLayers,outputIndex,'');
% calculate the expression for neuron number 'neuron' in layer 'layer'
function s = expressionForNeuron(net,layer,neuron,s)
% should the transfer function names be replaced by their equations
flattenTfunc = 1;
% precision for the weights and biases
precision = '%0.6d';
% base case of the recursion, the input layer
if(layer == 1)
% each input is denoted by xi
str = '';
% get the matrix of weights from the input layer to the inputs
L = net.IW{layer,1};
% NOTE: we only consider the netsum function (neuron activation is a weighted sum of inputs)
for i=1:net.inputs{layer}.size;
w = L(neuron,i);
input = [ 'x' num2str(i)];
str = [str input '*' sprintf(precision,w) ' + '];
%str = [str input '*' num2str(w) ' + '];
end
% dont forget to add the bias
biases = net.b{layer};
bias = biases(neuron);
str = [str sprintf(precision,bias)];
%str = [str num2str(bias)];
if(flattenTfunc)
str = ['(' expressionForTfunc(net.layers{layer}.transferFcn,str) ')'];
else
str = [net.layers{layer}.transferFcn '(' str ')'];
end
s = str;
%disp(sprintf('Input neuron %i: %s',neuron,s));
else
% calculate the expression for one of the hidden or output neurons
fanIn = net.layers{layer-1}.size;
L = net.LW{layer,layer-1};
s = '';
% again we only consider a weighted sum of inputs
for i=1:fanIn
w = L(neuron,i);
% recurively call the function for each neuron in the fan-in
input = expressionForNeuron(net,layer-1,i,s);
s = [s input '*' sprintf(precision,w) ' + '];
%s = [s input '*' num2str(w) ' + '];
end
% dont forget to add the bias
biases = net.b{layer};
bias = biases(neuron);
s = [s sprintf(precision,bias)];
%s = [s num2str(bias)];
if(flattenTfunc)
s = ['(' expressionForTfunc(net.layers{layer}.transferFcn,s) ')'];
else
s = [net.layers{layer}.transferFcn '(' s ')'];
end
%disp(sprintf('Layer %i neuron %i: %s',layer,neuron,s));
end
end
% given a name of a matlab neural network transfer function, return the mathematical expression
% applied to the given input argument s
function expr = expressionForTfunc(tfunc,s)
if(strcmp(tfunc,'tansig'))
% definition: a = tansig(n) = 2/(1+exp(-2*n))-1
expr = ['2/(1+exp(-2*(' s ')))-1'];
elseif(strcmp(tfunc,'logsig'))
% definition: logsig(n) = 1 / (1 + exp(-n))
expr = ['1/(1+exp(-(' s ')))'];
elseif(strcmp(tfunc,'radbas'))
% definition: exp(-n^2)
expr = ['exp(-(' s ')^2)'];
elseif(strcmp(tfunc,'purelin'))
expr = ['(' s ')'];
else
error('Unsupported transfer function');
end
end
end
% Information from the Mathworks website
% Weight and Bias Values
%
% These properties define the network's adjustable parameters: its weight matrices and bias vectors.
%
% net.IW
%
% This property defines the weight matrices of weights going to layers from network inputs. It is always an Nl x Ni cell array, where Nl is the number of network layers (net.numLayers), and Ni is the number of network inputs (net.numInputs).
%
% The weight matrix for the weight going to the ith layer from the jth input (or a null matrix []) is located at net.IW{i,j} if net.inputConnect(i,j) is 1 (or 0).
%
% The weight matrix has as many rows as the size of the layer it goes to (net.layers{i}.size). It has as many columns as the product of the input size with the number of delays associated with the weight.
%
% *
%
% net.inputs{j}.size * length(net.inputWeights{i,j}.delays)
%
% These dimensions can also be obtained from the input weight properties.
%
% *
%
% net.inputWeights{i,j}.size
%
% net.LW
%
% This property defines the weight matrices of weights going to layers from other layers. It is always an Nl x Nl cell array, where Nl is the number of network layers (net.numLayers).
%
% The weight matrix for the weight going to the ith layer from the jth layer (or a null matrix []) is located at net.LW{i,j} if net.layerConnect(i,j) is 1 (or 0).
%
% The weight matrix has as many rows as the size of the layer it goes to (net.layers{i}.size). It has as many columns as the product of the size of the layer it comes from with the number of delays associated with the weight.
%
% *
%
% net.layers{j}.size * length(net.layerWeights{i,j}.delays)
%
% These dimensions can also be obtained from the layer weight properties.
%
% *
%
% net.layerWeights{i,j}.size
%
% net.b
%
% This property defines the bias vectors for each layer with a bias. It is always an Nl x 1 cell array, where Nl is the number of network layers (net.numLayers).
%
% The bias vector for the ith layer (or a null matrix []) is located at net.b{i} if net.biasConnect(i) is 1 (or 0).
%
% The number of elements in the bias vector is always equal to the size of the layer it is associated with (net.layers{i}.size).
%
% This dimension can also be obtained from the bias properties.
%
% *
%
% net.biases{i}.size
