file: ~/courses/spring2002/434/perceptron_setup.txt    RWN   02/07-10/02

The steps to follow, with an example, using MatLab NNET programs for 
handling perceptrons:

1. setup the network:
	net1 = newp([0 1; -2 1; 3 9; -0.8 0.8; -3 3],2);
		This uses the new_perceptron call to set up a
	double neuron perceptron with five inputs, for which 
	weights will be generated; the five entries in an input
	vector are in the five ranges indicated.
2. setup the training set of inputs P and corresponding outputs T: 
	P1=[0 0.1 0.2 0.8; 1 -1 -2 2; 3 4 5 6; -0.6 -0.5 0.5 0.6; ...
	-3 -2 -1 2]
	T1=[1 0 0 1; 0 1 1 1]
3. check the initial weight and bias values:
	net1.IW{1,1}
	net1.b{1}
		These should return two rows each all with zeros
3. next check, by simulating the net, that P1 works with net1:
	Y=sim(net1,P1)
		This returns Y =
			     1     1     1     1
			     1     1     1     1
		These values are as expected since the inputs
	to the hardlimit activation functions are all zero.
4. next train via the adapt function on P1 & T1:
	first set how many epochs=passes to use
		net1.adaptParam.passes = 10;
	then adapt the weights and bias
		net1= adapt(net1,P1,T1);
	then check if T1 results as the output now 	
		Y = sim(net1,P1)
	Note that this returns Y that is T1.
5. To see the final weights and bias repeat 3.:
	net1.IW{1,1}
ans =
    0.5000    6.0000         0         0    2.0000
    1.1000   -4.0000    6.0000    2.4000    8.0000
	net1.b{1}
ans =
     0
     0
6. try out on a new input vector:
	P2=[0.3; -1.2; 4; 0.6; 2];
	Y=sim(net1,P2)
		This returns
Y =
     0
     1
7. next we use the train program to do the same thing:
	First reset the weights and bias to zero
		net1 = init(net1);
	to check that zero results use 
		net1.IW{1,1}
		Y = sim(net1,P2)
	This returns:
	    Y =
		1     1     1     1
		1     1     1     1
	Set the upper limit of the number of epochs=passes
		net1.trainParam.epochs = 20;
	Now do the training	
		net1 = train(net1,P2,T2);
		Y = sim(net1,P2)
	The result is
		TRAINC, Epoch 0/20
		TRAINC, Epoch 4/20
		TRAINC, Performance goal met.
	    Y =
		1      0     0     1
		0     1     1     1

	along with a plot of the performance showing the error behavior. 
8. With this perceptron defined it can be put into simulink.
	For that the weight vectors and biases can be set into gain
	blocks which call their gains from the "command window." For
	the weight-gain for the first neuron one can enter the gain as
	W1net1 and enter in the command window
		W1net1=net1.IW{1,1}(1,:)
	and then for the second neuron
		W2net1=net1.IW{1,1}(2,:)
	while for the gain of bias blocks
		B1net1=net1.b{1}(1)
		B2net1=net1.b{1}(2)

4'. To see what is available for this net1:
	net1= adapt(net1,P1,T1)
		entered without the ;
	This returns: 
net1 =

    Neural Network object:

    architecture:

         numInputs: 1
         numLayers: 1
       biasConnect: [1]
      inputConnect: [1]
      layerConnect: [0]
     outputConnect: [1]
     targetConnect: [1]

        numOutputs: 1  (read-only)
        numTargets: 1  (read-only)
    numInputDelays: 0  (read-only)
    numLayerDelays: 0  (read-only)

    subobject structures:

            inputs: {1x1 cell} of inputs
            layers: {1x1 cell} of layers
           outputs: {1x1 cell} containing 1 output
           targets: {1x1 cell} containing 1 target
            biases: {1x1 cell} containing 1 bias
      inputWeights: {1x1 cell} containing 1 input weight
      layerWeights: {1x1 cell} containing no layer weights

    functions:

          adaptFcn: 'trains'
           initFcn: 'initlay'
        performFcn: 'mae'
          trainFcn: 'trainc'

    parameters:

        adaptParam: .passes
         initParam: (none)
      performParam: (none)
        trainParam: .epochs, .goal, .show, .time

    weight and bias values:

                IW: {1x1 cell} containing 1 input weight matrix
                LW: {1x1 cell} containing no layer weight matrices
                 b: {1x1 cell} containing 1 bias vector

    other:

          userdata: (user stuff)
4''. Now one can find various data: 
	net1.trainParam.epochs

ans =

   20

	net1.trainParam.goal

ans =

     0
	net1.layers{1}.transferFcn
ans =

	hardlim