wolffd@0: function [e, edata, eprior] = mlperr(net, x, t)
wolffd@0: %MLPERR	Evaluate error function for 2-layer network.
wolffd@0: %
wolffd@0: %	Description
wolffd@0: %	E = MLPERR(NET, X, T) takes a network data structure NET together
wolffd@0: %	with a matrix X of input vectors and a matrix T of target vectors,
wolffd@0: %	and evaluates the error function E. The choice of error function
wolffd@0: %	corresponds to the output unit activation function. Each row of X
wolffd@0: %	corresponds to one input vector and each row of T corresponds to one
wolffd@0: %	target vector.
wolffd@0: %
wolffd@0: %	[E, EDATA, EPRIOR] = MLPERR(NET, X, T) additionally returns the data
wolffd@0: %	and prior components of the error, assuming a zero mean Gaussian
wolffd@0: %	prior on the weights with inverse variance parameters ALPHA and BETA
wolffd@0: %	taken from the network data structure NET.
wolffd@0: %
wolffd@0: %	See also
wolffd@0: %	MLP, MLPPAK, MLPUNPAK, MLPFWD, MLPBKP, MLPGRAD
wolffd@0: %
wolffd@0: 
wolffd@0: %	Copyright (c) Ian T Nabney (1996-2001)
wolffd@0: 
wolffd@0: % Check arguments for consistency
wolffd@0: errstring = consist(net, 'mlp', x, t);
wolffd@0: if ~isempty(errstring);
wolffd@0:   error(errstring);
wolffd@0: end
wolffd@0: [y, z, a] = mlpfwd(net, x);
wolffd@0: 
wolffd@0: switch net.outfn
wolffd@0: 
wolffd@0:   case 'linear'        % Linear outputs
wolffd@0:     edata = 0.5*sum(sum((y - t).^2));
wolffd@0: 
wolffd@0:   case 'logistic'      % Logistic outputs
wolffd@0:     % Ensure that log(1-y) is computable: need exp(a) > eps
wolffd@0:     maxcut = -log(eps);
wolffd@0:     % Ensure that log(y) is computable
wolffd@0:     mincut = -log(1/realmin - 1);
wolffd@0:     a = min(a, maxcut);
wolffd@0:     a = max(a, mincut);
wolffd@0:     y = 1./(1 + exp(-a));
wolffd@0:     edata = - sum(sum(t.*log(y) + (1 - t).*log(1 - y)));
wolffd@0: 
wolffd@0:   case 'softmax'       % Softmax outputs
wolffd@0:     nout = size(a,2);
wolffd@0:     % Ensure that sum(exp(a), 2) does not overflow
wolffd@0:     maxcut = log(realmax) - log(nout);
wolffd@0:     % Ensure that exp(a) > 0
wolffd@0:     mincut = log(realmin);
wolffd@0:     a = min(a, maxcut);
wolffd@0:     a = max(a, mincut);
wolffd@0:     temp = exp(a);
wolffd@0:     y = temp./(sum(temp, 2)*ones(1,nout));
wolffd@0:     % Ensure that log(y) is computable
wolffd@0:     y(y<realmin) = realmin;
wolffd@0:     edata = - sum(sum(t.*log(y)));
wolffd@0: 
wolffd@0:   otherwise
wolffd@0:     error(['Unknown activation function ', net.outfn]);  
wolffd@0: end
wolffd@0: [e, edata, eprior] = errbayes(net, edata);