traincgp_snn.m

来自「神经网络的工具箱, 神经网络的工具箱,」· M 代码 · 共 213 行

function [net, result] = traincgp_snn(net, dataLV, dataVV, dataTV)
%TRAINCGP_SNN Conjugate gradient training (Polak - Ribiere).
%
%  Syntax
%
%   [net, tr_info] = traincgp_snn(net, dataLV)
%   [net, tr_info] = traincgp_snn(net, dataLV, dataVV)
%   [net, tr_info] = traincgp_snn(net, dataLV, dataVV, dataTV)
%
%   trainfcn_struct = traincgp_snn('pdefaults')
%
%  Description
%
%   TRAINCGP_SNN is a network training function that updates weight and
%   bias values according to a conjugate gradient algorithm as
%   proposed by Polak and Ribiere.
%
%   TRAINCGP_SNN takes
%     net    - a net_struct containing the initial network.
%     dataLV - training data set.
%     dataVV - validation data set (optional).
%     dataTV - test data set (optional).
%   and returns
%     net     - the trained network.
%     tr_info - a structure containing information about the
%               training process.
%
%   The field 'trainFcn' of 'net' must contain a structure with the
%   training parameters. The default parameters can be set with
%
%     net.trainFcn = traincgp_snn('pdefaults')
%
%   or can be set manually. The parameters are:
%
%     net.trainFcn.name     - 'traincgp_snn'
%     net.trainFcn.epochs   - maximum number of training epochs
%     net.trainFcn.goal     - training goal for cost function
%     net.trainFcn.max_fail - maximum number of epochs in which cost
%                             on validation set increases.
%     net.trainFcn.max_suc_fail - maximum number of succesive epochs
%                                 in which cost on validation set
%                                 increases.
%     net.trainFcn.tol      - tolerance parameter. 
%     net.trainFcn.searchFcn - 'srchbre_snn' (or another line search function)
%     net.trainFcn.progressFcn - progressfcn_struct containing parameters
%                                for showing progress of training
%     net.trainFcn.time     - maximum time for training.
%
%  See also
%
%   NET_STRUCT_SNN, PROGRESSFCN_STRUCT_SNN
%

% FUNCTION INFO
% =============
if isstr(net)
  switch (net)
    case 'pnames',
          net = {'name';'epochs';'goal';'lr';'max_fail';'min_grad';...
            'show';'time'};
    case 'pdefaults',
	  trainFcn.name = 'traincgp_snn';
          trainFcn.epochs = 1000;
          trainFcn.goal = 0;
          trainFcn.max_fail = 20;
          trainFcn.max_suc_fail = 10;
          trainFcn.tol = 1e-7;
          trainFcn.searchFcn = 'srchbre_snn';
          trainFcn.progressFcn = progress_snn('pdefaults');
          trainFcn.time = inf;
          net = trainFcn;
    otherwise,
          error('Unrecognized code.')
  end   
  return
end     


        
% CALCULATION
% ===========

if (nargin <3)
   doValidation = 0;
else
   doValidation = 1;
end

if (nargin <4)
   doTest = 0;
else
   doTest = 1;
end

% Constants
epochs = net.trainFcn.epochs;
goal = net.trainFcn.goal;
time = net.trainFcn.time;
max_fail = net.trainFcn.max_fail;
max_suc_fail = net.trainFcn.max_suc_fail;
tol = net.trainFcn.tol;

% Initialize
result.stop = '';
startTime = clock;
X = getx_snn(net);
result.tr = tr_struct_snn(epochs, 'perf', 'vperf', 'tperf');
%#function se_snn
%#function relerr_snn
%#function loglikelihood_snn
%#function crosslogistic_snn
%#function crossentropy_snn
result.tr.perf(1) = feval(net.costFcn.name, net, dataLV);
fX = result.tr.perf(1);
if (doValidation)
   result.best_X = X;
   result.tr.vperf(1) = feval(net.costFcn.name, net, dataVV);
   result.vperf_min = result.tr.vperf(1);
   result.num_fail = 0;
   result.num_suc_fail = 0;
end
if (doTest)
   result.tr.tperf(1) = feval(net.costFcn.name, net, dataTV);
end
xi = gradff_snn(net, dataLV);
g = -xi;
xi = g;
h = g;
gg = 0;

% Train
for epoch = 1:epochs
 
    % Train with Conjugate Gradient (Polak Ribiere) 
    %#function srchbre_snn
    [X, fmin] = feval(net.trainFcn.searchFcn, ...
                     net.costFcn.name, X, xi, net, dataLV);
    if (2*abs(fmin - fX) <= tol*(abs(fmin)+abs(fX)) + 1e-10)
       result.stop = 'Stop; minimum reached for costFcn, performance goal was not met.';
    else
       fX = fmin;
       xi = gradff_snn(X, net, dataLV);
       gg = g'*g;
       dgg = (xi+g)'*xi;
       if (~gg)
          result.stop = 'Stop; gradient for costFcn = 0, performance goal was not met.';
       else
          gam = dgg/gg;
	  g = -xi;
	  xi = g + gam*h;
	  h = xi;
       end
    end
    net = setx_snn(net, X); 

    % Save performance
    epochPlus1 = epoch + 1;
    result.tr.perf(epochPlus1) = fmin;
        
    % Validation
    if (doValidation)
       %#function wcf_snn
       result.tr.vperf(epochPlus1) = feval(net.costFcn.name, net, dataVV);
       if (result.tr.vperf(epochPlus1) < result.vperf_min)
          result.vperf_min = result.tr.vperf(epochPlus1);
          result.best_X = X;
          result.num_fail = 0;
       elseif (result.tr.vperf(epochPlus1) > result.vperf_min)
          result.num_fail = result.num_fail + 1;
       end
       if (result.tr.vperf(epochPlus1) > result.tr.vperf(epoch))
          result.num_suc_fail = result.num_suc_fail + 1;
       else
          result.num_suc_fail = 0;
       end
    end
    if (doTest)
       %#function wcf_snn
       result.tr.tperf(epochPlus1) = feval(net.costFcn.name, net, dataTV);
    end
    
    % Stopping criteria
    result.runtime = etime(clock, startTime);
    if (result.tr.perf(epochPlus1) <= goal)
       result.stop = 'Performance goal met.';
    elseif (epoch == epochs)
       result.stop = 'Maximum epoch reached, performance goal was not met.';
    elseif (result.runtime > time)
       result.stop = 'Maximum time elapsed, performance goal was not met.';
    elseif (doValidation) & (result.num_fail > max_fail)
       result.stop = 'Validation stop; maximum total failures';
       net = setx_snn(net, result.best_X);
    elseif (doValidation) & (result.num_suc_fail > max_suc_fail)
       result.stop = 'Validation stop; maximum successive failures';
       net = setx_snn(net, result.best_X);
    end
    result.epoch = epoch;
    if length(result.stop) 
	%#function progress_snn
        feval(net.trainFcn.progressFcn.name, net, [], result);
       stdout_snn('%s, %s\n', upper(net.trainFcn.name), result.stop); 
       break; 
    end

    % Show progress
    %#function progress_snn
    feval(net.trainFcn.progressFcn.name, net, [], result);
end

% Finish
result.tr = cliptr_snn(result.tr, epoch);