progress_snn.m

来自「神经网络的工具箱, 神经网络的工具箱,」· M 代码 · 共 62 行

function r = progress_snn(net, data, result)
%PROGRESS_SNN Show plot of training progress.
%
% Syntax
%
%   progress_snn(net, [], tr_info)
%
%   progressfcn_struct = progress_snn('pdefaults')
%
% Description
%
%   PROGRESS_SNN takes
%     net     - the network in training.
%     tr_info - a structure containing information about the training
%               process.
%   and plots a figure showing the evoluation of the cost function.  
%
%   The field 'trainFcn.progressFcn' of 'net' must contain a structure
%   with the progress function paramters. The default parameters for
%   this function can be set with:
%
%     net.trainFcn.progressFcn = progress_snn('pdefaults')
%
%   or can be set manually. The parameters are:
%
%     net.trainFcn.progressFcn.name   - 'progress_snn'
%     net.trainFcn.progressFcn.show   - a number indicating after how
%                                       many training epochs the plot 
%                                       must be updated.
%
% See also
%
%    PROGRESSFCN_STRUCT_SNN
%


if isstr(net)
   switch (net)
     case 'pdefaults',
          progressFcn.name = 'progress_snn';
	  progressFcn.show = 25;
          r = progressFcn;
     otherwise,
          error('Unrecognized code.')
   end
   return
end

if (~rem(result.epoch,net.trainFcn.progressFcn.show) | (length(result.stop)))
   stdout_snn(upper(net.trainFcn.name));
   if isfinite(net.trainFcn.epochs) 
      stdout_snn(', Epoch %g/%g', result.epoch, net.trainFcn.epochs); 
   end
   if isfinite(net.trainFcn.goal) 
      stdout_snn(', %s %g/%g', upper(net.costFcn.name), ...
               result.tr.perf(result.epoch+1), net.trainFcn.goal);
   end
   stdout_snn('\n');
   plotperf_snn(result.tr, net.trainFcn.goal, upper(net.trainFcn.name), result.epoch);
end