fit_rgnn.m

介绍如何用较复杂的方法去优化对负载具有感知特性的“黑匣子”模型的神经网络控制。

function [fitness, x] = fit_rgnn(x,fcn_opts)
%[fitness, x] = fit_rgnn(x,fcn_opts)
%this calculates the fitness for a system identification problem, a
%recurrent generalized neural network.  The fitness is the negative RMS
%error, or, in the case of nan, the number of finite values before nan.
%fcn_opts.U is the input matrix (N_in x N_points)
%fcn_opts.Y is the target output vector (1 x N_points)
%fcn_opts.activation is the activation vector (N_neurons x 1)

%    Copyright Travis Wiens 2008
%
%    This program is free software: you can redistribute it and/or modify
%    it under the terms of the GNU General Public License as published by
%    the Free Software Foundation, either version 3 of the License, or
%    (at your option) any later version.
%
%    This program is distributed in the hope that it will be useful,
%    but WITHOUT ANY WARRANTY; without even the implied warranty of
%    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
%    GNU General Public License for more details.
%
%    You should have received a copy of the GNU General Public License
%    along with this program.  If not, see <http://www.gnu.org/licenses/>.
%
%    If you would like to request a commerical (or other) license, please
%    feel free to contact [e-mail]

use_mex=false;%set to true to use precompiled mex functions

N_in=size(fcn_opts.U,1);%number of inputs

W=params_to_W(x,N_in);%convert row vector of params to square W matrix

x0=zeros(size(W,1),1);%initial states

if use_mex
	[Y_hat]=rgnn_sim_mex(fcn_opts.U,W,fcn_opts.activation,x0);%perform NN calculation
else
	[Y_hat]=rgnn_sim(fcn_opts.U,W,fcn_opts.activation,x0);
end

k_skip=100;%skip first few data points

E=sqrt(mean((fcn_opts.Y(k_skip:end)-Y_hat(k_skip:end)).^2));%RMS error

fitness=-E;%maximize the negative error

if isnan(fitness)
	%if the error is "not a number" then the fitness is the number of finite values
	fitness=-(numel(Y_hat)-max(find(~isnan(Y_hat))));
end