想要自定义神经网络内部，但是网上找了很久都没有找到源代码，于是自己去搬运了一波matlab官方的源代码
代码如下
function out1 = newff(varargin)
%NEWFF Create a feed-forward backpropagation network.
%
% Obsoleted in R2010b NNET 7.0. Last used in R2010a NNET 6.0.4.
% The recommended function is feedforwardnet.
%
% Syntax
%
% net = newff(P,T,S)
% net = newff(P,T,S,TF,BTF,BLF,PF,IPF,OPF,DDF)
%
% Description
%
% NEWFF(P,T,S) takes,
% P - RxQ1 matrix of Q1 representative R-element input vectors.
% T - SNxQ2 matrix of Q2 representative SN-element target vectors.
% Si - Sizes of N-1 hidden layers, S1 to S(N-1), default = [].
% (Output layer size SN is determined from T.)
% and returns an N layer feed-forward backprop network.
%
% NEWFF(P,T,S,TF,BTF,BLF,PF,IPF,OPF,DDF) takes optional inputs,
% TFi - Transfer function of ith layer. Default is 'tansig' for
% hidden layers, and 'purelin' for output layer.
% BTF - Backprop network training function, default = 'trainlm'.
% BLF - Backprop weight/bias learning function, default = 'learngdm'.
% PF - Performance function, default = 'mse'.
% IPF - Row cell array of input processing functions.
% Default is {'fixunknowns','remconstantrows','mapminmax'}.
% OPF - Row cell array of output processing functions.
% Default is {'remconstantrows','mapminmax'}.
% DDF - Data division function, default = 'dividerand';
% and returns an N layer feed-forward backprop network.
%
% The transfer functions TF{i} can be any differentiable transfer
% function such as TANSIG, LOGSIG, or PURELIN.
%
% The training function BTF can be any of the backprop training
% functions such as TRAINLM, TRAINBFG, TRAINRP, TRAINGD, etc.
%
% *WARNING*: TRAINLM is the default training function because it
% is very fast, but it requires a lot of memory to run. If you get
% an "out-of-memory" error when training try doing one of these:
%
% (1) Slow TRAINLM training, but reduce memory requirements, by
% setting NET.efficiency.memoryReduction to 2 or more. (See HELP TRAINLM.)
% (2) Use TRAINBFG, which is slower but more memory efficient than TRAINLM.
% (3) Use TRAINRP which is slower but more memory efficient than TRAINBFG.
%
% The learning function BLF can be either of the backpropagation
% learning functions such as LEARNGD, or LEARNGDM.
%
% The performance function can be any of the differentiable performance
% functions such as MSE or MSEREG.
%
% Examples
%
% [inputs,targets] = simplefitdata;
% net = newff(inputs,targets,20);
% net = train(net,inputs,targets);
% outputs = net(inputs);
% errors = outputs - targets;
% perf = perform(net,outputs,targets)
%
% Algorithm
%
% Feed-forward networks consist of Nl layers using the DOTPROD
% weight function, NETSUM net input function, and the specified
% transfer functions.
%
% The first layer has weights coming from the input. Each subsequent
% layer has a weight coming from the previous layer. All layers
% have biases. The last layer is the network output.
%
% Each layer's weights and biases are initialized with INITNW.
%
% Adaption is done with TRAINS which updates weights with the
% specified learning function. Training is done with the specified
% training function. Performance is measured according to the specified
% performance function.
%
% See also NEWCF, NEWELM, SIM, INIT, ADAPT, TRAIN, TRAINS
% Mark Beale, 11-31-97
% Copyright 1992-2010 The MathWorks, Inc.
%disp('NEWFF is no longer recommended. FEEDFORWARD is simpler and more efficient.');
% TODO - Recommendation function NNRECOMMEND
%% Boilerplate Code - Same for all Network Functions
persistent INFO;
if (nargin < 1), error(message('nnet:Args:NotEnough')); end
in1 = varargin{1};
if ischar(in1)
switch in1
case 'info',
if isempty(INFO), INFO = get_info; end
out1 = INFO;
end
else
out1 = create_network(varargin{:});
end
%% Boilerplate Code - Same for all Network Functions
%%
function info = get_info
info.function = mfilename;
info.name = 'Feed-Forward';
info.description = nnfcn.get_mhelp_title(mfilename);
info.type = 'nntype.network_fcn';
info.version = 6.0;
%%
function net = create_network(varargin)
if nargin < 2, error(message('nnet:Args:NotEnough')), end
v1 = varargin{1};
if isa(v1,'cell'), v1 = cell2mat(v1); end
v2 = varargin{2};
if nargin > 2, v3 = varargin{3}; end
if (nargin<= 6) && (size(v1,2)==2) && (~iscell(v2)) && (size(v2,1)==1) && ((nargin<3)||iscell(v3))
nnerr.obs_use(mfilename,['See help for ' upper(mfilename) ' to update calls to the new argument list.']);
net = new_5p0(varargin{:});
else
net = new_5p1(varargin{:});
end
%=============================================================
function net = new_5p1(p,t,s,tf,btf,blf,pf,ipf,tpf,ddf)
if nargin < 2, error(message('nnet:Args:NotEnough')), end
% Defaults
if (nargin < 3), s = []; end
if (nargin < 4), tf = {}; end
if (nargin < 5), btf = 'trainlm'; end
if (nargin < 6), blf = 'learngdm'; end
if (nargin < 7), pf = 'mse'; end
if (nargin < 8), ipf = {'fixunknowns','removeconstantrows','mapminmax'}; end
if (nargin < 9), tpf = {'removeconstantrows','mapminmax'}; end
if (nargin < 10), ddf = 'dividerand'; end
% Format
if isa(p,'cell'), p = cell2mat(p); end
if isa(t,'cell'), t = cell2mat(t); end
% Error checking
if ~(isa(p,'double') || isreal(p) || islogical(t))
error(message('nnet:NNet:XNotLegal'))
end
if ~(isa(t,'double') || isreal(t) || islogical(t))
error(message('nnet:NNet:TNotLegal'))
end
if isa(s,'cell')
if (size(s,1) ~= 1)
error(message('nnet:NNet:LayerSizes'))
end
for i=1:length(s)
si = s{i};
if ~isa(si,'double') || ~isreal(si) || any(size(si) ~= 1) || any(si<1) || any(round(si) ~= si)
error(message('nnet:NNet:LayerSizes'))
end
end
s = cell2mat(s);
end
if (~isa(s,'double')) || ~isreal(s) || (size(s,1) > 1) || any(s<1) || any(round(s) ~= s)
error(message('nnet:NNet:LayerSizes'))
end
% Architecture
Nl = length(s)+1;
net = network;
net.numInputs = 1;
net.numLayers = Nl;
net.biasConnect = ones(Nl,1);
net.inputConnect(1,1) = 1;
[j,i] = meshgrid(1:Nl,1:Nl);
net.layerConnect = (j == (i-1));
net.outputConnect(Nl) = 1;
% Simulation
net.inputs{1}.processFcns = ipf;
for i=1:Nl
if (i < Nl)
net.layers{i}.size = s(i);
if (Nl == 2)
net.layers{i}.name = 'Hidden Layer';
else
net.layers{i}.name = ['Hidden Layer ' num2str(i)];
end
else
net.layers{i}.name = 'Output Layer';
end
if (length(tf) < i) || all(isnan(tf{i}))
if (i
net.layers{i}.transferFcn = 'tansig';
else
net.layers{i}.transferFcn = 'purelin';
end
else
net.layers{i}.transferFcn = tf{i};
end
end
net.outputs{Nl}.processFcns = tpf;
% Adaption
net.adaptfcn = 'adaptwb';
net.inputWeights{1,1}.learnFcn = blf;
for i=1:Nl
net.biases{i}.learnFcn = blf;
net.layerWeights{i,:}.learnFcn = blf;
end
% Training
net.trainfcn = btf;
net.dividefcn = ddf;
net.performFcn = pf;
% Initialization
net.initFcn = 'initlay';
for i=1:Nl
net.layers{i}.initFcn = 'initnw';
end
% Configuration
% Warning: Use of these properties is no longer recommended
net.inputs{1}.exampleInput = p;
net.outputs{Nl}.exampleOutput = t;
% Initialize
net = init(net);
% Plots
net.plotFcns = {'plotperform','plottrainstate','plotregression'};
%================================================================
function net = new_5p0(p,s,tf,btf,blf,pf)
% Backward compatible to NNT 5.0
if nargin < 2, error(message('nnet:Args:NotEnough')), end
% Defaults
Nl = length(s);
if nargin < 3, tf = {'tansig'}; tf = tf(ones(1,Nl)); end
if nargin < 4, btf = 'trainlm'; end
if nargin < 5, blf = 'learngdm'; end
if nargin < 6, pf = 'mse'; end
% Error checking
if isa(p,'cell') && all(size(p)==[1 1]), p = p{1,1}; end
if (~isa(p,'double')) || ~isreal(p)
error(message('nnet:NNData:XNotMatorCell1Mat'))
end
if isa(s,'cell')
if (size(s,1) ~= 1)
error(message('nnet:NNet:LayerSizes'))
end
for i=1:length(s)
si = s{i};
if ~isa(si,'double') || ~isreal(si) || any(size(si) ~= 1) || any(si<1) || any(round(si) ~= si)
error(message('nnet:NNet:LayerSizes'))
end
end
s = cell2mat(s);
end
if (~isa(s,'double')) || ~isreal(s) || (size(s,1) ~= 1) || any(s<1) || any(round(s) ~= s)
error(message('nnet:NNet:LayerSizes'))
end
% Architecture
net = network(1,Nl);
net.biasConnect = ones(Nl,1);
net.inputConnect(1,1) = 1;
[j,i] = meshgrid(1:Nl,1:Nl);
net.layerConnect = (j == (i-1));
net.outputConnect(Nl) = 1;
% Simulation
for i=1:Nl
net.layers{i}.size = s(i);
net.layers{i}.transferFcn = tf{i};
end
% Performance
net.performFcn = pf;
% Adaption
net.adaptfcn = 'adaptwb';
net.inputWeights{1,1}.learnFcn = blf;
for i=1:Nl
net.biases{i}.learnFcn = blf;
net.layerWeights{i,:}.learnFcn = blf;
end
% Training
net.trainfcn = btf;
% Initialization
net.initFcn = 'initlay';
for i=1:Nl
net.layers{i}.initFcn = 'initnw';
end
% Warning: this property is no longer recommended for use
net.inputs{1}.exampleInput = p;
net = init(net);
% Plots
net.plotFcns = {'plotperform','plottrainstate','plotregression'};
网友评论