network

matlab

创建一个神经网络。

语法

net = network
net = network(numInputs,numLayers,biasConnect,inputConnect,layerConnect,outputConnect)

获取帮助

键入help network/network

描述

network创建一个新的自定义的网络。
net = network没有任何参数创建一个没有输入，层与输出的网络。
net = network(numInputs,numLayers,biasConnect,inputConnect,layerConnect,outputConnect)

返回一个神经网络

属性

架构属性

子结构属性

函数属性

参数属性

权重与偏置值属性

其他属性

net.userdata随意保存用户自定义信息。

例子

创建一个单输入两隐藏层的网络

net = network
net.numInputs = 1
net.numLayers = 2

也可以使用net = network(1,2)来创建。

创建前向网络并且浏览属性

创建一个单输入，两隐藏层，前向网络。仅第一层有偏置。一个输入权重链接到层1从输入层1.一个层权重链接从层1到层2.层2是网络输出层并且有一个目标。
net = network(1,2,[1;0],[1; 0],[0 0; 1 0],[0 1])

可以使用下列命令显示子对象的属性：

net.inputs{1}
net.layers{1}, net.layers{2}
net.biases{1}
net.inputWeights{1,1}, net.layerWeights{2,1}
net.outputs{2}

可以通过下列代码修改子结构的属性，例如：

net.layers{1}.transferFcn = 'tansig';
net.layers{2}.transferFcn = 'logsig';

也可以浏览去权重值，例如从输入层1到层1的权重。
net.IW{1,1}

从层1到层2的权重值：
net.LW{2,1}

层1的偏置值
net.b{1}

修改输入单元1为两个元素，并且设置每个元素的值范围：
net.inputs{1}.rang = [0 1; -1 1];

对其进行仿真

p = [0.5; -0.1];
y = sim(net,p)

更多知识

• 神经网络对象属性
• 神经网络子对象属性

代码分析

function net = network(varargin)
%NETWORK 创建一个自定义的神经网络.
%
%  <a href="matlab:doc network">network</a> without arguments returns a new neural network with no
%  inputs, layers or outputs.
%
%  <a href="matlab:doc network">network</a>(numInputs,numLayers,biasConnect,inputConnect,layerConnect,
%  outputConnect) takes additional optional arguments and returns a neural
%  network with the following properties defined:
%    numInputs     - Number of inputs, 0.
%    numLayers     - Number of layers, 0.
%    biasConnect   - numLayers-by-1 Boolean vector, zeros.
%    inputConnect  - numLayers-by-numInputs Boolean matrix, zeros.
%    layerConnect  - numLayers-by-numLayers Boolean matrix, zeros.
%    outputConnect - 1-by-numLayers Boolean vector, zeros.
%
%  以下是一个一输入两层的网络被创建.
%
%      net = <a href="matlab:doc network">network</a>(1,2)
%
%  以下代码将创建一个一输入，二层，前馈网络.
%  仅仅第一层需要拥有一个bias.输入权重将链接层1. 层1的权重将链接层2
%  层2是输出层.
%
%    net = <a href="matlab:doc network">network</a>(1,2,[1;0],[1; 0],[0 0; 1 0],[0 1])
%
%  其他的子结构属性:
%
%    net.<a href="matlab:doc nnproperty.net_inputs">inputs</a>{1}
%    net.<a href="matlab:doc nnproperty.net_layers">layers</a>{1}, net.<a href="matlab:doc nnproperty.net_layers">layers</a>{2}
%    net.<a href="matlab:doc nnproperty.net_biases">biases</a>{1}
%    net.<a href="matlab:doc nnproperty.net_inputWeights">inputWeights</a>{1,1}, net.<a href="matlab:doc nnproperty.net_layerWeights">layerWeights</a>{2,1}
%    net.<a href="matlab:doc nnproperty.net_outputs">outputs</a>{2}
%
%  我们能获取权值矩阵与偏导向量：
%
%    net.<a href="matlab:doc nnproperty.net_IW">IW</a>{1,1}, net.<a href="matlab:doc nnproperty.net_IW">IW</a>{2,1}, net.<a href="matlab:doc nnproperty.net_b">b</a>{1}
%
%  我们能改变这些任意的子对象的属性。我们能同时改变两层的传输函数（激励函数）：
%
%    net.<a href="matlab:doc nnproperty.net_layers">layers</a>{1}.<a href="matlab:doc nnproperty.layer_transferFcn">transferFcn</a> = 'tansig';
%    net.<a href="matlab:doc nnproperty.net_layers">layers</a>{2}.<a href="matlab:doc nnproperty.layer_transferFcn">transferFcn</a> = 'logsig';
%
%  我们能改变一个输入元素为两个，通过设定每个元素的范围：
%
%    net.<a href="matlab:doc nnproperty.net_inputs">inputs</a>{1}.<a href="matlab:doc nnproperty.input_range">range</a> = [0 1; -1 1];
%
%  接下来我能仿真一个两输入元素向量的网络：
%
%    p = [0.5; -0.1];
%    y = net(p)
%
%  See also INIT, REVERT, SIM, ADAPT, TRAIN, VIEW.
%  Mark Beale, 11-31-97
%  Copyright 1992-2012 The MathWorks, Inc.
if (nargin == 1)
  in1 = varargin{1};
  % 如果输入的是结构，则直接转成network类
  if isa(in1,'struct')
    net = class(in1,'network');
  % 如果参数已经是network类则直接赋值
  elseif isa(in1,'network')
    net = in1;
  else
    % 单个参数构建一个网络
    net = new_network(in1);
  end
else
  % 多参数构建一个新的网络
  net = new_network(varargin{:});
end
function net = new_network(numInputs,numLayers,biasConnect,inputConnect, ...
  layerConnect,outputConnect,ignore) %#ok<INUSD>
% Defaults
if nargin < 1, numInputs = 0; end
if nargin < 2, numLayers = 0; end
if nargin < 3, biasConnect = false(numLayers,1); end
if nargin < 4, inputConnect = false(numLayers,numInputs); end
if nargin < 5, layerConnect = false(numLayers,numLayers); end
if nargin < 6, outputConnect = false(1,numLayers); end
% Checking
% TODO - Error checking
% NETWORK PROPERTIES
% Note: "Param" and "Config" properties in NETWORK and subobject
% (nnetInput, nnetOutput, nnetLayer, nnetWeight, nnetBias) properties must
% always occur directly after their associated "Fcn" properties for
% NN_STRUCT2OBJECT conversions to work properly.
% Version
net.version = '8';
% Basics
net.name = 'Custom Neural Network';
% 在下一次升级中可能会删除
% BACKWARDS COMPATIBILITY TO 8.0
% Use TRAIN(net,x,t,'Reduction',num) notation instead.
net.efficiency.cacheDelayedInputs = true;
net.efficiency.flattenTime = true;
net.efficiency.memoryReduction = 1;
% User Data
net.userdata.note = 'Put your custom network information here.';
% Sizes
net.numInputs = 0;
net.numLayers = 0;
net.numOutputs = 0;
net.numInputDelays = 0;
net.numLayerDelays = 0;
net.numFeedbackDelays = 0;
net.numWeightElements = 0;
net.sampleTime = 1;
% Connections
net.biasConnect = false(0,1);
net.inputConnect = false(0,0);
net.layerConnect = false(0,0);
net.outputConnect = false(1,0);
% 子对象
net.inputs = cell(0,1);
net.layers = cell(0,1);
net.biases = cell(0,1);
net.outputs = cell(1,0);
net.inputWeights = cell(0,0);
net.layerWeights = cell(0,0);
% 函数和参数
net.adaptFcn = '';
net.adaptParam = struct;
net.divideFcn = '';
net.divideParam = struct;
net.divideMode = 'sample';
net.initFcn = 'initlay';
net.performFcn = 'mse';
net.performParam = mse('defaultParam');
net.plotFcns = cell(1,0);
net.plotParams = cell(1,0);
net.derivFcn = 'defaultderiv';
net.trainFcn = '';
net.trainParam = nnetParam;
% 权值 & 偏差值
net.IW = cell(0,0);
net.LW = cell(0,0);
net.b = cell(0,1);
% 隐藏属性
net.revert.IW = {};    % 输入层权重
net.revert.LW = {};    % 隐藏层权重
net.revert.b = {};     % 偏差
% 要删掉的属性
% NNET 6.0 Compatibility
net.gradientFcn = ''; % Obsolete
net.gradientParam = struct; % Obsolete
% CLASS
% 生成一个网络类型
net = class(net,'network');
% ARCHITECTURE
net = setnet(net,'numInputs',numInputs);
net = setnet(net,'numLayers',numLayers);
net = setnet(net,'biasConnect',biasConnect);
net = setnet(net,'inputConnect',inputConnect);
net = setnet(net,'layerConnect',layerConnect);
net = setnet(net,'outputConnect',outputConnect);
function net = setnet(net,field,value)
subscripts.type = '.';
subscripts.subs = field;
net = subsasgn(net,subscripts,value);