net_struct_snn.m

神经网络的工具箱, 神经网络的工具箱,

function net = net_struct_snn (s, tf, trainFcn)
%NET_STRUCT_SNN Create a feed-forward network in net_struct format.
%
%  Syntax
%
%   net_struct = net_struct_snn([S0 S1 ... SL], {TF1 TF2 ... TFL})
%   net_struct = net_struct_snn([S0 S1 ... SL], {TF1 TF2 ... TFL}, trainFcn)
%   net_struct = net_struct_snn(...
%                   [S0 S1 ... SL], {TF1 TF2 ... TFL}, trainFcn, costFcn)
%
%  Description
%
%   NET_STRUCT_SNN takes
%     Sl  - Number of units in the l'th layer, for L layers.
%     TFl - Transfer function for the l'th layer. 
%     trainFcn - The network training function (default TRAINLM_SNN).
%     costFcn - The cost fucntion which is minimized in training
%               (default WCF_SNN). 
%   and returns a net_struct that can be used by TRAIN_SNN.
%
%   In the SNN Feed Forward Neural Network toolbox, all parameters
%   regarding the network are stored in a net_struct containing the
%   following fields:
%
%   net_struct.numLayers   - number of layers in the network (L)
%   net_sturct.transferFcn - 1 x L cell matrix of transfer function names. 
%   net_struct.biases      - 1 x L cell matrix of N(l-1) x 1 matrices of
%                            unit biases.
%   net_struct.weights     - 1 x L cell matrix of N(l-1) x Nl matrices of 
%                            connection weights.
%   net_struct.trainFcn    - trainfcn_struct with trainings
%                            algorithm related parameters.
%   net_struct.costFcn    - costfcn_struct with cost function
%                            related parameters.
%   (L = number of layers, Nl = number of units in l'th layer)
%
%  Example
%
%   net = net_struct_snn([11 5 6], {'tansigtf_snn' 'lintf_snn'}, 'trainlm_snn');
%
%  See also
%
%    TRAIN_SNN, TRANSFERFCN_SNN, TRAINFCN_SNN, COSTFCN_SNN,
%    TRAINFCN_STRUCT_SNN, COSTFCN_STRUCT_SNN
%

if (nargin < 4)
   costFcn = 'wcf_snn'; 
end

if (nargin < 3)
   trainFcn = 'trainlm_snn';
end

net.numLayers = size(s,2) - 1;
if (size(tf,2) ~= net.numLayers)
   error('TF and S must be equal is size');
end

net.transferFcn = tf;
for i = 1:net.numLayers
   net.biases{i} = 0.1*randn(s(1,i+1),1);
   net.weights{i} = 0.1*randn(s(1,i+1), s(1,i)); 
end

%#function traingd_snn traincgp_snn trainlm_snn
net.trainFcn = feval(trainFcn, 'pdefaults');
%#function wcf_snn
net.costFcn = feval(costFcn, 'pdefaults');