1.matlab BP神经网络的训练算法中训练函数（traingdm 、trainlm、webcrack源码trainbr）的龙轩导航 源码实现过程及相应的VC源代码

matlab BP神经网络的训练算法中训练函数（traingdm 、trainlm、163邮箱登录源码trainbr）的ubuntu 查看linux源码实现过程及相应的VC源代码

       VC源代码？你很搞笑嘛。。比利猫跳跃 源码

       给你trainlm的m码

       function [out1,out2] = trainlm(varargin)

       %TRAINLM Levenberg-Marquardt backpropagation.

       %

       % <a href="matlab:doc trainlm">trainlm</a> is a network training function that updates weight and

       % bias states according to Levenberg-Marquardt optimization.

       %

       % <a href="matlab:doc trainlm">trainlm</a> is often the fastest backpropagation algorithm in the toolbox,

       % and is highly recommended as a first choice supervised algorithm,

       % although it does require more memory than other algorithms.

       %

       % [NET,TR] = <a href="matlab:doc trainlm">trainlm</a>(NET,X,T) takes a network NET, input data X

       % and target data T and returns the network after training it, and a

       % a training record TR.

       %

       % [NET,TR] = <a href="matlab:doc trainlm">trainlm</a>(NET,X,T,Xi,Ai,EW) takes additional optional

       % arguments suitable for training dynamic networks and training with

       % error weights. Xi and Ai are the initial input and layer delays states

       % respectively and EW defines error weights used to indicate

       % the relative importance of each target value.

       %

       % Training occurs according to training parameters, with default values.

       % Any or all of these can be overridden with parameter name/value argument

       % pairs appended to the input argument list, or by appending a structure

       % argument with fields having one or more of these names.

       % show Epochs between displays

       % showCommandLine 0 generate command line output

       % showWindow 1 show training GUI

       % epochs Maximum number of epochs to train

       % goal 0 Performance goal

       % max_fail 5 Maximum validation failures

       % min_grad 1e- Minimum performance gradient

       % mu 0. Initial Mu

       % mu_dec 0.1 Mu decrease factor

       % mu_inc Mu increase factor

       % mu_max 1e Maximum Mu

       % time inf Maximum time to train in seconds

       %

       % To make this the default training function for a network, and view

       % and/or change parameter settings, use these two properties:

       %

       % net.<a href="matlab:doc nnproperty.net_trainFcn">trainFcn</a> = 'trainlm';

       % net.<a href="matlab:doc nnproperty.net_trainParam">trainParam</a>

       %

       % See also trainscg, feedforwardnet, narxnet.

       % Mark Beale, --, ODJ //

       % Updated by Orlando De Jes鷖, Martin Hagan, Dynamic Training 7--

       % Copyright - The MathWorks, Inc.

       % $Revision: 1.1.6..2.2 $ $Date: // :: $

       %% =======================================================

       % BOILERPLATE_START

       % This code is the same for all Training Functions.

        persistent INFO;

        if isempty(INFO), INFO = get_info; end

        nnassert.minargs(nargin,1);

        in1 = varargin{ 1};

        if ischar(in1)

        switch (in1)

        case 'info'

        out1 = INFO;

        case 'check_param'

        nnassert.minargs(nargin,2);

        param = varargin{ 2};

        err = nntest.param(INFO.parameters,param);

        if isempty(err)

        err = check_param(param);

        end

        if nargout > 0

        out1 = err;

        elseif ~isempty(err)

        nnerr.throw('Type',err);

        end

        otherwise,

        try

        out1 = eval(['INFO.' in1]);

        catch me, nnerr.throw(['Unrecognized first argument: ''' in1 ''''])

        end

        end

        return

        end

        nnassert.minargs(nargin,2);

        net = nn.hints(nntype.network('format',in1,'NET'));

        oldTrainFcn = net.trainFcn;

        oldTrainParam = net.trainParam;

        if ~strcmp(net.trainFcn,mfilename)

        net.trainFcn = mfilename;

        net.trainParam = INFO.defaultParam;

        end

        [args,param] = nnparam.extract_param(varargin(2:end),net.trainParam);

        err = nntest.param(INFO.parameters,param);

        if ~isempty(err), nnerr.throw(nnerr.value(err,'NET.trainParam')); end

        if INFO.isSupervised && isempty(net.performFcn) % TODO - fill in MSE

        nnerr.throw('Training function is supervised but NET.performFcn is undefined.');

        end

        if INFO.usesGradient && isempty(net.derivFcn) % TODO - fill in

        nnerr.throw('Training function uses derivatives but NET.derivFcn is undefined.');

        end

        if net.hint.zeroDelay, nnerr.throw('NET contains a zero-delay loop.'); end

        [X,T,Xi,Ai,EW] = nnmisc.defaults(args,{ },{ },{ },{ },{ 1});

        X = nntype.data('format',X,'Inputs X');

        T = nntype.data('format',T,'Targets T');

        Xi = nntype.data('format',Xi,'Input states Xi');

        Ai = nntype.data('format',Ai,'Layer states Ai');

        EW = nntype.nndata_pos('format',EW,'Error weights EW');

        % Prepare Data

        [net,data,tr,~,err] = nntraining.setup(net,mfilename,X,Xi,Ai,T,EW);

        if ~isempty(err), nnerr.throw('Args',err), end

        % Train

        net = struct(net);

        fcns = nn.subfcns(net);

        [net,tr] = train_network(net,tr,data,fcns,param);

        tr = nntraining.tr_clip(tr);

        if isfield(tr,'perf')

        tr.best_perf = tr.perf(tr.best_epoch+1);

        end

        if isfield(tr,'vperf')

        tr.best_vperf = tr.vperf(tr.best_epoch+1);

        end

        if isfield(tr,'tperf')

        tr.best_tperf = tr.tperf(tr.best_epoch+1);

        end

        net.trainFcn = oldTrainFcn;

        net.trainParam = oldTrainParam;

        out1 = network(net);

        out2 = tr;

       end

       % BOILERPLATE_END

       %% =======================================================

       % TODO - MU => MU_START

       % TODO - alternate parameter names (i.e. MU for MU_START)

       function info = get_info()

        info = nnfcnTraining(mfilename,'Levenberg-Marquardt',7.0,true,true,...

        [ ...

        nnetParamInfo('showWindow','Show Training Window Feedback','nntype.bool_scalar',true,...

        'Display training window during training.'), ...

        nnetParamInfo('showCommandLine','Show Command Line Feedback','nntype.bool_scalar',false,...

        'Generate command line output during training.'), ...

        nnetParamInfo('show','Command Line Frequency','nntype.strict_pos_int_inf_scalar',,...

        'Frequency to update command line.'), ...

        ...

        nnetParamInfo('epochs','Maximum Epochs','nntype.pos_int_scalar',,...

        'Maximum number of training iterations before training is stopped.'), ...

        nnetParamInfo('time','Maximum Training Time','nntype.pos_inf_scalar',inf,...

        'Maximum time in seconds before training is stopped.'), ...

        ...

        nnetParamInfo('goal','Performance Goal','nntype.pos_scalar',0,...

        'Performance goal.'), ...

        nnetParamInfo('min_grad','Minimum Gradient','nntype.pos_scalar',1e-5,...

        'Minimum performance gradient before training is stopped.'), ...

        nnetParamInfo('max_fail','Maximum Validation Checks','nntype.strict_pos_int_scalar',6,...

        'Maximum number of validation checks before training is stopped.'), ...

        ...

        nnetParamInfo('mu','Mu','nntype.pos_scalar',0.,...

        'Mu.'), ...

        nnetParamInfo('mu_dec','Mu Decrease Ratio','nntype.real_0_to_1',0.1,...

        'Ratio to decrease mu.'), ...

        nnetParamInfo('mu_inc','Mu Increase Ratio','nntype.over1',,...

        'Ratio to increase mu.'), ...

        nnetParamInfo('mu_max','Maximum mu','nntype.strict_pos_scalar',1e,...

        'Maximum mu before training is stopped.'), ...

        ], ...

        [ ...

        nntraining.state_info('gradient','Gradient','continuous','log') ...

        nntraining.state_info('mu','Mu','continuous','log') ...

        nntraining.state_info('val_fail','Validation Checks','discrete','linear') ...

        ]);

       end

       function err = check_param(param)

        err = '';

       end

       function [net,tr] = train_network(net,tr,data,fcns,param)

        % Checks

        if isempty(net.performFcn)

        warning('nnet:trainlm:Performance',nnwarning.empty_performfcn_corrected);

        net.performFcn = 'mse';

        net.performParam = mse('defaultParam');

        tr.performFcn = net.performFcn;

        tr.performParam = net.performParam;

        end

        if isempty(strmatch(net.performFcn,{ 'sse','mse'},'exact'))

        warning('nnet:trainlm:Performance',nnwarning.nonjacobian_performfcn_replaced);

        net.performFcn = 'mse';

        net.performParam = mse('defaultParam');

        tr.performFcn = net.performFcn;

        tr.performParam = net.performParam;

        end

        % Initialize

        startTime = clock;

        original_net = net;

        [perf,vperf,tperf,je,jj,gradient] = nntraining.perfs_jejj(net,data,fcns);

        [best,val_fail] = nntraining.validation_start(net,perf,vperf);

        WB = getwb(net);

        lengthWB = length(WB);

        ii = sparse(1:lengthWB,1:lengthWB,ones(1,lengthWB));

        mu = param.mu;

        % Training Record

        tr.best_epoch = 0;

        tr.goal = param.goal;

        tr.states = { 'epoch','time','perf','vperf','tperf','mu','gradient','val_fail'};

        % Status

        status = ...

        [ ...

        nntraining.status('Epoch','iterations','linear','discrete',0,param.epochs,0), ...

        nntraining.status('Time','seconds','linear','discrete',0,param.time,0), ...

        nntraining.status('Performance','','log','continuous',perf,param.goal,perf) ...

        nntraining.status('Gradient','','log','continuous',gradient,param.min_grad,gradient) ...

        nntraining.status('Mu','','log','continuous',mu,param.mu_max,mu) ...

        nntraining.status('Validation Checks','','linear','discrete',0,param.max_fail,0) ...

        ];

        nn_train_feedback('start',net,status);

        % Train

        for epoch = 0:param.epochs

        % Stopping Criteria

        current_time = etime(clock,startTime);

        [userStop,userCancel] = nntraintool('check');

        if userStop, tr.stop = 'User stop.'; net = best.net;

        elseif userCancel, tr.stop = 'User cancel.'; net = original_net;

        elseif (perf <= param.goal), tr.stop = 'Performance goal met.'; net = best.net;

        elseif (epoch == param.epochs), tr.stop = 'Maximum epoch reached.'; net = best.net;

        elseif (current_time >= param.time), tr.stop = 'Maximum time elapsed.'; net = best.net;

        elseif (gradient <= param.min_grad), tr.stop = 'Minimum gradient reached.'; net = best.net;

        elseif (mu >= param.mu_max), tr.stop = 'Maximum MU reached.'; net = best.net;

        elseif (val_fail >= param.max_fail), tr.stop = 'Validation stop.'; net = best.net;

        end

        % Feedback

        tr = nntraining.tr_update(tr,[epoch current_time perf vperf tperf mu gradient val_fail]);

        nn_train_feedback('update',net,status,tr,data, ...

        [epoch,current_time,best.perf,gradient,mu,val_fail]);

        % Stop

        if ~isempty(tr.stop), break, end

        % Levenberg Marquardt

        while (mu <= param.mu_max)

        % CHECK FOR SINGULAR MATRIX

        [msgstr,msgid] = lastwarn;

        lastwarn('MATLAB:nothing','MATLAB:nothing')

        warnstate = warning('off','all');

        dWB = -(jj+ii*mu) \ je;

        [~,msgid1] = lastwarn;

        flag_inv = isequal(msgid1,'MATLAB:nothing');

        if flag_inv, lastwarn(msgstr,msgid); end;

        warning(warnstate)

        WB2 = WB + dWB;

        net2 = setwb(net,WB2);

        perf2 = nntraining.train_perf(net2,data,fcns);

        % TODO - possible speed enhancement

        % - retain intermediate variables for Memory Reduction = 1

        if (perf2 < perf) && flag_inv

        WB = WB2; net = net2;

        mu = max(mu*param.mu_dec,1e-);

        break

        end

        mu = mu * param.mu_inc;

        end

        % Validation

        [perf,vperf,tperf,je,jj,gradient] = nntraining.perfs_jejj(net,data,fcns);

        [best,tr,val_fail] = nntraining.validation(best,tr,val_fail,net,perf,vperf,epoch);

        end

       end