trainsm.c

nnToolKit 神经网络工具包是基于 MATLAB 神经网络工具箱自行开发的一组神经网络算法函数库

     * max_nb = max(max(m));
     */
    mlfAssign(
      &max_nb,
      mlfMax(NULL, mlfMax(NULL, mclVa(m, "m"), NULL, NULL), NULL, NULL));
    /*
     * 
     * % SIZES
     * [R,Q] = size(p);
     */
    mlfSize(mlfVarargout(&R, &Q, NULL), mclVa(p, "p"), NULL);
    /*
     * [S,R] = size(w);
     */
    mlfSize(mlfVarargout(&S, &R, NULL), mclVa(w, "w"), NULL);
    /*
     * 
     * % PLOTTING
     * message = sprintf('TRAINSM: %%g/%g epochs, neighborhood = %%g, lr = %%g.\n',max_pres);
     */
    mlfAssign(
      &message,
      mlfSprintf(NULL, _mxarray9_, mclVv(max_pres, "max_pres"), NULL));
    /*
     * fprintf(message,0,max_nb,init_lr)
     */
    mclPrintAns(
      &ans,
      mlfNFprintf(
        0,
        mclVv(message, "message"),
        _mxarray11_,
        mclVv(max_nb, "max_nb"),
        mclVv(init_lr, "init_lr"),
        NULL));
    /*
     * if R > 1 & S < 100
     */
    {
        mxArray * a_ = mclInitialize(mclGt(mclVv(R, "R"), _mxarray13_));
        if (mlfTobool(a_)
            && mlfTobool(mclAnd(a_, mclLt(mclVv(S, "S"), _mxarray12_)))) {
            mxDestroyArray(a_);
            /*
             * plotsm(w,m);
             */
            mlfPlotsm(mclVa(w, "w"), mclVa(m, "m"));
        } else {
            mxDestroyArray(a_);
        }
    /*
     * end
     */
    }
    /*
     * 
     * lr_x = (1/max_pres)^(1/sqrt(max_pres*1000));
     */
    mlfAssign(
      &lr_x,
      mclMpower(
        mclMrdivide(_mxarray13_, mclVv(max_pres, "max_pres")),
        mclMrdivide(
          _mxarray13_,
          mlfSqrt(mclMtimes(mclVv(max_pres, "max_pres"), _mxarray14_)))));
    /*
     * base_lr = init_lr / (1+lr_x);
     */
    mlfAssign(
      &base_lr,
      mclMrdivide(
        mclVv(init_lr, "init_lr"), mclPlus(_mxarray13_, mclVv(lr_x, "lr_x"))));
    /*
     * nb_x = (1/max_nb)^(10/max_pres);
     */
    mlfAssign(
      &nb_x,
      mclMpower(
        mclMrdivide(_mxarray13_, mclVv(max_nb, "max_nb")),
        mclMrdivide(_mxarray15_, mclVv(max_pres, "max_pres"))));
    /*
     * z = ones(S,1)*(1/S);
     */
    mlfAssign(
      &z,
      mclMtimes(
        mlfOnes(mclVv(S, "S"), _mxarray13_, NULL),
        mclMrdivide(_mxarray13_, mclVv(S, "S"))));
    /*
     * for i=1:max_pres
     */
    {
        int v_ = mclForIntStart(1);
        int e_ = mclForIntEnd(mclVv(max_pres, "max_pres"));
        if (v_ > e_) {
            mlfAssign(&i, _mxarray6_);
        } else {
            /*
             * 
             * % TRAINING PARAMETER UPDATE
             * nb = max(1,max_nb*nb_x^i);
             * lr = base_lr*(5/(4+i) + lr_x^i);
             * 
             * % PRESENTATION PHASE
             * j = fix(rand*Q) + 1;
             * P = p(:,j);
             * a = simusm(P,w,m,nb);
             * 
             * % LEARNING PHASE
             * dw = learnis(w,P,a,lr);
             * w = w + dw;
             * 
             * % PLOTTING 
             * if rem(i,df) == 0
             * fprintf(message,i,nb,lr)
             * if R > 1 & S < 100
             * plotsm(w,m);
             * end
             * end
             * end
             */
            for (; ; ) {
                mlfAssign(
                  &nb,
                  mlfMax(
                    NULL,
                    _mxarray13_,
                    mclMtimes(
                      mclVv(max_nb, "max_nb"),
                      mclMpower(mclVv(nb_x, "nb_x"), mlfScalar(v_))),
                    NULL));
                mlfAssign(
                  &lr,
                  mclMtimes(
                    mclVv(base_lr, "base_lr"),
                    mclPlus(
                      mlfScalar(svDoubleScalarRdivide(5.0, (double) (4 + v_))),
                      mclMpower(mclVv(lr_x, "lr_x"), mlfScalar(v_)))));
                mlfAssign(
                  &j,
                  mclPlus(
                    mlfFix(mclMtimes(mlfNRand(1, NULL), mclVv(Q, "Q"))),
                    _mxarray13_));
                mlfAssign(
                  &P,
                  mclArrayRef2(
                    mclVa(p, "p"), mlfCreateColonIndex(), mclVv(j, "j")));
                mlfAssign(
                  &a,
                  mlfSimusm(
                    mclVv(P, "P"),
                    mclVa(w, "w"),
                    mclVa(m, "m"),
                    mclVv(nb, "nb")));
                mlfAssign(
                  &dw,
                  mlfLearnis(
                    NULL,
                    mclVa(w, "w"),
                    mclVv(P, "P"),
                    mclVv(a, "a"),
                    mclVv(lr, "lr"),
                    NULL,
                    NULL,
                    NULL,
                    NULL,
                    NULL,
                    NULL,
                    NULL,
                    NULL));
                mlfAssign(&w, mclPlus(mclVa(w, "w"), mclVv(dw, "dw")));
                if (mclEqBool(
                      mlfRem(mlfScalar(v_), mclVv(df, "df")), _mxarray11_)) {
                    mclPrintAns(
                      &ans,
                      mlfNFprintf(
                        0,
                        mclVv(message, "message"),
                        mlfScalar(v_),
                        mclVv(nb, "nb"),
                        mclVv(lr, "lr"),
                        NULL));
                    {
                        mxArray * a_
                          = mclInitialize(mclGt(mclVv(R, "R"), _mxarray13_));
                        if (mlfTobool(a_)
                            && mlfTobool(
                                 mclAnd(
                                   a_, mclLt(mclVv(S, "S"), _mxarray12_)))) {
                            mxDestroyArray(a_);
                            mlfPlotsm(mclVa(w, "w"), mclVa(m, "m"));
                        } else {
                            mxDestroyArray(a_);
                        }
                    }
                }
                if (v_ == e_) {
                    break;
                }
                ++v_;
            }
            mlfAssign(&i, mlfScalar(v_));
        }
    }
    /*
     * 
     * if rem(i,df) ~= 0
     */
    if (mclNeBool(mlfRem(mclVv(i, "i"), mclVv(df, "df")), _mxarray11_)) {
        /*
         * fprintf(message,i,nb,lr)
         */
        mclPrintAns(
          &ans,
          mlfNFprintf(
            0,
            mclVv(message, "message"),
            mclVv(i, "i"),
            mclVv(nb, "nb"),
            mclVv(lr, "lr"),
            NULL));
    /*
     * end
     */
    }
    mclValidateOutput(w, 1, nargout_, "w", "trainsm");
    mxDestroyArray(ans);
    mxDestroyArray(df);
    mxDestroyArray(max_pres);
    mxDestroyArray(init_lr);
    mxDestroyArray(max_nb);
    mxDestroyArray(R);
    mxDestroyArray(Q);
    mxDestroyArray(S);
    mxDestroyArray(message);
    mxDestroyArray(lr_x);
    mxDestroyArray(base_lr);
    mxDestroyArray(nb_x);
    mxDestroyArray(z);
    mxDestroyArray(i);
    mxDestroyArray(nb);
    mxDestroyArray(lr);
    mxDestroyArray(j);
    mxDestroyArray(P);
    mxDestroyArray(a);
    mxDestroyArray(dw);
    mxDestroyArray(tp);
    mxDestroyArray(p);
    mxDestroyArray(m);
    mclSetCurrentLocalFunctionTable(save_local_function_table_);
    return w;
}