scm2cpp-lasso 0.13.2__py3-none-win_amd64.whl

scm2cpp_lasso/_generated/gramops_blas/gramops.hpp

@@ -0,0 +1,3033 @@
+
+#ifndef GRAMOPS_HPP
+#define GRAMOPS_HPP
+#define SCM2CPP_MINIMAL
+#include	<cmath>
+#include	"scm2cpp-blas.hpp"
+#include	<array>
+#include	<math.h>
+#include	<vector>
+#include	"scm2cpp.hpp"
+
+SCM2CPP_FN int
+spd_inverse( scm2cpp::cspan<double> a,  scm2cpp::span<double> inv,  scm2cpp::span<double> l,  scm2cpp::span<double> z,  int  p );
+SCM2CPP_FN int
+enet_gap( scm2cpp::cspan<double> w,  scm2cpp::cspan<double> q,  scm2cpp::cspan<double> qw,  scm2cpp::span<double> xta,  std::array<double,2> & out,  double  alpha,  double  ynorm2,  int  m );
+SCM2CPP_FN int
+enet_safe_rule( scm2cpp::span<double> w,  scm2cpp::cspan<double> qm,  scm2cpp::span<double> qw,  scm2cpp::cspan<double> xta,  scm2cpp::span<int> active,  scm2cpp::span<int> excl,  const std::array<double,2> & out,  double  alpha,  int  m );
+SCM2CPP_FN int
+enet_gram( scm2cpp::span<double> w,  scm2cpp::cspan<double> qm,  scm2cpp::cspan<double> q,  scm2cpp::span<double> qw,  scm2cpp::span<double> xta,  scm2cpp::span<int> active,  scm2cpp::span<int> excl,  std::array<double,2> & out,  double  alpha,  int  maxiter,  double  tol,  int  m );
+int
+graphical_lasso( scm2cpp::cspan<double> emp,  scm2cpp::span<double> cov,  scm2cpp::span<double> prec,  double  alpha,  int  maxiter,  double  tol,  double  enettol,  int  p );
+int
+omp( scm2cpp::cspan<double> x,  scm2cpp::cspan<double> y,  scm2cpp::span<double> coef,  scm2cpp::span<double> path,  int  rec,  int  kmax,  int  n,  int  p );
+SCM2CPP_FN double
+logistic_fg( scm2cpp::cspan<double> coef,  scm2cpp::span<double> g,  scm2cpp::cspan<double> x,  scm2cpp::cspan<double> y,  scm2cpp::span<double> z,  double  l2,  int  multi,  int  fitint,  int  n,  int  p,  int  k );
+int
+logistic_lbfgs( scm2cpp::cspan<double> x,  scm2cpp::cspan<double> y,  scm2cpp::span<double> coef,  scm2cpp::span<double> out,  double  c,  int  maxiter,  double  tol,  int  multi,  int  fitint,  int  n,  int  p,  int  k );
+SCM2CPP_FN double
+huber_fg( scm2cpp::cspan<double> coef,  scm2cpp::span<double> g,  scm2cpp::cspan<double> x,  scm2cpp::cspan<double> y,  scm2cpp::span<double> z,  double  epsilon,  double  alpha,  int  fitint,  int  n,  int  p );
+int
+huber( scm2cpp::cspan<double> x,  scm2cpp::cspan<double> y,  scm2cpp::span<double> coef,  scm2cpp::span<double> out,  double  epsilon,  double  alpha,  int  maxiter,  double  tol,  int  fitint,  int  n,  int  p );
+SCM2CPP_FN double
+glm_fg( scm2cpp::cspan<double> coef,  scm2cpp::span<double> g,  scm2cpp::cspan<double> x,  scm2cpp::cspan<double> y,  scm2cpp::span<double> z,  int  family,  double  power,  double  l2,  int  fitint,  int  n,  int  p );
+int
+glm( scm2cpp::cspan<double> x,  scm2cpp::cspan<double> y,  scm2cpp::span<double> coef,  scm2cpp::span<double> out,  int  family,  double  power,  double  alpha,  int  maxiter,  double  tol,  int  fitint,  int  n,  int  p );
+SCM2CPP_FN double
+lb_abs( double  a );
+SCM2CPP_FN double
+lb_max( double  a,  double  b );
+SCM2CPP_FN double
+lb_min( double  a,  double  b );
+SCM2CPP_FN int
+lb_dcstep( std::array<double,8> & cs,  double  fp,  double  dp,  double  stpmin,  double  stpmax );
+SCM2CPP_FN int
+lb_search_start( std::array<double,16> & ls,  double  f,  double  g,  double  stp,  double  stpmin,  double  stpmax,  double  ftol );
+SCM2CPP_FN int
+lb_search_step( std::array<double,16> & ls,  std::array<double,8> & cs,  double  f,  double  g,  double  ftol,  double  gtol,  double  xtol,  double  stpmin,  double  stpmax );
+SCM2CPP_FN double
+lb_dot( scm2cpp::cspan<double> a,  scm2cpp::cspan<double> b,  int  nv );
+SCM2CPP_FN double
+lb_dot_row( scm2cpp::cspan<double> mem,  int  slot,  scm2cpp::cspan<double> v,  int  nv );
+SCM2CPP_FN double
+lb_dot_rows( scm2cpp::cspan<double> a,  int  ra,  scm2cpp::cspan<double> b,  int  rb,  int  nv );
+SCM2CPP_FN double
+lb_projgr( scm2cpp::cspan<double> g,  scm2cpp::cspan<double> w,  int  ib,  double  lbv,  int  nv );
+SCM2CPP_FN int
+lb_hv( scm2cpp::span<double> q,  scm2cpp::cspan<double> smem,  scm2cpp::cspan<double> ymem,  scm2cpp::cspan<double> rho,  scm2cpp::span<double> al,  double  theta,  int  col,  int  head,  int  m,  int  nv );
+SCM2CPP_FN double
+lb_vbv( scm2cpp::cspan<double> v,  scm2cpp::cspan<double> smem,  scm2cpp::cspan<double> ymem,  scm2cpp::cspan<double> rho,  scm2cpp::span<double> bm,  scm2cpp::span<double> sb,  double  theta,  int  col,  int  head,  int  m,  int  nv );
+SCM2CPP_FN int
+lb_direction( scm2cpp::cspan<double> w,  scm2cpp::cspan<double> g,  scm2cpp::span<double> d,  scm2cpp::span<double> q,  scm2cpp::span<double> e,  scm2cpp::cspan<double> smem,  scm2cpp::cspan<double> ymem,  scm2cpp::cspan<double> rho,  scm2cpp::span<double> al,  scm2cpp::span<double> bm,  scm2cpp::span<double> sb,  double  theta,  int  col,  int  head,  int  m,  int  ib,  double  lbv,  int  nv );
+int
+lbfgs( int  kind,  scm2cpp::span<double> w,  scm2cpp::cspan<double> x,  scm2cpp::cspan<double> y,  const std::array<double,4> & aux,  scm2cpp::span<double> work,  scm2cpp::span<double> out,  int  m,  int  maxiter,  double  pgtol,  double  ftol,  int  maxls,  int  ib,  double  lbv,  int  nv,  int  n,  int  p,  int  k );
+SCM2CPP_FN double
+lbfgs_fg( int  kind,  scm2cpp::cspan<double> w,  scm2cpp::span<double> g,  scm2cpp::cspan<double> x,  scm2cpp::cspan<double> y,  const std::array<double,4> & aux,  scm2cpp::span<double> work,  int  n,  int  p,  int  k );
+SCM2CPP_FN int
+gmm_mstep( scm2cpp::cspan<double> x,  scm2cpp::cspan<double> resp,  scm2cpp::span<double> weights,  scm2cpp::span<double> means,  scm2cpp::span<double> prec,  scm2cpp::span<double> dw,  scm2cpp::span<double> cov,  scm2cpp::span<double> l,  scm2cpp::span<double> u,  double  reg,  int  norm,  int  n,  int  p,  int  kc );
+SCM2CPP_FN double
+gmm_estep( scm2cpp::cspan<double> x,  scm2cpp::span<double> resp,  scm2cpp::cspan<double> weights,  scm2cpp::cspan<double> means,  scm2cpp::cspan<double> prec,  scm2cpp::span<double> wlp,  scm2cpp::span<double> y,  scm2cpp::span<double> u,  scm2cpp::span<double> mu,  int  n,  int  p,  int  kc );
+SCM2CPP_FN double
+gmm_abs( double  a );
+int
+gmm( scm2cpp::cspan<double> x,  scm2cpp::span<double> resp,  scm2cpp::span<double> weights,  scm2cpp::span<double> means,  scm2cpp::span<double> prec,  scm2cpp::span<double> labels,  scm2cpp::span<double> out,  int  maxiter,  double  tol,  double  reg,  int  n,  int  p,  int  kc );
+SCM2CPP_FN double
+eig_hypot( double  a,  double  b );
+SCM2CPP_FN int
+sym_eigen( scm2cpp::span<double> a,  scm2cpp::span<double> v,  scm2cpp::span<double> ev,  scm2cpp::span<double> e,  int  m );
+SCM2CPP_FN double
+eig_abs( double  a );
+SCM2CPP_FN int
+eig_order( scm2cpp::cspan<double> ev,  scm2cpp::span<int> ord,  int  m );
+SCM2CPP_FN double
+ica_tanh( double  x );
+SCM2CPP_FN int
+ica_sym( scm2cpp::span<double> w,  scm2cpp::span<double> a,  scm2cpp::span<double> v,  scm2cpp::span<double> ev,  scm2cpp::span<double> e,  scm2cpp::span<double> t,  int  m );
+int
+fastica( scm2cpp::cspan<double> x,  scm2cpp::cspan<double> winit,  scm2cpp::span<double> comps,  scm2cpp::span<double> white,  scm2cpp::span<double> unmix,  scm2cpp::span<double> mean,  int  maxiter,  double  tol,  int  n,  int  p,  int  nc );
+int
+pca( scm2cpp::cspan<double> x,  scm2cpp::span<double> comps,  scm2cpp::span<double> ev,  scm2cpp::span<double> sv,  scm2cpp::span<double> mean,  int  n,  int  p );
+SCM2CPP_FN int
+ard_pinvh( scm2cpp::span<double> a,  scm2cpp::span<double> s,  scm2cpp::span<double> v,  scm2cpp::span<double> ev,  scm2cpp::span<double> e,  int  k );
+SCM2CPP_FN int
+ard_sigma_coef( scm2cpp::cspan<double> xy,  scm2cpp::cspan<int> idx,  scm2cpp::cspan<double> lam,  scm2cpp::span<double> coef,  scm2cpp::span<double> sigma,  scm2cpp::span<double> a,  scm2cpp::span<double> v,  scm2cpp::span<double> ev,  scm2cpp::span<double> e,  scm2cpp::cspan<double> gk,  double  alpha,  int  k );
+int
+ard( scm2cpp::cspan<double> x,  scm2cpp::cspan<double> y,  scm2cpp::span<double> coef,  scm2cpp::span<double> sigma,  scm2cpp::span<double> lam,  scm2cpp::span<double> out,  int  maxiter,  double  tol,  double  alpha1,  double  alpha2,  double  lambda1,  double  lambda2,  double  thresh,  int  hoist,  int  n,  int  p );
+int
+f_regression( scm2cpp::cspan<double> x,  scm2cpp::cspan<double> y,  scm2cpp::span<double> r,  scm2cpp::span<double> f,  int  center,  int  n,  int  p,  int  q );
+int
+f_classif( scm2cpp::cspan<double> x,  scm2cpp::cspan<int> lab,  scm2cpp::span<double> f,  int  n,  int  p,  int  k );
+SCM2CPP_FN double
+min_pos( scm2cpp::cspan<double> v,  int  m );
+SCM2CPP_FN double
+round_even( double  y );
+SCM2CPP_FN int
+cholesky_delete( scm2cpp::span<double> l,  int  go,  int  n,  int  k );
+SCM2CPP_FN int
+swap_rc( scm2cpp::span<double> g,  int  a,  int  b,  int  p );
+int
+lars( scm2cpp::span<double> g,  scm2cpp::cspan<double> cov,  scm2cpp::span<double> coef,  scm2cpp::span<double> path,  scm2cpp::span<double> alphas,  int  rec,  double  nsamples,  double  alphamin,  int  method,  int  maxiter,  double  eps,  int  p );
+SCM2CPP_FN int
+rfe_chol( scm2cpp::cspan<double> g,  scm2cpp::cspan<int> idx,  scm2cpp::span<double> l,  double  alpha,  int  k,  int  p );
+SCM2CPP_FN int
+rfe_solve( scm2cpp::cspan<double> c,  scm2cpp::cspan<int> idx,  scm2cpp::span<double> w,  scm2cpp::cspan<double> l,  int  k,  int  p );
+int
+rfe( scm2cpp::cspan<double> x,  scm2cpp::cspan<double> y,  scm2cpp::span<double> coef,  scm2cpp::span<int> support,  scm2cpp::span<int> ranking,  double  alpha,  int  nsel,  int  step,  int  n,  int  p );
+SCM2CPP_FN int
+gram_xx( scm2cpp::cspan<double> x,  scm2cpp::span<double> g,  int  n,  int  p );
+SCM2CPP_FN int
+gram_xy( scm2cpp::cspan<double> x,  scm2cpp::cspan<double> y,  scm2cpp::span<double> c,  int  n,  int  p,  int  q );
+
+
+
+SCM2CPP_FN inline int
+spd_inverse( scm2cpp::cspan<double> a,  scm2cpp::span<double> inv,  scm2cpp::span<double> l,  scm2cpp::span<double> z,  int  p )
+{   for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+        double s = a[((j*p)+j)];
+        for( int k = 0 ; !(k == j) ; k = (k+1) ) {
+            s = (s-(l[((j*p)+k)]*l[((j*p)+k)]))  ;
+        }  ;
+        l[ ((j*p)+j) ] = sqrt(s)   ;
+        for( int i = (j+1) ; !(i >= p) ; i = (i+1) ) {
+            double t = a[((i*p)+j)];
+            for( int k = 0 ; !(k == j) ; k = (k+1) ) {
+                t = (t-(l[((i*p)+k)]*l[((j*p)+k)]))  ;
+            }  ;
+            l[ ((i*p)+j) ] = (t/l[((j*p)+j)])   ;
+            ;
+        }  ;
+        ;
+    }  ;
+    for( int c = 0 ; !(c == p) ; c = (c+1) ) {
+        for( int i = 0 ; !(i == p) ; i = (i+1) ) {
+            double t = ( ( i == c ) ? (1.0) : (0.0) );
+            for( int k = 0 ; !(k == i) ; k = (k+1) ) {
+                t = (t-(l[((i*p)+k)]*z[k]))  ;
+            }  ;
+            z[ i ] = (t/l[((i*p)+i)])   ;
+            ;
+        }  ;
+        for( int ii = 0 ; !(ii == p) ; ii = (ii+1) ) {
+            int i = ((p-1)-ii);
+            double t = z[i];
+            for( int k = (i+1) ; !(k >= p) ; k = (k+1) ) {
+                t = (t-(l[((k*p)+i)]*inv[((k*p)+c)]))  ;
+            }  ;
+            inv[ ((i*p)+c) ] = (t/l[((i*p)+i)])   ;
+            ;
+            ;
+        }  ;
+    }  ;
+    return 0 ;
+}
+
+
+SCM2CPP_FN inline int
+enet_gap( scm2cpp::cspan<double> w,  scm2cpp::cspan<double> q,  scm2cpp::cspan<double> qw,  scm2cpp::span<double> xta,  std::array<double,2> & out,  double  alpha,  double  ynorm2,  int  m )
+{   double qdotw = 0.0;
+    double wqw = 0.0;
+    double l1 = 0.0;
+    double dn = 0.0;
+    for( int j = 0 ; !(j == m) ; j = (j+1) ) {
+        double wj = w[j];
+        qdotw = (qdotw+(wj*q[j]))  ;
+        wqw = (wqw+(wj*qw[j]))  ;
+        l1 = (l1+( ( wj < 0.0 ) ? ((0.0-wj)) : (wj) ))  ;
+        double x = (q[j]-qw[j]);
+        xta[ j ] = x   ;
+        double ax = ( ( x < 0.0 ) ? ((0.0-x)) : (x) );
+        if( ax > dn ) {
+            dn = ax  ;
+        } else {
+            0  ;
+        }  ;
+        ;
+        ;
+        ;
+    }  ;
+    double rnorm2 = ((ynorm2+wqw)-(2.0*qdotw));
+    double ry = (ynorm2-qdotw);
+    double scale = ( ( dn > alpha ) ? ((alpha/dn)) : (1.0) );
+    out[ 0 ] = (((0.5*rnorm2)+(alpha*l1))-((-0.5*((scale*scale)*rnorm2))+(scale*ry)))   ;
+    out[ 1 ] = dn   ;
+    ;
+    ;
+    ;
+    return 0 ;
+}
+
+
+SCM2CPP_FN inline int
+enet_safe_rule( scm2cpp::span<double> w,  scm2cpp::cspan<double> qm,  scm2cpp::span<double> qw,  scm2cpp::cspan<double> xta,  scm2cpp::span<int> active,  scm2cpp::span<int> excl,  const std::array<double,2> & out,  double  alpha,  int  m )
+{
+    struct let17063 {
+        let17063() {}
+        double operator()( double  g  )
+        {
+            if( g < 0.0 ) {
+                return (0.0-g) ;
+            }
+            else {
+                return g ;
+            }
+        }
+    }  ;
+    double radius = (sqrt((2.0*let17063()(out[0])))/alpha);
+    double dn = out[1];
+    int nact = 0;
+    for( int j = 0 ; !(j == m) ; j = (j+1) ) {
+        if( excl[j] == 0 ) {
+            double xjt = (xta[j]/( ( alpha > dn ) ? (alpha) : (dn) ));
+            double dj = ((1.0-( ( xjt < 0.0 ) ? ((0.0-xjt)) : (xjt) ))/sqrt(qm[((j*m)+j)]));
+            if( dj <= radius ) {
+                active[ nact ] = j   ;
+                nact = (nact+1)  ;
+            } else {
+                if( !(w[j] == 0.0) ) {
+                    double wj = w[j];
+                    for( int r = 0 ; !(r == m) ; r = (r+1) ) {
+                        qw[ r ] = (qw[r]-(wj*qm[((j*m)+r)]))   ;
+                    }  ;
+                    w[ j ] = 0.0   ;
+                    ;
+                } else {
+                    0  ;
+                }  ;
+                excl[ j ] = 1   ;
+            }  ;
+            ;
+            ;
+        } else {
+            0  ;
+        }  ;
+    }  ;
+    return nact ;
+}
+
+
+SCM2CPP_FN inline int
+enet_gram( scm2cpp::span<double> w,  scm2cpp::cspan<double> qm,  scm2cpp::cspan<double> q,  scm2cpp::span<double> qw,  scm2cpp::span<double> xta,  scm2cpp::span<int> active,  scm2cpp::span<int> excl,  std::array<double,2> & out,  double  alpha,  int  maxiter,  double  tol,  int  m )
+{   double ynorm2 = 0.0;
+    double dwtol = tol;
+    int nact = 0;
+    int done = 0;
+    int iter = 0;
+    for( int j = 0 ; !(j == m) ; j = (j+1) ) {
+        ynorm2 = (ynorm2+(q[j]*q[j]))  ;
+        excl[ j ] = 0   ;
+    }  ;
+    scm2cpp::blas_gemv(qw, qm, w, 1.0, 0.0, 0, m, m)  ;
+    double gtol = (tol*ynorm2);
+    enet_gap(w,q,qw,xta,out,alpha,ynorm2,m)  ;
+    if( (out[0] >= 0.0 && out[0] <= gtol) ) {
+        done = 1  ;
+    } else {
+        for( int j = 0 ; !(j == m) ; j = (j+1) ) {
+            if( qm[((j*m)+j)] == 0.0 ) {
+                w[ j ] = 0.0   ;
+                excl[ j ] = 1   ;
+            } else {
+                0  ;
+            }  ;
+        }  ;
+        nact = enet_safe_rule(w,qm,qw,xta,active,excl,out,alpha,m)  ;
+    }  ;
+    for(  ; !((done == 1 || iter >= maxiter)) ;  ) {
+        double wmax = 0.0;
+        double dwmax = 0.0;
+        for( int f = 0 ; !(f == nact) ; f = (f+1) ) {
+            int j = active[f];
+            double qjj = qm[((j*m)+j)];
+            if( !(qjj == 0.0) ) {
+                double wj = w[j];
+                double tmp = ((q[j]-qw[j])+(wj*qjj));
+                double mag = (( ( tmp < 0.0 ) ? ((0.0-tmp)) : (tmp) )-alpha);
+                double wn = ( ( mag > 0.0 ) ? ((( ( tmp < 0.0 ) ? ((0.0-mag)) : (mag) )/qjj)) : (0.0) );
+                w[ j ] = wn   ;
+                if( !(wn == wj) ) {
+                    for( int r = 0 ; !(r == m) ; r = (r+1) ) {
+                        qw[ r ] = (qw[r]+((wn-wj)*qm[((j*m)+r)]))   ;
+                    }  ;
+                } else {
+                    0  ;
+                }  ;
+                double dw = ( ( wn < wj ) ? ((wj-wn)) : ((wn-wj)) );
+                double aw = ( ( wn < 0.0 ) ? ((0.0-wn)) : (wn) );
+                if( dw > dwmax ) {
+                    dwmax = dw  ;
+                } else {
+                    0  ;
+                }  ;
+                if( aw > wmax ) {
+                    wmax = aw  ;
+                } else {
+                    0  ;
+                }  ;
+                ;
+                ;
+                ;
+                ;
+                ;
+            } else {
+                0  ;
+            }  ;
+            ;
+            ;
+        }  ;
+        if( (wmax == 0.0 || (dwmax/wmax) <= dwtol || iter == (maxiter-1)) ) {
+            enet_gap(w,q,qw,xta,out,alpha,ynorm2,m)  ;
+            if( out[0] <= gtol ) {
+                done = 1  ;
+            } else {
+                nact = enet_safe_rule(w,qm,qw,xta,active,excl,out,alpha,m)  ;
+            }  ;
+        } else {
+            0  ;
+        }  ;
+        iter = (iter+1)  ;
+        ;
+    }  ;
+    ;
+    return iter ;
+}
+
+
+inline int
+graphical_lasso( scm2cpp::cspan<double> emp,  scm2cpp::span<double> cov,  scm2cpp::span<double> prec,  double  alpha,  int  maxiter,  double  tol,  double  enettol,  int  p )
+{   int m = (p-1);
+    std::vector<double> sub((m*m),0.0);
+    std::vector<double> row(m,0.0);
+    std::vector<double> b(m,0.0);
+    std::vector<double> bnew(m,0.0);
+    std::vector<double> qw(m,0.0);
+    std::vector<double> xta(m,0.0);
+    std::vector<int> active(m,0);
+    std::vector<int> excl(m,0);
+    std::array<double,2> out = scm2cpp::filled_array<double,2>(0.0);
+    std::vector<double> l((p*p),0.0);
+    std::vector<double> z(p,0.0);
+    int done = 0;
+    int it = 0;
+    int innermax = ( ( maxiter > 20 ) ? (maxiter) : (20) );
+    for( int i = 0 ; !(i == p) ; i = (i+1) ) {
+        for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+            cov[ ((i*p)+j) ] = ( ( i == j ) ? (emp[((i*p)+j)]) : ((0.95*emp[((i*p)+j)])) )   ;
+        }  ;
+    }  ;
+    spd_inverse(cov,prec,l,z,p)  ;
+    for(  ; !((done == 1 || it >= maxiter)) ;  ) {
+        for( int idx = 0 ; !(idx == p) ; idx = (idx+1) ) {
+            if( idx > 0 ) {
+                int di = (idx-1);
+                for( int s = 0 ; !(s == m) ; s = (s+1) ) {
+                    int o = ( ( s < idx ) ? (s) : ((s+1)) );
+                    sub[ ((di*m)+s) ] = cov[((di*p)+o)]   ;
+                    sub[ ((s*m)+di) ] = cov[((o*p)+di)]   ;
+                    ;
+                }  ;
+                ;
+            } else {
+                for( int s = 0 ; !(s == m) ; s = (s+1) ) {
+                    for( int t = 0 ; !(t == m) ; t = (t+1) ) {
+                        sub[ ((s*m)+t) ] = cov[(((s+1)*p)+(t+1))]   ;
+                    }  ;
+                }  ;
+            }  ;
+            double pii = prec[((idx*p)+idx)];
+            for( int s = 0 ; !(s == m) ; s = (s+1) ) {
+                int o = ( ( s < idx ) ? (s) : ((s+1)) );
+                row[ s ] = emp[((idx*p)+o)]   ;
+                b[ s ] = (0.0-(prec[((o*p)+idx)]/(pii+(1000.0*2.220446049250313e-16))))   ;
+                ;
+            }  ;
+            ;
+            enet_gram(b,sub,row,qw,xta,active,excl,out,alpha,innermax,enettol,m)  ;
+            double acc = 0.0;
+            for( int s = 0 ; !(s == m) ; s = (s+1) ) {
+                int o = ( ( s < idx ) ? (s) : ((s+1)) );
+                acc = (acc+(cov[((o*p)+idx)]*b[s]))  ;
+                ;
+            }  ;
+            double pnew = (1.0/(cov[((idx*p)+idx)]-acc));
+            prec[ ((idx*p)+idx) ] = pnew   ;
+            for( int s = 0 ; !(s == m) ; s = (s+1) ) {
+                int o = ( ( s < idx ) ? (s) : ((s+1)) );
+                prec[ ((o*p)+idx) ] = (0.0-(pnew*b[s]))   ;
+                prec[ ((idx*p)+o) ] = (0.0-(pnew*b[s]))   ;
+                ;
+            }  ;
+            ;
+            ;
+            scm2cpp::blas_gemv(bnew, sub, b, 1.0, 0.0, 0, m, m)  ;
+            for( int s = 0 ; !(s == m) ; s = (s+1) ) {
+                int o = ( ( s < idx ) ? (s) : ((s+1)) );
+                cov[ ((idx*p)+o) ] = bnew[s]   ;
+                cov[ ((o*p)+idx) ] = bnew[s]   ;
+                ;
+            }  ;
+        }  ;
+        double tr = 0.0;
+        double offl1 = 0.0;
+        for( int i = 0 ; !(i == p) ; i = (i+1) ) {
+            for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+                double t = prec[((i*p)+j)];
+                tr = (tr+(emp[((i*p)+j)]*t))  ;
+                if( !(i == j) ) {
+                    offl1 = (offl1+( ( t < 0.0 ) ? ((0.0-t)) : (t) ))  ;
+                } else {
+                    0  ;
+                }  ;
+                ;
+            }  ;
+        }  ;
+        double gap = ((tr-(1.0*p))+(alpha*offl1));
+        it = (it+1)  ;
+        if( ( ( gap < 0.0 ) ? ((0.0-gap)) : (gap) ) < tol ) {
+            done = 1  ;
+        } else {
+            0  ;
+        }  ;
+        ;
+        ;
+    }  ;
+    return it ;
+}
+
+
+inline int
+omp( scm2cpp::cspan<double> x,  scm2cpp::cspan<double> y,  scm2cpp::span<double> coef,  scm2cpp::span<double> path,  int  rec,  int  kmax,  int  n,  int  p )
+{   std::vector<double> g((p*p),0.0);
+    std::vector<double> xy(p,0.0);
+    std::vector<double> alpha(p,0.0);
+    std::vector<double> beta(p,0.0);
+    std::vector<int> index(p,0);
+    std::vector<double> l((kmax*kmax),0.0);
+    std::vector<double> gamma(kmax,0.0);
+    std::vector<double> z(kmax,0.0);
+    double eps = 2.220446049250313e-16;
+    int na = 0;
+    int done = 0;
+    scm2cpp::blas_syrk(g, x, 1.0, 0.0, 1, p, n)  ;
+    scm2cpp::blas_gemv(xy, x, y, 1.0, 0.0, 1, p, n)  ;
+    for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+        index[ j ] = j   ;
+        alpha[ j ] = xy[j]   ;
+    }  ;
+    for(  ; !(done == 1) ;  ) {
+        int lam = 0;
+        double best = -1.0;
+        for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+            double a = alpha[j];
+            double aa = ( ( a < 0.0 ) ? ((0.0-a)) : (a) );
+            if( aa > best ) {
+                best = aa  ;
+                lam = j  ;
+            } else {
+                0  ;
+            }  ;
+            ;
+            ;
+        }  ;
+        if( (lam < na || (alpha[lam]*alpha[lam]) < eps) ) {
+            done = 1  ;
+        } else {
+            if( na > 0 ) {
+                double v = 0.0;
+                for( int i = 0 ; !(i == na) ; i = (i+1) ) {
+                    double t = g[((lam*p)+i)];
+                    for( int s = 0 ; !(s == i) ; s = (s+1) ) {
+                        t = (t-(l[((i*kmax)+s)]*l[((na*kmax)+s)]))  ;
+                    }  ;
+                    double li = (t/l[((i*kmax)+i)]);
+                    l[ ((na*kmax)+i) ] = li   ;
+                    v = (v+(li*li))  ;
+                    ;
+                    ;
+                }  ;
+                double lkk = (g[((lam*p)+lam)]-v);
+                if( lkk <= eps ) {
+                    done = 1  ;
+                } else {
+                    l[ ((na*kmax)+na) ] = sqrt(lkk)   ;
+                }  ;
+                ;
+                ;
+            } else {
+                l[ 0 ] = sqrt(g[((lam*p)+lam)])   ;
+            }  ;
+            if( done == 0 ) {
+                for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+                    double grow = g[((na*p)+j)];
+                    g[ ((na*p)+j) ] = g[((lam*p)+j)]   ;
+                    g[ ((lam*p)+j) ] = grow   ;
+                    ;
+                }  ;
+                for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+                    double gcol = g[((j*p)+na)];
+                    g[ ((j*p)+na) ] = g[((j*p)+lam)]   ;
+                    g[ ((j*p)+lam) ] = gcol   ;
+                    ;
+                }  ;
+                int itmp = index[na];
+                index[ na ] = index[lam]   ;
+                index[ lam ] = itmp   ;
+                ;
+                double xtmp = xy[na];
+                xy[ na ] = xy[lam]   ;
+                xy[ lam ] = xtmp   ;
+                ;
+                na = (na+1)  ;
+                for( int i = 0 ; !(i == na) ; i = (i+1) ) {
+                    double t = xy[i];
+                    for( int s = 0 ; !(s == i) ; s = (s+1) ) {
+                        t = (t-(l[((i*kmax)+s)]*z[s]))  ;
+                    }  ;
+                    z[ i ] = (t/l[((i*kmax)+i)])   ;
+                    ;
+                }  ;
+                for( int ii = 0 ; !(ii == na) ; ii = (ii+1) ) {
+                    int i = ((na-1)-ii);
+                    double t = z[i];
+                    for( int s = (i+1) ; !(s >= na) ; s = (s+1) ) {
+                        t = (t-(l[((s*kmax)+i)]*gamma[s]))  ;
+                    }  ;
+                    gamma[ i ] = (t/l[((i*kmax)+i)])   ;
+                    ;
+                    ;
+                }  ;
+                if( rec == 1 ) {
+                    for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+                        path[ (((na-1)*p)+j) ] = 0.0   ;
+                    }  ;
+                    for( int s = 0 ; !(s == na) ; s = (s+1) ) {
+                        path[ (((na-1)*p)+index[s]) ] = gamma[s]   ;
+                    }  ;
+                } else {
+                    0  ;
+                }  ;
+                for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+                    double acc = 0.0;
+                    for( int s = 0 ; !(s == na) ; s = (s+1) ) {
+                        acc = (acc+(g[((j*p)+s)]*gamma[s]))  ;
+                    }  ;
+                    beta[ j ] = acc   ;
+                    alpha[ j ] = (xy[j]-acc)   ;
+                    ;
+                }  ;
+                if( na == kmax ) {
+                    done = 1  ;
+                } else {
+                    0  ;
+                }  ;
+            } else {
+                0  ;
+            }  ;
+        }  ;
+        ;
+    }  ;
+    for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+        coef[ j ] = 0.0   ;
+    }  ;
+    for( int s = 0 ; !(s == na) ; s = (s+1) ) {
+        coef[ index[s] ] = gamma[s]   ;
+    }  ;
+    return na ;
+}
+
+
+SCM2CPP_FN inline double
+logistic_fg( scm2cpp::cspan<double> coef,  scm2cpp::span<double> g,  scm2cpp::cspan<double> x,  scm2cpp::cspan<double> y,  scm2cpp::span<double> z,  double  l2,  int  multi,  int  fitint,  int  n,  int  p,  int  k )
+{   double loss = 0.0;
+    double pen = 0.0;
+    scm2cpp::blas_gemm(z, x, coef, 1.0, 0.0, 0, 1, n, k, p)  ;
+    for( int i = 0 ; !(i == n) ; i = (i+1) ) {
+        for( int c = 0 ; !(c == k) ; c = (c+1) ) {
+            z[ ((i*k)+c) ] = (z[((i*k)+c)]+coef[((k*p)+c)])   ;
+        }  ;
+    }  ;
+    if( multi == 0 ) {
+        for( int i = 0 ; !(i == n) ; i = (i+1) ) {
+            double zi = z[i];
+            double yi = y[i];
+            if( zi <= -2.0 ) {
+                double e1 = exp(zi);
+                if( zi <= -37.0 ) {
+                    loss = (loss+(e1-(yi*zi)))  ;
+                    z[ i ] = (e1-yi)   ;
+                } else {
+                    loss = (loss+(log((1.0+e1))-(yi*zi)))  ;
+                    z[ i ] = ((((1.0-yi)*e1)-yi)/(1.0+e1))   ;
+                }  ;
+                ;
+            } else {
+                double e2 = exp((0.0-zi));
+                loss = (loss+(( ( zi <= 18.0 ) ? (log((1.0+e2))) : (e2) )+((1.0-yi)*zi)))  ;
+                z[ i ] = (((1.0-yi)-(yi*e2))/(1.0+e2))   ;
+                ;
+            }  ;
+            ;
+        }  ;
+    } else {
+        for( int i = 0 ; !(i == n) ; i = (i+1) ) {
+            double zmax = z[(i*k)];
+            double se = 0.0;
+            double yc = y[i];
+            for( int c = 1 ; !(c == k) ; c = (c+1) ) {
+                if( z[((i*k)+c)] > zmax ) {
+                    zmax = z[((i*k)+c)]  ;
+                } else {
+                    0  ;
+                }  ;
+            }  ;
+            loss = (loss-z[((i*k)+int(yc))])  ;
+            for( int c = 0 ; !(c == k) ; c = (c+1) ) {
+                double ec = exp((z[((i*k)+c)]-zmax));
+                z[ ((i*k)+c) ] = ec   ;
+                se = (se+ec)  ;
+                ;
+            }  ;
+            loss = (loss+(zmax+log(se)))  ;
+            for( int c = 0 ; !(c == k) ; c = (c+1) ) {
+                z[ ((i*k)+c) ] = (z[((i*k)+c)]/se)   ;
+            }  ;
+            int iy = ((i*k)+int(yc));
+            z[ iy ] = (z[iy]-1.0)   ;
+            ;
+            ;
+        }  ;
+    }  ;
+    scm2cpp::blas_gemm(g, z, x, 1.0, 0.0, 1, 0, k, p, n)  ;
+    for( int t = 0 ; !(t == (k*p)) ; t = (t+1) ) {
+        double wt = coef[t];
+        pen = (pen+(wt*wt))  ;
+        g[ t ] = ((g[t]/double(n))+(l2*wt))   ;
+        ;
+    }  ;
+    for( int c = 0 ; !(c == k) ; c = (c+1) ) {
+        double acc = 0.0;
+        if( fitint == 1 ) {
+            for( int i = 0 ; !(i == n) ; i = (i+1) ) {
+                acc = (acc+z[((i*k)+c)])  ;
+            }  ;
+        } else {
+            0  ;
+        }  ;
+        g[ ((k*p)+c) ] = (acc/double(n))   ;
+        ;
+    }  ;
+    return ((loss/double(n))+(0.5*(l2*pen))) ;
+}
+
+
+inline int
+logistic_lbfgs( scm2cpp::cspan<double> x,  scm2cpp::cspan<double> y,  scm2cpp::span<double> coef,  scm2cpp::span<double> out,  double  c,  int  maxiter,  double  tol,  int  multi,  int  fitint,  int  n,  int  p,  int  k )
+{   std::array<double,4> aux = scm2cpp::filled_array<double,4>(0.0);
+    std::vector<double> work((n*k),0.0);
+    aux[ 0 ] = (1.0/(c*double(n)))   ;
+    aux[ 1 ] = ( ( multi == 1 ) ? (1.0) : (0.0) )   ;
+    aux[ 2 ] = ( ( fitint == 1 ) ? (1.0) : (0.0) )   ;
+    return lbfgs(0,coef,x,y,aux,work,out,10,maxiter,tol,(64.0*2.220446049250313e-16),50,-1,0.0,(k*(p+1)),n,p,k) ;
+}
+
+
+SCM2CPP_FN inline double
+huber_fg( scm2cpp::cspan<double> coef,  scm2cpp::span<double> g,  scm2cpp::cspan<double> x,  scm2cpp::cspan<double> y,  scm2cpp::span<double> z,  double  epsilon,  double  alpha,  int  fitint,  int  n,  int  p )
+{   double b = coef[p];
+    double sigma = coef[(p+1)];
+    int nout = 0;
+    double outsum = 0.0;
+    double sq = 0.0;
+    double csum = 0.0;
+    double ww = 0.0;
+    scm2cpp::blas_gemv(z, x, coef, 1.0, 0.0, 0, n, p)  ;
+    for( int i = 0 ; !(i == n) ; i = (i+1) ) {
+        double r = ((y[i]-z[i])-b);
+        double a = lb_abs(r);
+        if( a > (epsilon*sigma) ) {
+            nout = (nout+1)  ;
+            outsum = (outsum+a)  ;
+            z[ i ] = (-2.0*(epsilon*( ( r < 0.0 ) ? (-1.0) : (1.0) )))   ;
+        } else {
+            sq = (sq+(r*r))  ;
+            z[ i ] = ((-2.0*r)/sigma)   ;
+        }  ;
+        csum = (csum+z[i])  ;
+        ;
+        ;
+    }  ;
+    scm2cpp::blas_gemv(g, x, z, 1.0, 0.0, 1, p, n)  ;
+    for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+        double wj = coef[j];
+        ww = (ww+(wj*wj))  ;
+        g[ j ] = (g[j]+(2.0*(alpha*wj)))   ;
+        ;
+    }  ;
+    g[ p ] = ( ( fitint == 1 ) ? (csum) : (0.0) )   ;
+    double sqloss = (sq/sigma);
+    double eps2 = (epsilon*epsilon);
+    g[ (p+1) ] = ((double(n)-(double(nout)*eps2))-(sqloss/sigma))   ;
+    return ((((double(n)*sigma)+sqloss)+((2.0*(epsilon*outsum))-(sigma*(double(nout)*eps2))))+(alpha*ww)) ;
+}
+
+
+inline int
+huber( scm2cpp::cspan<double> x,  scm2cpp::cspan<double> y,  scm2cpp::span<double> coef,  scm2cpp::span<double> out,  double  epsilon,  double  alpha,  int  maxiter,  double  tol,  int  fitint,  int  n,  int  p )
+{   std::array<double,4> aux = scm2cpp::filled_array<double,4>(0.0);
+    std::vector<double> work(n,0.0);
+    aux[ 0 ] = epsilon   ;
+    aux[ 1 ] = alpha   ;
+    aux[ 2 ] = ( ( fitint == 1 ) ? (1.0) : (0.0) )   ;
+    return lbfgs(1,coef,x,y,aux,work,out,10,maxiter,tol,(10000000.0*2.220446049250313e-16),20,(p+1),(10.0*2.220446049250313e-16),(p+2),n,p,1) ;
+}
+
+
+SCM2CPP_FN inline double
+glm_fg( scm2cpp::cspan<double> coef,  scm2cpp::span<double> g,  scm2cpp::cspan<double> x,  scm2cpp::cspan<double> y,  scm2cpp::span<double> z,  int  family,  double  power,  double  l2,  int  fitint,  int  n,  int  p )
+{   double loss = 0.0;
+    double pen = 0.0;
+    double b = coef[p];
+    scm2cpp::blas_gemv(z, x, coef, 1.0, 0.0, 0, n, p)  ;
+    for( int i = 0 ; !(i == n) ; i = (i+1) ) {
+        double zi = (z[i]+b);
+        double yi = y[i];
+        if( (family == 0 || (family == 2 && power == 1.0)) ) {
+            double e = exp(zi);
+            loss = (loss+(e-(yi*zi)))  ;
+            z[ i ] = (e-yi)   ;
+            ;
+        } else {
+            if( (family == 1 || (family == 2 && power == 2.0)) ) {
+                double e2 = exp((0.0-zi));
+                loss = (loss+(zi+(yi*e2)))  ;
+                z[ i ] = (1.0-(yi*e2))   ;
+                ;
+            } else {
+                if( family == 3 ) {
+                    double r = (zi-yi);
+                    loss = (loss+(0.5*(r*r)))  ;
+                    z[ i ] = r   ;
+                    ;
+                } else {
+                    if( power == 0.0 ) {
+                        double e3 = exp(zi);
+                        loss = (loss+(0.5*((e3-yi)*(e3-yi))))  ;
+                        z[ i ] = (e3*(e3-yi))   ;
+                        ;
+                    } else {
+                        double e1 = exp(((1.0-power)*zi));
+                        double e4 = exp(((2.0-power)*zi));
+                        loss = (loss+((e4/(2.0-power))-((yi*e1)/(1.0-power))))  ;
+                        z[ i ] = (e4-(yi*e1))   ;
+                        ;
+                    }  ;
+                }  ;
+            }  ;
+        }  ;
+        ;
+    }  ;
+    scm2cpp::blas_gemv(g, x, z, 1.0, 0.0, 1, p, n)  ;
+    for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+        double wj = coef[j];
+        pen = (pen+(wj*wj))  ;
+        g[ j ] = ((g[j]/double(n))+(l2*wj))   ;
+        ;
+    }  ;
+    double acc = 0.0;
+    if( fitint == 1 ) {
+        for( int i = 0 ; !(i == n) ; i = (i+1) ) {
+            acc = (acc+z[i])  ;
+        }  ;
+    } else {
+        0  ;
+    }  ;
+    g[ p ] = (acc/double(n))   ;
+    ;
+    return ((loss/double(n))+(0.5*(l2*pen))) ;
+}
+
+
+inline int
+glm( scm2cpp::cspan<double> x,  scm2cpp::cspan<double> y,  scm2cpp::span<double> coef,  scm2cpp::span<double> out,  int  family,  double  power,  double  alpha,  int  maxiter,  double  tol,  int  fitint,  int  n,  int  p )
+{   std::array<double,4> aux = scm2cpp::filled_array<double,4>(0.0);
+    std::vector<double> work(n,0.0);
+    double ym = 0.0;
+    aux[ 0 ] = alpha   ;
+    aux[ 1 ] = double(family)   ;
+    aux[ 2 ] = ( ( fitint == 1 ) ? (1.0) : (0.0) )   ;
+    aux[ 3 ] = power   ;
+    for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+        coef[ j ] = 0.0   ;
+    }  ;
+    for( int i = 0 ; !(i == n) ; i = (i+1) ) {
+        ym = (ym+y[i])  ;
+    }  ;
+    ym = (ym/double(n))  ;
+    coef[ p ] = ( ( fitint == 1 ) ? (( ( family == 3 ) ? (ym) : (log(ym)) )) : (0.0) )   ;
+    return lbfgs(2,coef,x,y,aux,work,out,10,maxiter,tol,(64.0*2.220446049250313e-16),50,-1,0.0,(p+1),n,p,1) ;
+}
+
+
+SCM2CPP_FN inline double
+lb_abs( double  a )
+{
+    if( a < 0.0 ) {
+        return (0.0-a) ;
+    }
+    else {
+        return a ;
+    }
+}
+
+
+SCM2CPP_FN inline double
+lb_max( double  a,  double  b )
+{
+    if( a > b ) {
+        return a ;
+    }
+    else {
+        return b ;
+    }
+}
+
+
+SCM2CPP_FN inline double
+lb_min( double  a,  double  b )
+{
+    if( a < b ) {
+        return a ;
+    }
+    else {
+        return b ;
+    }
+}
+
+
+SCM2CPP_FN inline int
+lb_dcstep( std::array<double,8> & cs,  double  fp,  double  dp,  double  stpmin,  double  stpmax )
+{   double stx = cs[0];
+    double fx = cs[1];
+    double dx = cs[2];
+    double sty = cs[3];
+    double fy = cs[4];
+    double dy = cs[5];
+    double stp = cs[6];
+    double brackt = cs[7];
+    double stpf = 0.0;
+    double sgnd = (dp*(dx/lb_abs(dx)));
+    if( fp > fx ) {
+        double th1 = ((((3.0*(fx-fp))/(stp-stx))+dx)+dp);
+        double s1 = lb_max(lb_abs(th1),lb_max(lb_abs(dx),lb_abs(dp)));
+        double ga1 = (s1*sqrt((((th1/s1)*(th1/s1))-((dx/s1)*(dp/s1)))));
+        double gm1 = ( ( stp < stx ) ? ((0.0-ga1)) : (ga1) );
+        double r1 = (((gm1-dx)+th1)/(((gm1-dx)+gm1)+dp));
+        double stpc1 = (stx+(r1*(stp-stx)));
+        double stpq1 = (stx+(((dx/(((fx-fp)/(stp-stx))+dx))/2.0)*(stp-stx)));
+        if( lb_abs(double((stpc1-stx))) < lb_abs(double((stpq1-stx))) ) {
+            stpf = stpc1  ;
+        } else {
+            stpf = (stpc1+((stpq1-stpc1)/2.0))  ;
+        }  ;
+        brackt = 1.0  ;
+        ;
+        ;
+        ;
+        ;
+        ;
+        ;
+        0  ;
+    } else {
+        if( sgnd < 0.0 ) {
+            double th2 = ((((3.0*(fx-fp))/(stp-stx))+dx)+dp);
+            double s2 = lb_max(lb_abs(th2),lb_max(lb_abs(dx),lb_abs(dp)));
+            double ga2 = (s2*sqrt((((th2/s2)*(th2/s2))-((dx/s2)*(dp/s2)))));
+            double gm2 = ( ( stp > stx ) ? ((0.0-ga2)) : (ga2) );
+            double r2 = (((gm2-dp)+th2)/(((gm2-dp)+gm2)+dx));
+            double stpc2 = (stp+(r2*(stx-stp)));
+            double stpq2 = (stp+((dp/(dp-dx))*(stx-stp)));
+            if( lb_abs(double((stpc2-stp))) > lb_abs(double((stpq2-stp))) ) {
+                stpf = stpc2  ;
+            } else {
+                stpf = stpq2  ;
+            }  ;
+            brackt = 1.0  ;
+            ;
+            ;
+            ;
+            ;
+            ;
+            ;
+            0  ;
+        } else {
+            if( lb_abs(dp) < lb_abs(dx) ) {
+                double th3 = ((((3.0*(fx-fp))/(stp-stx))+dx)+dp);
+                double s3 = lb_max(lb_abs(th3),lb_max(lb_abs(dx),lb_abs(dp)));
+                double ga3 = (s3*sqrt(lb_max(0.0,double((((th3/s3)*(th3/s3))-((dx/s3)*(dp/s3)))))));
+                double gm3 = ( ( stp > stx ) ? ((0.0-ga3)) : (ga3) );
+                double r3 = (((gm3-dp)+th3)/((gm3+(dx-dp))+gm3));
+                double stpc3 = ( ( (r3 < 0.0 && !(gm3 == 0.0)) ) ? ((stp+(r3*(stx-stp)))) : (( ( stp > stx ) ? (stpmax) : (stpmin) )) );
+                double stpq3 = (stp+((dp/(dp-dx))*(stx-stp)));
+                if( brackt == 1.0 ) {
+                    if( lb_abs(double((stpc3-stp))) < lb_abs(double((stpq3-stp))) ) {
+                        stpf = stpc3  ;
+                    } else {
+                        stpf = stpq3  ;
+                    }  ;
+                    if( stp > stx ) {
+                        stpf = lb_min(double((stp+(0.66*(sty-stp)))),stpf)  ;
+                    } else {
+                        stpf = lb_max(double((stp+(0.66*(sty-stp)))),stpf)  ;
+                    }  ;
+                } else {
+                    if( lb_abs(double((stpc3-stp))) > lb_abs(double((stpq3-stp))) ) {
+                        stpf = stpc3  ;
+                    } else {
+                        stpf = stpq3  ;
+                    }  ;
+                    stpf = lb_min(stpmax,stpf)  ;
+                    stpf = lb_max(stpmin,stpf)  ;
+                }  ;
+                ;
+                ;
+                ;
+                ;
+                ;
+                ;
+                0  ;
+            } else {
+                if( brackt == 1.0 ) {
+                    double th4 = ((((3.0*(fp-fy))/(sty-stp))+dy)+dp);
+                    double s4 = lb_max(lb_abs(th4),lb_max(lb_abs(dy),lb_abs(dp)));
+                    double ga4 = (s4*sqrt((((th4/s4)*(th4/s4))-((dy/s4)*(dp/s4)))));
+                    double gm4 = ( ( stp > sty ) ? ((0.0-ga4)) : (ga4) );
+                    double r4 = (((gm4-dp)+th4)/(((gm4-dp)+gm4)+dy));
+                    stpf = (stp+(r4*(sty-stp)))  ;
+                    ;
+                    ;
+                    ;
+                    ;
+                    ;
+                    0  ;
+                } else {
+                    if( stp > stx ) {
+                        stpf = stpmax  ;
+                    } else {
+                        stpf = stpmin  ;
+                    }  ;
+                }  ;
+            }  ;
+        }  ;
+    }  ;
+    if( fp > fx ) {
+        cs[ 3 ] = stp   ;
+        cs[ 4 ] = fp   ;
+        cs[ 5 ] = dp   ;
+    } else {
+        if( sgnd < 0.0 ) {
+            cs[ 3 ] = stx   ;
+            cs[ 4 ] = fx   ;
+            cs[ 5 ] = dx   ;
+        } else {
+            0  ;
+        }  ;
+        cs[ 0 ] = stp   ;
+        cs[ 1 ] = fp   ;
+        cs[ 2 ] = dp   ;
+    }  ;
+    cs[ 6 ] = stpf   ;
+    cs[ 7 ] = brackt   ;
+    return 0 ;
+}
+
+
+SCM2CPP_FN inline int
+lb_search_start( std::array<double,16> & ls,  double  f,  double  g,  double  stp,  double  stpmin,  double  stpmax,  double  ftol )
+{   ls[ 0 ] = 0.0   ;
+    ls[ 1 ] = f   ;
+    ls[ 2 ] = g   ;
+    ls[ 3 ] = 0.0   ;
+    ls[ 4 ] = f   ;
+    ls[ 5 ] = g   ;
+    ls[ 6 ] = 0.0   ;
+    ls[ 7 ] = 1.0   ;
+    ls[ 8 ] = f   ;
+    ls[ 9 ] = g   ;
+    ls[ 10 ] = (ftol*g)   ;
+    ls[ 11 ] = (stpmax-stpmin)   ;
+    ls[ 12 ] = ((stpmax-stpmin)/0.5)   ;
+    ls[ 13 ] = 0.0   ;
+    ls[ 14 ] = (stp+(4.0*stp))   ;
+    ls[ 15 ] = stp   ;
+    return 0 ;
+}
+
+
+SCM2CPP_FN inline int
+lb_search_step( std::array<double,16> & ls,  std::array<double,8> & cs,  double  f,  double  g,  double  ftol,  double  gtol,  double  xtol,  double  stpmin,  double  stpmax )
+{   double stp = ls[15];
+    double finit = ls[8];
+    double ginit = ls[9];
+    double gtest = ls[10];
+    int task = 0;
+    double ftest = (finit+(stp*gtest));
+    if( (ls[7] == 1.0 && (f <= ftest && g >= 0.0)) ) {
+        ls[ 7 ] = 2.0   ;
+    } else {
+        0  ;
+    }  ;
+    if( (ls[6] == 1.0 && (stp <= ls[13] || stp >= ls[14])) ) {
+        task = 2  ;
+    } else {
+        0  ;
+    }  ;
+    if( (ls[6] == 1.0 && (ls[14]-ls[13]) <= (xtol*ls[14])) ) {
+        task = 2  ;
+    } else {
+        0  ;
+    }  ;
+    if( (stp == stpmax && (f <= ftest && g <= gtest)) ) {
+        task = 2  ;
+    } else {
+        0  ;
+    }  ;
+    if( (stp == stpmin && (f > ftest || g >= gtest)) ) {
+        task = 2  ;
+    } else {
+        0  ;
+    }  ;
+    if( (f <= ftest && lb_abs(g) <= (gtol*(0.0-ginit))) ) {
+        task = 1  ;
+    } else {
+        0  ;
+    }  ;
+    if( task == 0 ) {
+        cs[ 0 ] = ls[0]   ;
+        cs[ 3 ] = ls[3]   ;
+        cs[ 6 ] = stp   ;
+        cs[ 7 ] = ls[6]   ;
+        if( (ls[7] == 1.0 && (f <= ls[1] && f > ftest)) ) {
+            cs[ 1 ] = (ls[1]-(ls[0]*gtest))   ;
+            cs[ 2 ] = (ls[2]-gtest)   ;
+            cs[ 4 ] = (ls[4]-(ls[3]*gtest))   ;
+            cs[ 5 ] = (ls[5]-gtest)   ;
+            lb_dcstep(cs,double((f-(stp*gtest))),double((g-gtest)),ls[13],ls[14])  ;
+            ls[ 1 ] = (cs[1]+(cs[0]*gtest))   ;
+            ls[ 2 ] = (cs[2]+gtest)   ;
+            ls[ 4 ] = (cs[4]+(cs[3]*gtest))   ;
+            ls[ 5 ] = (cs[5]+gtest)   ;
+        } else {
+            cs[ 1 ] = ls[1]   ;
+            cs[ 2 ] = ls[2]   ;
+            cs[ 4 ] = ls[4]   ;
+            cs[ 5 ] = ls[5]   ;
+            lb_dcstep(cs,f,g,ls[13],ls[14])  ;
+            ls[ 1 ] = cs[1]   ;
+            ls[ 2 ] = cs[2]   ;
+            ls[ 4 ] = cs[4]   ;
+            ls[ 5 ] = cs[5]   ;
+        }  ;
+        ls[ 0 ] = cs[0]   ;
+        ls[ 3 ] = cs[3]   ;
+        ls[ 6 ] = cs[7]   ;
+        double nstp = cs[6];
+        double stx = cs[0];
+        double sty = cs[3];
+        if( ls[6] == 1.0 ) {
+            if( lb_abs(double((sty-stx))) >= (0.66*ls[12]) ) {
+                nstp = (stx+(0.5*(sty-stx)))  ;
+            } else {
+                0  ;
+            }  ;
+            ls[ 12 ] = ls[11]   ;
+            ls[ 11 ] = lb_abs(double((sty-stx)))   ;
+        } else {
+            0  ;
+        }  ;
+        if( ls[6] == 1.0 ) {
+            ls[ 13 ] = lb_min(stx,sty)   ;
+            ls[ 14 ] = lb_max(stx,sty)   ;
+        } else {
+            ls[ 13 ] = (nstp+(1.1*(nstp-stx)))   ;
+            ls[ 14 ] = (nstp+(4.0*(nstp-stx)))   ;
+        }  ;
+        nstp = lb_max(nstp,stpmin)  ;
+        nstp = lb_min(nstp,stpmax)  ;
+        if( (ls[6] == 1.0 && ((nstp <= ls[13] || nstp >= ls[14]) || (ls[14]-ls[13]) <= (xtol*ls[14]))) ) {
+            nstp = stx  ;
+        } else {
+            0  ;
+        }  ;
+        ls[ 15 ] = nstp   ;
+        ;
+    } else {
+        0  ;
+    }  ;
+    return task ;
+}
+
+
+SCM2CPP_FN inline double
+lb_dot( scm2cpp::cspan<double> a,  scm2cpp::cspan<double> b,  int  nv )
+{   double acc = 0.0;
+    for( int j = 0 ; !(j == nv) ; j = (j+1) ) {
+        acc = (acc+(a[j]*b[j]))  ;
+    }  ;
+    return acc ;
+}
+
+
+SCM2CPP_FN inline double
+lb_dot_row( scm2cpp::cspan<double> mem,  int  slot,  scm2cpp::cspan<double> v,  int  nv )
+{   double acc = 0.0;
+    for( int j = 0 ; !(j == nv) ; j = (j+1) ) {
+        acc = (acc+(mem[((slot*nv)+j)]*v[j]))  ;
+    }  ;
+    return acc ;
+}
+
+
+SCM2CPP_FN inline double
+lb_dot_rows( scm2cpp::cspan<double> a,  int  ra,  scm2cpp::cspan<double> b,  int  rb,  int  nv )
+{   double acc = 0.0;
+    for( int j = 0 ; !(j == nv) ; j = (j+1) ) {
+        acc = (acc+(a[((ra*nv)+j)]*b[((rb*nv)+j)]))  ;
+    }  ;
+    return acc ;
+}
+
+
+SCM2CPP_FN inline double
+lb_projgr( scm2cpp::cspan<double> g,  scm2cpp::cspan<double> w,  int  ib,  double  lbv,  int  nv )
+{   double acc = 0.0;
+    for( int j = 0 ; !(j == nv) ; j = (j+1) ) {
+        double gj = g[j];
+        double gp = ( ( (j == ib && gj >= 0.0) ) ? (lb_min(double((w[j]-lbv)),gj)) : (gj) );
+        double v = lb_abs(gp);
+        if( v > acc ) {
+            acc = v  ;
+        } else {
+            0  ;
+        }  ;
+        ;
+        ;
+        ;
+    }  ;
+    return acc ;
+}
+
+
+SCM2CPP_FN inline int
+lb_hv( scm2cpp::span<double> q,  scm2cpp::cspan<double> smem,  scm2cpp::cspan<double> ymem,  scm2cpp::cspan<double> rho,  scm2cpp::span<double> al,  double  theta,  int  col,  int  head,  int  m,  int  nv )
+{   for( int ii = 0 ; !(ii == col) ; ii = (ii+1) ) {
+        int slot = ((head+((col-1)-ii)) % m);
+        double a = (rho[slot]*lb_dot_row(smem,slot,q,nv));
+        al[ slot ] = a   ;
+        for( int j = 0 ; !(j == nv) ; j = (j+1) ) {
+            q[ j ] = (q[j]-(a*ymem[((slot*nv)+j)]))   ;
+        }  ;
+        ;
+        ;
+    }  ;
+    for( int j = 0 ; !(j == nv) ; j = (j+1) ) {
+        q[ j ] = (q[j]/theta)   ;
+    }  ;
+    for( int ii = 0 ; !(ii == col) ; ii = (ii+1) ) {
+        int slot2 = ((head+ii) % m);
+        double bb = (rho[slot2]*lb_dot_row(ymem,slot2,q,nv));
+        for( int j = 0 ; !(j == nv) ; j = (j+1) ) {
+            q[ j ] = (q[j]+((al[slot2]-bb)*smem[((slot2*nv)+j)]))   ;
+        }  ;
+        ;
+        ;
+    }  ;
+    return 0 ;
+}
+
+
+SCM2CPP_FN inline double
+lb_vbv( scm2cpp::cspan<double> v,  scm2cpp::cspan<double> smem,  scm2cpp::cspan<double> ymem,  scm2cpp::cspan<double> rho,  scm2cpp::span<double> bm,  scm2cpp::span<double> sb,  double  theta,  int  col,  int  head,  int  m,  int  nv )
+{   for( int ii = 0 ; !(ii == col) ; ii = (ii+1) ) {
+        int si = ((head+ii) % m);
+        for( int j = 0 ; !(j == nv) ; j = (j+1) ) {
+            bm[ ((ii*nv)+j) ] = (theta*smem[((si*nv)+j)])   ;
+        }  ;
+        for( int jj = 0 ; !(jj == ii) ; jj = (jj+1) ) {
+            int sj = ((head+jj) % m);
+            double cb = (lb_dot_rows(bm,jj,smem,si,nv)/sb[jj]);
+            double cy = (rho[sj]*lb_dot_rows(ymem,sj,smem,si,nv));
+            for( int j = 0 ; !(j == nv) ; j = (j+1) ) {
+                bm[ ((ii*nv)+j) ] = ((bm[((ii*nv)+j)]-(cb*bm[((jj*nv)+j)]))+(cy*ymem[((sj*nv)+j)]))   ;
+            }  ;
+            ;
+            ;
+        }  ;
+        sb[ ii ] = lb_dot_rows(bm,ii,smem,si,nv)   ;
+        ;
+    }  ;
+    double acc = (theta*lb_dot(v,v,nv));
+    for( int ii = 0 ; !(ii == col) ; ii = (ii+1) ) {
+        int si2 = ((head+ii) % m);
+        double bv = lb_dot_row(bm,ii,v,nv);
+        double yv = lb_dot_row(ymem,si2,v,nv);
+        acc = ((acc-((bv*bv)/sb[ii]))+(rho[si2]*(yv*yv)))  ;
+        ;
+        ;
+    }  ;
+    return acc ;
+}
+
+
+SCM2CPP_FN inline int
+lb_direction( scm2cpp::cspan<double> w,  scm2cpp::cspan<double> g,  scm2cpp::span<double> d,  scm2cpp::span<double> q,  scm2cpp::span<double> e,  scm2cpp::cspan<double> smem,  scm2cpp::cspan<double> ymem,  scm2cpp::cspan<double> rho,  scm2cpp::span<double> al,  scm2cpp::span<double> bm,  scm2cpp::span<double> sb,  double  theta,  int  col,  int  head,  int  m,  int  ib,  double  lbv,  int  nv )
+{   for( int j = 0 ; !(j == nv) ; j = (j+1) ) {
+        q[ j ] = g[j]   ;
+    }  ;
+    lb_hv(q,smem,ymem,rho,al,theta,col,head,m,nv)  ;
+    for( int j = 0 ; !(j == nv) ; j = (j+1) ) {
+        d[ j ] = (0.0-q[j])   ;
+    }  ;
+    if( ib >= 0 ) {
+        double gs = g[ib];
+        double tl = (w[ib]-lbv);
+        int fixed = 0;
+        double delta = 0.0;
+        double dtm = 0.0;
+        if( (tl <= 0.0 && gs >= 0.0) ) {
+            fixed = 1  ;
+        } else {
+            dtm = (lb_dot(g,g,nv)/lb_vbv(g,smem,ymem,rho,bm,sb,theta,col,head,m,nv))  ;
+            if( (gs > 0.0 && dtm >= (tl/gs)) ) {
+                fixed = 1  ;
+                delta = (0.0-tl)  ;
+            } else {
+                0  ;
+            }  ;
+        }  ;
+        if( fixed == 1 ) {
+            for( int j = 0 ; !(j == nv) ; j = (j+1) ) {
+                e[ j ] = 0.0   ;
+            }  ;
+            e[ ib ] = 1.0   ;
+            lb_hv(e,smem,ymem,rho,al,theta,col,head,m,nv)  ;
+            double lam = ((delta+q[ib])/e[ib]);
+            for( int j = 0 ; !(j == nv) ; j = (j+1) ) {
+                d[ j ] = (d[j]+(lam*e[j]))   ;
+            }  ;
+            ;
+            d[ ib ] = delta   ;
+        } else {
+            if( (w[ib]+d[ib]) < lbv ) {
+                double ddp = ((lb_dot(d,g,nv)-(d[ib]*gs))+((lbv-w[ib])*gs));
+                if( ddp > 0.0 ) {
+                    double dus = (d[ib]-((0.0-dtm)*gs));
+                    double xcps = (w[ib]-(dtm*gs));
+                    double alpha = 1.0;
+                    if( dus < 0.0 ) {
+                        double temp2 = (lbv-xcps);
+                        if( temp2 >= 0.0 ) {
+                            alpha = 0.0  ;
+                        } else {
+                            if( dus < temp2 ) {
+                                alpha = (temp2/dus)  ;
+                            } else {
+                                0  ;
+                            }  ;
+                        }  ;
+                        ;
+                    } else {
+                        0  ;
+                    }  ;
+                    for( int j = 0 ; !(j == nv) ; j = (j+1) ) {
+                        double xcpj = ((0.0-dtm)*g[j]);
+                        d[ j ] = (xcpj+(alpha*(d[j]-xcpj)))   ;
+                        ;
+                    }  ;
+                    if( alpha < 1.0 ) {
+                        d[ ib ] = (lbv-w[ib])   ;
+                    } else {
+                        0  ;
+                    }  ;
+                    ;
+                } else {
+                    d[ ib ] = (lbv-w[ib])   ;
+                }  ;
+                ;
+            } else {
+                0  ;
+            }  ;
+        }  ;
+        ;
+    } else {
+        0  ;
+    }  ;
+    return 0 ;
+}
+
+
+inline int
+lbfgs( int  kind,  scm2cpp::span<double> w,  scm2cpp::cspan<double> x,  scm2cpp::cspan<double> y,  const std::array<double,4> & aux,  scm2cpp::span<double> work,  scm2cpp::span<double> out,  int  m,  int  maxiter,  double  pgtol,  double  ftol,  int  maxls,  int  ib,  double  lbv,  int  nv,  int  n,  int  p,  int  k )
+{   std::vector<double> g(nv,0.0);
+    std::vector<double> gold(nv,0.0);
+    std::vector<double> d(nv,0.0);
+    std::vector<double> t(nv,0.0);
+    std::vector<double> q(nv,0.0);
+    std::vector<double> e(nv,0.0);
+    std::vector<double> smem((m*nv),0.0);
+    std::vector<double> ymem((m*nv),0.0);
+    std::vector<double> bm((m*nv),0.0);
+    std::vector<double> rho(m,0.0);
+    std::vector<double> sb(m,0.0);
+    std::vector<double> al(m,0.0);
+    std::array<double,16> ls = scm2cpp::filled_array<double,16>(0.0);
+    std::array<double,8> cs = scm2cpp::filled_array<double,8>(0.0);
+    double eps = 2.220446049250313e-16;
+    int iter = 0;
+    int col = 0;
+    int head = 0;
+    double theta = 1.0;
+    int nfg = 0;
+    int done = 0;
+    double f = lbfgs_fg(kind,w,g,x,y,aux,work,n,p,k);
+    nfg = 1  ;
+    if( lb_projgr(g,w,ib,lbv,nv) <= pgtol ) {
+        done = 1  ;
+    } else {
+        0  ;
+    }  ;
+    for(  ; !(done == 1) ;  ) {
+        lb_direction(w,g,d,q,e,smem,ymem,rho,al,bm,sb,theta,col,head,m,ib,lbv,nv)  ;
+        for( int j = 0 ; !(j == nv) ; j = (j+1) ) {
+            t[ j ] = w[j]   ;
+            gold[ j ] = g[j]   ;
+        }  ;
+        double fold = f;
+        double dnorm = sqrt(lb_dot(d,d,nv));
+        double gdold = lb_dot(g,d,nv);
+        double stpmx = 10000000000.0;
+        double gd = 0.0;
+        int task = 0;
+        int ifun = 0;
+        int fail = 0;
+        if( ib >= 0 ) {
+            if( iter == 0 ) {
+                stpmx = 1.0  ;
+            } else {
+                if( d[ib] < 0.0 ) {
+                    double a2 = (lbv-w[ib]);
+                    if( a2 >= 0.0 ) {
+                        stpmx = 0.0  ;
+                    } else {
+                        if( (d[ib]*stpmx) < a2 ) {
+                            stpmx = (a2/d[ib])  ;
+                        } else {
+                            0  ;
+                        }  ;
+                    }  ;
+                    ;
+                } else {
+                    0  ;
+                }  ;
+            }  ;
+        } else {
+            0  ;
+        }  ;
+        if( gdold >= 0.0 ) {
+            fail = 1  ;
+        } else {
+            lb_search_start(ls,f,gdold,( ( iter == 0 ) ? (lb_min(double((1.0/dnorm)),stpmx)) : (1.0) ),0.0,stpmx,0.001)  ;
+            for(  ; !((fail == 1 || !(task == 0))) ;  ) {
+                ifun = (ifun+1)  ;
+                if( ifun > maxls ) {
+                    fail = 1  ;
+                } else {
+                    double stp = ls[15];
+                    for( int j = 0 ; !(j == nv) ; j = (j+1) ) {
+                        w[ j ] = (t[j]+(stp*d[j]))   ;
+                    }  ;
+                    f = lbfgs_fg(kind,w,g,x,y,aux,work,n,p,k)  ;
+                    nfg = (nfg+1)  ;
+                    gd = lb_dot(g,d,nv)  ;
+                    task = lb_search_step(ls,cs,f,gd,0.001,0.9,0.1,0.0,stpmx)  ;
+                    ;
+                }  ;
+            }  ;
+        }  ;
+        if( fail == 1 ) {
+            for( int j = 0 ; !(j == nv) ; j = (j+1) ) {
+                w[ j ] = t[j]   ;
+                g[ j ] = gold[j]   ;
+            }  ;
+            f = fold  ;
+            if( col == 0 ) {
+                done = 1  ;
+            } else {
+                col = 0  ;
+                head = 0  ;
+                theta = 1.0  ;
+            }  ;
+        } else {
+            double stp2 = ls[15];
+            iter = (iter+1)  ;
+            if( iter >= maxiter ) {
+                done = 1  ;
+            } else {
+                0  ;
+            }  ;
+            if( lb_projgr(g,w,ib,lbv,nv) <= pgtol ) {
+                done = 1  ;
+            } else {
+                0  ;
+            }  ;
+            if( (fold-f) <= (ftol*lb_max(lb_max(lb_abs(fold),lb_abs(f)),1.0)) ) {
+                done = 1  ;
+            } else {
+                0  ;
+            }  ;
+            if( done == 0 ) {
+                double dr = ((gd-gdold)*stp2);
+                double ddum = ((0.0-gdold)*stp2);
+                if( dr > (eps*ddum) ) {
+                    int slot3 = ( ( col < m ) ? (((head+col) % m)) : (head) );
+                    double rr = 0.0;
+                    for( int j = 0 ; !(j == nv) ; j = (j+1) ) {
+                        double yj = (g[j]-gold[j]);
+                        smem[ ((slot3*nv)+j) ] = (stp2*d[j])   ;
+                        ymem[ ((slot3*nv)+j) ] = yj   ;
+                        rr = (rr+(yj*yj))  ;
+                        ;
+                    }  ;
+                    rho[ slot3 ] = (1.0/dr)   ;
+                    theta = (rr/dr)  ;
+                    if( col < m ) {
+                        col = (col+1)  ;
+                    } else {
+                        head = ((head+1) % m)  ;
+                    }  ;
+                    ;
+                    ;
+                    0  ;
+                } else {
+                    0  ;
+                }  ;
+                ;
+            } else {
+                0  ;
+            }  ;
+            ;
+        }  ;
+        ;
+    }  ;
+    out[ 0 ] = f   ;
+    out[ 1 ] = double(nfg)   ;
+    return iter ;
+}
+
+
+SCM2CPP_FN inline double
+lbfgs_fg( int  kind,  scm2cpp::cspan<double> w,  scm2cpp::span<double> g,  scm2cpp::cspan<double> x,  scm2cpp::cspan<double> y,  const std::array<double,4> & aux,  scm2cpp::span<double> work,  int  n,  int  p,  int  k )
+{
+    if( kind == 0 ) {
+        return logistic_fg(w,g,x,y,work,aux[0],( ( aux[1] == 1.0 ) ? (1) : (0) ),( ( aux[2] == 1.0 ) ? (1) : (0) ),n,p,k) ;
+    }
+    else {
+        if( kind == 1 ) {
+            return huber_fg(w,g,x,y,work,aux[0],aux[1],( ( aux[2] == 1.0 ) ? (1) : (0) ),n,p) ;
+        }
+        else {
+            return glm_fg(w,g,x,y,work,int(aux[1]),aux[3],aux[0],( ( aux[2] == 1.0 ) ? (1) : (0) ),n,p) ;
+        }
+    }
+}
+
+
+SCM2CPP_FN inline int
+gmm_mstep( scm2cpp::cspan<double> x,  scm2cpp::cspan<double> resp,  scm2cpp::span<double> weights,  scm2cpp::span<double> means,  scm2cpp::span<double> prec,  scm2cpp::span<double> dw,  scm2cpp::span<double> cov,  scm2cpp::span<double> l,  scm2cpp::span<double> u,  double  reg,  int  norm,  int  n,  int  p,  int  kc )
+{   double tot = 0.0;
+    for( int c = 0 ; !(c == kc) ; c = (c+1) ) {
+        weights[ c ] = 0.0   ;
+    }  ;
+    for( int i = 0 ; !(i == n) ; i = (i+1) ) {
+        for( int c = 0 ; !(c == kc) ; c = (c+1) ) {
+            weights[ c ] = (weights[c]+resp[((i*kc)+c)])   ;
+        }  ;
+    }  ;
+    for( int c = 0 ; !(c == kc) ; c = (c+1) ) {
+        weights[ c ] = (weights[c]+(10.0*2.220446049250313e-16))   ;
+        tot = (tot+weights[c])  ;
+    }  ;
+    scm2cpp::blas_gemm(means, resp, x, 1.0, 0.0, 1, 0, kc, p, n)  ;
+    for( int c = 0 ; !(c == kc) ; c = (c+1) ) {
+        double nk = weights[c];
+        for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+            means[ ((c*p)+j) ] = (means[((c*p)+j)]/nk)   ;
+        }  ;
+        for( int i = 0 ; !(i == n) ; i = (i+1) ) {
+            double r = resp[((i*kc)+c)];
+            for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+                double dv = (x[((i*p)+j)]-means[((c*p)+j)]);
+                u[ ((i*p)+j) ] = dv   ;
+                dw[ ((i*p)+j) ] = (r*dv)   ;
+                ;
+            }  ;
+            ;
+        }  ;
+        scm2cpp::blas_gemm(cov, dw, u, 1.0, 0.0, 1, 0, p, p, n)  ;
+        for( int t = 0 ; !(t == (p*p)) ; t = (t+1) ) {
+            cov[ t ] = (cov[t]/nk)   ;
+        }  ;
+        for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+            cov[ ((j*p)+j) ] = (cov[((j*p)+j)]+reg)   ;
+        }  ;
+        for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+            double s = cov[((j*p)+j)];
+            for( int m = 0 ; !(m == j) ; m = (m+1) ) {
+                s = (s-(l[((j*p)+m)]*l[((j*p)+m)]))  ;
+            }  ;
+            l[ ((j*p)+j) ] = sqrt(s)   ;
+            for( int i = (j+1) ; !(i >= p) ; i = (i+1) ) {
+                double t2 = cov[((i*p)+j)];
+                for( int m = 0 ; !(m == j) ; m = (m+1) ) {
+                    t2 = (t2-(l[((i*p)+m)]*l[((j*p)+m)]))  ;
+                }  ;
+                l[ ((i*p)+j) ] = (t2/l[((j*p)+j)])   ;
+                ;
+            }  ;
+            ;
+        }  ;
+        for( int t = 0 ; !(t == (p*p)) ; t = (t+1) ) {
+            prec[ ((c*(p*p))+t) ] = 0.0   ;
+        }  ;
+        for( int e = 0 ; !(e == p) ; e = (e+1) ) {
+            for( int i = e ; !(i == p) ; i = (i+1) ) {
+                double t3 = ( ( i == e ) ? (1.0) : (0.0) );
+                for( int m = e ; !(m == i) ; m = (m+1) ) {
+                    t3 = (t3-(l[((i*p)+m)]*prec[((c*(p*p))+((e*p)+m))]))  ;
+                }  ;
+                prec[ ((c*(p*p))+((e*p)+i)) ] = (t3/l[((i*p)+i)])   ;
+                ;
+            }  ;
+        }  ;
+        ;
+    }  ;
+    for( int c = 0 ; !(c == kc) ; c = (c+1) ) {
+        weights[ c ] = (weights[c]/( ( norm == 1 ) ? (double(n)) : (tot) ))   ;
+    }  ;
+    return 0 ;
+}
+
+
+SCM2CPP_FN inline double
+gmm_estep( scm2cpp::cspan<double> x,  scm2cpp::span<double> resp,  scm2cpp::cspan<double> weights,  scm2cpp::cspan<double> means,  scm2cpp::cspan<double> prec,  scm2cpp::span<double> wlp,  scm2cpp::span<double> y,  scm2cpp::span<double> u,  scm2cpp::span<double> mu,  int  n,  int  p,  int  kc )
+{   for( int c = 0 ; !(c == kc) ; c = (c+1) ) {
+        double logdet = 0.0;
+        double lw = log(weights[c]);
+        for( int t = 0 ; !(t == (p*p)) ; t = (t+1) ) {
+            u[ t ] = prec[((c*(p*p))+t)]   ;
+        }  ;
+        for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+            logdet = (logdet+log(u[((j*p)+j)]))  ;
+            double acc = 0.0;
+            for( int m = 0 ; !(m == p) ; m = (m+1) ) {
+                acc = (acc+(means[((c*p)+m)]*u[((m*p)+j)]))  ;
+            }  ;
+            mu[ j ] = acc   ;
+            ;
+        }  ;
+        scm2cpp::blas_gemm(y, x, u, 1.0, 0.0, 0, 0, n, p, p)  ;
+        for( int i = 0 ; !(i == n) ; i = (i+1) ) {
+            double sq = 0.0;
+            for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+                double v = (y[((i*p)+j)]-mu[j]);
+                sq = (sq+(v*v))  ;
+                ;
+            }  ;
+            wlp[ ((i*kc)+c) ] = (((-0.5*((double(p)*1.8378770664093453)+sq))+logdet)+lw)   ;
+            ;
+        }  ;
+        ;
+    }  ;
+    double lb = 0.0;
+    for( int i = 0 ; !(i == n) ; i = (i+1) ) {
+        double amax = wlp[(i*kc)];
+        double se = 0.0;
+        for( int c = 1 ; !(c == kc) ; c = (c+1) ) {
+            if( wlp[((i*kc)+c)] > amax ) {
+                amax = wlp[((i*kc)+c)]  ;
+            } else {
+                0  ;
+            }  ;
+        }  ;
+        for( int c = 0 ; !(c == kc) ; c = (c+1) ) {
+            se = (se+exp((wlp[((i*kc)+c)]-amax)))  ;
+        }  ;
+        double lse = (log(se)+amax);
+        lb = (lb+lse)  ;
+        for( int c = 0 ; !(c == kc) ; c = (c+1) ) {
+            resp[ ((i*kc)+c) ] = exp((wlp[((i*kc)+c)]-lse))   ;
+        }  ;
+        ;
+        ;
+    }  ;
+    return (lb/double(n)) ;
+}
+
+
+SCM2CPP_FN inline double
+gmm_abs( double  a )
+{
+    if( a < 0.0 ) {
+        return (0.0-a) ;
+    }
+    else {
+        return a ;
+    }
+}
+
+
+inline int
+gmm( scm2cpp::cspan<double> x,  scm2cpp::span<double> resp,  scm2cpp::span<double> weights,  scm2cpp::span<double> means,  scm2cpp::span<double> prec,  scm2cpp::span<double> labels,  scm2cpp::span<double> out,  int  maxiter,  double  tol,  double  reg,  int  n,  int  p,  int  kc )
+{   std::vector<double> dw((n*p),0.0);
+    std::vector<double> y((n*p),0.0);
+    std::vector<double> wlp((n*kc),0.0);
+    std::vector<double> cov((p*p),0.0);
+    std::vector<double> l((p*p),0.0);
+    std::vector<double> u((p*p),0.0);
+    std::vector<double> mu(p,0.0);
+    double lb = -1e+300;
+    int it = 0;
+    int conv = 0;
+    int done = 0;
+    gmm_mstep(x,resp,weights,means,prec,dw,cov,l,y,reg,1,n,p,kc)  ;
+    for(  ; !((done == 1 || it >= maxiter)) ;  ) {
+        double prev = lb;
+        lb = gmm_estep(x,resp,weights,means,prec,wlp,y,u,mu,n,p,kc)  ;
+        gmm_mstep(x,resp,weights,means,prec,dw,cov,l,y,reg,0,n,p,kc)  ;
+        it = (it+1)  ;
+        if( (it > 1 && gmm_abs(double((lb-prev))) < tol) ) {
+            conv = 1  ;
+            done = 1  ;
+        } else {
+            0  ;
+        }  ;
+        ;
+    }  ;
+    gmm_estep(x,resp,weights,means,prec,wlp,y,u,mu,n,p,kc)  ;
+    for( int i = 0 ; !(i == n) ; i = (i+1) ) {
+        int best = 0;
+        double bv = resp[(i*kc)];
+        for( int c = 1 ; !(c == kc) ; c = (c+1) ) {
+            if( resp[((i*kc)+c)] > bv ) {
+                bv = resp[((i*kc)+c)]  ;
+                best = c  ;
+            } else {
+                0  ;
+            }  ;
+        }  ;
+        labels[ i ] = double(best)   ;
+        ;
+    }  ;
+    out[ 0 ] = lb   ;
+    out[ 1 ] = double(conv)   ;
+    return it ;
+}
+
+
+SCM2CPP_FN inline double
+eig_hypot( double  a,  double  b )
+{
+    double aa = ( ( a < 0.0 ) ? ((0.0-a)) : (a) );
+    double bb = ( ( b < 0.0 ) ? ((0.0-b)) : (b) );
+    if( aa > bb ) {
+        return (aa*sqrt((1.0+((bb/aa)*(bb/aa))))) ;
+    }
+    else {
+        if( bb == 0.0 ) {
+            return 0.0 ;
+        }
+        else {
+            return (bb*sqrt((1.0+((aa/bb)*(aa/bb))))) ;
+        }
+    }
+}
+
+
+SCM2CPP_FN inline int
+sym_eigen( scm2cpp::span<double> a,  scm2cpp::span<double> v,  scm2cpp::span<double> ev,  scm2cpp::span<double> e,  int  m )
+{   for( int ii = 0 ; !(ii == (m-1)) ; ii = (ii+1) ) {
+        int i = ((m-1)-ii);
+        int l = (i-1);
+        double h = 0.0;
+        double scale = 0.0;
+        if( l > 0 ) {
+            for( int k = 0 ; !(k > l) ; k = (k+1) ) {
+                double ak = a[((i*m)+k)];
+                scale = (scale+( ( ak < 0.0 ) ? ((0.0-ak)) : (ak) ))  ;
+                ;
+            }  ;
+            if( scale == 0.0 ) {
+                e[ i ] = a[((i*m)+l)]   ;
+            } else {
+                for( int k = 0 ; !(k > l) ; k = (k+1) ) {
+                    a[ ((i*m)+k) ] = (a[((i*m)+k)]/scale)   ;
+                    h = (h+(a[((i*m)+k)]*a[((i*m)+k)]))  ;
+                }  ;
+                double f = a[((i*m)+l)];
+                double g = ( ( f >= 0.0 ) ? ((0.0-sqrt(h))) : (sqrt(h)) );
+                e[ i ] = (scale*g)   ;
+                h = (h-(f*g))  ;
+                a[ ((i*m)+l) ] = (f-g)   ;
+                double ff = 0.0;
+                for( int j = 0 ; !(j > l) ; j = (j+1) ) {
+                    a[ ((j*m)+i) ] = (a[((i*m)+j)]/h)   ;
+                    double gg = 0.0;
+                    for( int k = 0 ; !(k > j) ; k = (k+1) ) {
+                        gg = (gg+(a[((j*m)+k)]*a[((i*m)+k)]))  ;
+                    }  ;
+                    for( int k = (j+1) ; !(k > l) ; k = (k+1) ) {
+                        gg = (gg+(a[((k*m)+j)]*a[((i*m)+k)]))  ;
+                    }  ;
+                    e[ j ] = (gg/h)   ;
+                    ff = (ff+(e[j]*a[((i*m)+j)]))  ;
+                    ;
+                }  ;
+                double hh = (ff/(h+h));
+                for( int j = 0 ; !(j > l) ; j = (j+1) ) {
+                    double f2 = a[((i*m)+j)];
+                    double g2 = (e[j]-(hh*f2));
+                    e[ j ] = g2   ;
+                    for( int k = 0 ; !(k > j) ; k = (k+1) ) {
+                        a[ ((j*m)+k) ] = (a[((j*m)+k)]-((f2*e[k])+(g2*a[((i*m)+k)])))   ;
+                    }  ;
+                    ;
+                    ;
+                }  ;
+                ;
+                ;
+                ;
+                ;
+            }  ;
+        } else {
+            e[ i ] = a[((i*m)+l)]   ;
+        }  ;
+        ev[ i ] = h   ;
+        ;
+        ;
+    }  ;
+    ev[ 0 ] = 0.0   ;
+    e[ 0 ] = 0.0   ;
+    for( int i = 0 ; !(i == m) ; i = (i+1) ) {
+        if( !(ev[i] == 0.0) ) {
+            for( int j = 0 ; !(j == i) ; j = (j+1) ) {
+                double g3 = 0.0;
+                for( int k = 0 ; !(k == i) ; k = (k+1) ) {
+                    g3 = (g3+(a[((i*m)+k)]*a[((k*m)+j)]))  ;
+                }  ;
+                for( int k = 0 ; !(k == i) ; k = (k+1) ) {
+                    a[ ((k*m)+j) ] = (a[((k*m)+j)]-(g3*a[((k*m)+i)]))   ;
+                }  ;
+                ;
+            }  ;
+        } else {
+            0  ;
+        }  ;
+        ev[ i ] = a[((i*m)+i)]   ;
+        a[ ((i*m)+i) ] = 1.0   ;
+        for( int j = 0 ; !(j == i) ; j = (j+1) ) {
+            a[ ((j*m)+i) ] = 0.0   ;
+            a[ ((i*m)+j) ] = 0.0   ;
+        }  ;
+    }  ;
+    for( int i = 0 ; !(i == m) ; i = (i+1) ) {
+        for( int j = 0 ; !(j == m) ; j = (j+1) ) {
+            v[ ((j*m)+i) ] = a[((i*m)+j)]   ;
+        }  ;
+    }  ;
+    for( int i = 1 ; !(i == m) ; i = (i+1) ) {
+        e[ (i-1) ] = e[i]   ;
+    }  ;
+    e[ (m-1) ] = 0.0   ;
+    for( int l = 0 ; !(l == m) ; l = (l+1) ) {
+        int fin = 0;
+        int iter = 0;
+        for(  ; !((fin == 1 || iter > 60)) ;  ) {
+            int mm = l;
+            int found = 0;
+            for(  ; !((found == 1 || mm >= (m-1))) ;  ) {
+                double dd = (eig_abs(ev[mm])+eig_abs(ev[(mm+1)]));
+                if( (eig_abs(e[mm])+dd) == dd ) {
+                    found = 1  ;
+                } else {
+                    mm = (mm+1)  ;
+                }  ;
+                ;
+            }  ;
+            if( mm == l ) {
+                fin = 1  ;
+            } else {
+                iter = (iter+1)  ;
+                double g = ((ev[(l+1)]-ev[l])/(2.0*e[l]));
+                double r = eig_hypot(g,1.0);
+                double g4 = ((ev[mm]-ev[l])+(e[l]/(g+( ( g >= 0.0 ) ? (eig_abs(r)) : ((0.0-eig_abs(r))) ))));
+                double s = 1.0;
+                double c = 1.0;
+                double pp = 0.0;
+                double gq = g4;
+                int under = 0;
+                int i = (mm-1);
+                for(  ; !((under == 1 || i < l)) ;  ) {
+                    double f = (s*e[i]);
+                    double b = (c*e[i]);
+                    double r2 = eig_hypot(f,gq);
+                    e[ (i+1) ] = r2   ;
+                    if( r2 == 0.0 ) {
+                        ev[ (i+1) ] = (ev[(i+1)]-pp)   ;
+                        e[ mm ] = 0.0   ;
+                        under = 1  ;
+                    } else {
+                        s = (f/r2)  ;
+                        c = (gq/r2)  ;
+                        double g5 = (ev[(i+1)]-pp);
+                        double r3 = (((ev[i]-g5)*s)+(2.0*(c*b)));
+                        pp = (s*r3)  ;
+                        ev[ (i+1) ] = (g5+pp)   ;
+                        gq = ((c*r3)-b)  ;
+                        ;
+                        ;
+                        for( int k = 0 ; !(k == m) ; k = (k+1) ) {
+                            double zf = v[(((i+1)*m)+k)];
+                            double zi = v[((i*m)+k)];
+                            v[ (((i+1)*m)+k) ] = ((s*zi)+(c*zf))   ;
+                            v[ ((i*m)+k) ] = ((c*zi)-(s*zf))   ;
+                            ;
+                        }  ;
+                        i = (i-1)  ;
+                    }  ;
+                    ;
+                    ;
+                }  ;
+                if( under == 0 ) {
+                    ev[ l ] = (ev[l]-pp)   ;
+                    e[ l ] = gq   ;
+                    e[ mm ] = 0.0   ;
+                } else {
+                    0  ;
+                }  ;
+                ;
+                ;
+                ;
+                ;
+            }  ;
+            ;
+        }  ;
+        ;
+    }  ;
+    for( int i = 0 ; !(i == m) ; i = (i+1) ) {
+        for( int j = (i+1) ; !(j >= m) ; j = (j+1) ) {
+            double tmp = v[((i*m)+j)];
+            v[ ((i*m)+j) ] = v[((j*m)+i)]   ;
+            v[ ((j*m)+i) ] = tmp   ;
+            ;
+        }  ;
+    }  ;
+    return 0 ;
+}
+
+
+SCM2CPP_FN inline double
+eig_abs( double  a )
+{
+    if( a < 0.0 ) {
+        return (0.0-a) ;
+    }
+    else {
+        return a ;
+    }
+}
+
+
+SCM2CPP_FN inline int
+eig_order( scm2cpp::cspan<double> ev,  scm2cpp::span<int> ord,  int  m )
+{   for( int j = 0 ; !(j == m) ; j = (j+1) ) {
+        ord[ j ] = j   ;
+    }  ;
+    for( int j = 0 ; !(j == m) ; j = (j+1) ) {
+        int best = j;
+        for( int r = (j+1) ; !(r >= m) ; r = (r+1) ) {
+            if( ev[ord[r]] > ev[ord[best]] ) {
+                best = r  ;
+            } else {
+                0  ;
+            }  ;
+        }  ;
+        int tmp = ord[j];
+        ord[ j ] = ord[best]   ;
+        ord[ best ] = tmp   ;
+        ;
+        ;
+    }  ;
+    return 0 ;
+}
+
+
+SCM2CPP_FN inline double
+ica_tanh( double  x )
+{
+    return (1.0-(2.0/(exp((2.0*x))+1.0))) ;
+}
+
+
+SCM2CPP_FN inline int
+ica_sym( scm2cpp::span<double> w,  scm2cpp::span<double> a,  scm2cpp::span<double> v,  scm2cpp::span<double> ev,  scm2cpp::span<double> e,  scm2cpp::span<double> t,  int  m )
+{   for( int i = 0 ; !(i == m) ; i = (i+1) ) {
+        for( int j = 0 ; !(j == m) ; j = (j+1) ) {
+            double acc = 0.0;
+            for( int r = 0 ; !(r == m) ; r = (r+1) ) {
+                acc = (acc+(w[((i*m)+r)]*w[((j*m)+r)]))  ;
+            }  ;
+            a[ ((i*m)+j) ] = acc   ;
+            ;
+        }  ;
+    }  ;
+    sym_eigen(a,v,ev,e,m)  ;
+    for( int i = 0 ; !(i == m) ; i = (i+1) ) {
+        for( int j = 0 ; !(j == m) ; j = (j+1) ) {
+            double acc = 0.0;
+            for( int r = 0 ; !(r == m) ; r = (r+1) ) {
+                double sr = ( ( ev[r] < 2.2250738585072014e-308 ) ? (2.2250738585072014e-308) : (ev[r]) );
+                acc = (acc+((v[((i*m)+r)]/sqrt(sr))*v[((j*m)+r)]))  ;
+                ;
+            }  ;
+            t[ ((i*m)+j) ] = acc   ;
+            ;
+        }  ;
+    }  ;
+    for( int i = 0 ; !(i == m) ; i = (i+1) ) {
+        for( int j = 0 ; !(j == m) ; j = (j+1) ) {
+            double acc = 0.0;
+            for( int r = 0 ; !(r == m) ; r = (r+1) ) {
+                acc = (acc+(t[((i*m)+r)]*w[((r*m)+j)]))  ;
+            }  ;
+            a[ ((i*m)+j) ] = acc   ;
+            ;
+        }  ;
+    }  ;
+    for( int t2 = 0 ; !(t2 == (m*m)) ; t2 = (t2+1) ) {
+        w[ t2 ] = a[t2]   ;
+    }  ;
+    return 0 ;
+}
+
+
+inline int
+fastica( scm2cpp::cspan<double> x,  scm2cpp::cspan<double> winit,  scm2cpp::span<double> comps,  scm2cpp::span<double> white,  scm2cpp::span<double> unmix,  scm2cpp::span<double> mean,  int  maxiter,  double  tol,  int  n,  int  p,  int  nc )
+{   std::vector<double> xc((n*p),0.0);
+    std::vector<double> g((p*p),0.0);
+    std::vector<double> u((p*p),0.0);
+    std::vector<double> ev(p,0.0);
+    std::vector<int> ord(p,0);
+    std::vector<double> x1((nc*n),0.0);
+    std::vector<double> gw((nc*n),0.0);
+    std::vector<double> w1((nc*nc),0.0);
+    std::vector<double> gx(nc,0.0);
+    std::vector<double> sa((nc*nc),0.0);
+    std::vector<double> sv((nc*nc),0.0);
+    std::vector<double> se(nc,0.0);
+    std::vector<double> se2(nc,0.0);
+    std::vector<double> pe(p,0.0);
+    std::vector<double> st((nc*nc),0.0);
+    int it = 0;
+    int done = 0;
+    for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+        mean[ j ] = 0.0   ;
+    }  ;
+    for( int i = 0 ; !(i == n) ; i = (i+1) ) {
+        for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+            mean[ j ] = (mean[j]+x[((i*p)+j)])   ;
+        }  ;
+    }  ;
+    for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+        mean[ j ] = (mean[j]/double(n))   ;
+    }  ;
+    for( int i = 0 ; !(i == n) ; i = (i+1) ) {
+        for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+            xc[ ((i*p)+j) ] = (x[((i*p)+j)]-mean[j])   ;
+        }  ;
+    }  ;
+    scm2cpp::blas_syrk(g, xc, 1.0, 0.0, 1, p, n)  ;
+    sym_eigen(g,u,ev,pe,p)  ;
+    eig_order(ev,ord,p)  ;
+    for( int c = 0 ; !(c == nc) ; c = (c+1) ) {
+        int col = ord[c];
+        double sg = ( ( u[col] < 0.0 ) ? (-1.0) : (( ( u[col] > 0.0 ) ? (1.0) : (0.0) )) );
+        double d = sqrt(ev[col]);
+        for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+            white[ ((c*p)+j) ] = ((sg*u[((j*p)+col)])/d)   ;
+        }  ;
+        ;
+        ;
+    }  ;
+    scm2cpp::blas_gemm(x1, white, xc, 1.0, 0.0, 0, 1, nc, n, p)  ;
+    double sn = sqrt(double(n));
+    for( int t = 0 ; !(t == (nc*n)) ; t = (t+1) ) {
+        x1[ t ] = (x1[t]*sn)   ;
+    }  ;
+    ;
+    for( int t = 0 ; !(t == (nc*nc)) ; t = (t+1) ) {
+        unmix[ t ] = winit[t]   ;
+    }  ;
+    ica_sym(unmix,sa,sv,se,se2,st,nc)  ;
+    for(  ; !((done == 1 || it >= maxiter)) ;  ) {
+        scm2cpp::blas_gemm(gw, unmix, x1, 1.0, 0.0, 0, 0, nc, n, nc)  ;
+        for( int c = 0 ; !(c == nc) ; c = (c+1) ) {
+            double acc = 0.0;
+            for( int i = 0 ; !(i == n) ; i = (i+1) ) {
+                double th = ica_tanh(gw[((c*n)+i)]);
+                gw[ ((c*n)+i) ] = th   ;
+                acc = (acc+(1.0-(th*th)))  ;
+                ;
+            }  ;
+            gx[ c ] = (acc/double(n))   ;
+            ;
+        }  ;
+        scm2cpp::blas_gemm(w1, gw, x1, 1.0, 0.0, 0, 1, nc, nc, n)  ;
+        for( int c = 0 ; !(c == nc) ; c = (c+1) ) {
+            for( int j = 0 ; !(j == nc) ; j = (j+1) ) {
+                w1[ ((c*nc)+j) ] = ((w1[((c*nc)+j)]/double(n))-(gx[c]*unmix[((c*nc)+j)]))   ;
+            }  ;
+        }  ;
+        ica_sym(w1,sa,sv,se,se2,st,nc)  ;
+        double lim = 0.0;
+        for( int c = 0 ; !(c == nc) ; c = (c+1) ) {
+            double acc = 0.0;
+            for( int j = 0 ; !(j == nc) ; j = (j+1) ) {
+                acc = (acc+(w1[((c*nc)+j)]*unmix[((c*nc)+j)]))  ;
+            }  ;
+            double dv = (( ( acc < 0.0 ) ? ((0.0-acc)) : (acc) )-1.0);
+            double av = ( ( dv < 0.0 ) ? ((0.0-dv)) : (dv) );
+            if( av > lim ) {
+                lim = av  ;
+            } else {
+                0  ;
+            }  ;
+            ;
+            ;
+            ;
+        }  ;
+        for( int t = 0 ; !(t == (nc*nc)) ; t = (t+1) ) {
+            unmix[ t ] = w1[t]   ;
+        }  ;
+        it = (it+1)  ;
+        if( lim < tol ) {
+            done = 1  ;
+        } else {
+            0  ;
+        }  ;
+        ;
+    }  ;
+    for( int c = 0 ; !(c == nc) ; c = (c+1) ) {
+        for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+            double acc = 0.0;
+            for( int r = 0 ; !(r == nc) ; r = (r+1) ) {
+                acc = (acc+(unmix[((c*nc)+r)]*white[((r*p)+j)]))  ;
+            }  ;
+            comps[ ((c*p)+j) ] = acc   ;
+            ;
+        }  ;
+    }  ;
+    scm2cpp::blas_gemm(gw, comps, xc, 1.0, 0.0, 0, 1, nc, n, p)  ;
+    for( int c = 0 ; !(c == nc) ; c = (c+1) ) {
+        double m1 = 0.0;
+        double v1 = 0.0;
+        for( int i = 0 ; !(i == n) ; i = (i+1) ) {
+            m1 = (m1+gw[((c*n)+i)])  ;
+        }  ;
+        m1 = (m1/double(n))  ;
+        for( int i = 0 ; !(i == n) ; i = (i+1) ) {
+            double dd = (gw[((c*n)+i)]-m1);
+            v1 = (v1+(dd*dd))  ;
+            ;
+        }  ;
+        double sd = sqrt((v1/double(n)));
+        for( int j = 0 ; !(j == nc) ; j = (j+1) ) {
+            unmix[ ((c*nc)+j) ] = (unmix[((c*nc)+j)]/sd)   ;
+        }  ;
+        ;
+        ;
+    }  ;
+    for( int c = 0 ; !(c == nc) ; c = (c+1) ) {
+        for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+            double acc = 0.0;
+            for( int r = 0 ; !(r == nc) ; r = (r+1) ) {
+                acc = (acc+(unmix[((c*nc)+r)]*white[((r*p)+j)]))  ;
+            }  ;
+            comps[ ((c*p)+j) ] = acc   ;
+            ;
+        }  ;
+    }  ;
+    return it ;
+}
+
+
+inline int
+pca( scm2cpp::cspan<double> x,  scm2cpp::span<double> comps,  scm2cpp::span<double> ev,  scm2cpp::span<double> sv,  scm2cpp::span<double> mean,  int  n,  int  p )
+{   std::vector<double> c((p*p),0.0);
+    std::vector<double> v((p*p),0.0);
+    std::vector<double> e(p,0.0);
+    std::vector<double> sc(p,0.0);
+    std::vector<int> ord(p,0);
+    for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+        mean[ j ] = 0.0   ;
+    }  ;
+    for( int i = 0 ; !(i == n) ; i = (i+1) ) {
+        for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+            mean[ j ] = (mean[j]+x[((i*p)+j)])   ;
+        }  ;
+    }  ;
+    for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+        mean[ j ] = (mean[j]/double(n))   ;
+    }  ;
+    scm2cpp::blas_syrk(c, x, 1.0, 0.0, 1, p, n)  ;
+    for( int i = 0 ; !(i == p) ; i = (i+1) ) {
+        for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+            c[ ((i*p)+j) ] = ((c[((i*p)+j)]-(double(n)*(mean[i]*mean[j])))/double((n-1)))   ;
+        }  ;
+    }  ;
+    sym_eigen(c,v,e,sc,p)  ;
+    eig_order(e,ord,p)  ;
+    for( int r = 0 ; !(r == p) ; r = (r+1) ) {
+        int col = ord[r];
+        double lam = ( ( e[col] < 0.0 ) ? (0.0) : (e[col]) );
+        int big = 0;
+        double bv = -1.0;
+        ev[ r ] = lam   ;
+        sv[ r ] = sqrt((lam*double((n-1))))   ;
+        for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+            double a = v[((j*p)+col)];
+            double aa = ( ( a < 0.0 ) ? ((0.0-a)) : (a) );
+            if( aa > bv ) {
+                bv = aa  ;
+                big = j  ;
+            } else {
+                0  ;
+            }  ;
+            ;
+            ;
+        }  ;
+        double sg = ( ( v[((big*p)+col)] < 0.0 ) ? (-1.0) : (1.0) );
+        for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+            comps[ ((r*p)+j) ] = (sg*v[((j*p)+col)])   ;
+        }  ;
+        ;
+        ;
+        ;
+    }  ;
+    return 0 ;
+}
+
+
+SCM2CPP_FN inline int
+ard_pinvh( scm2cpp::span<double> a,  scm2cpp::span<double> s,  scm2cpp::span<double> v,  scm2cpp::span<double> ev,  scm2cpp::span<double> e,  int  k )
+{   sym_eigen(a,v,ev,e,k)  ;
+    double mx = 0.0;
+    for( int r = 0 ; !(r == k) ; r = (r+1) ) {
+        double av = ( ( ev[r] < 0.0 ) ? ((0.0-ev[r])) : (ev[r]) );
+        if( av > mx ) {
+            mx = av  ;
+        } else {
+            0  ;
+        }  ;
+        ;
+    }  ;
+    double cut = (mx*(double(k)*2.220446049250313e-16));
+    for( int i = 0 ; !(i == k) ; i = (i+1) ) {
+        for( int j = 0 ; !(j == k) ; j = (j+1) ) {
+            double acc = 0.0;
+            for( int r = 0 ; !(r == k) ; r = (r+1) ) {
+                double er = ev[r];
+                if( ( ( er < 0.0 ) ? ((0.0-er)) : (er) ) > cut ) {
+                    acc = (acc+((v[((i*k)+r)]*v[((j*k)+r)])/er))  ;
+                } else {
+                    0  ;
+                }  ;
+                ;
+            }  ;
+            s[ ((i*k)+j) ] = acc   ;
+            ;
+        }  ;
+    }  ;
+    ;
+    ;
+    return 0 ;
+}
+
+
+SCM2CPP_FN inline int
+ard_sigma_coef( scm2cpp::cspan<double> xy,  scm2cpp::cspan<int> idx,  scm2cpp::cspan<double> lam,  scm2cpp::span<double> coef,  scm2cpp::span<double> sigma,  scm2cpp::span<double> a,  scm2cpp::span<double> v,  scm2cpp::span<double> ev,  scm2cpp::span<double> e,  scm2cpp::cspan<double> gk,  double  alpha,  int  k )
+{   for( int r = 0 ; !(r == k) ; r = (r+1) ) {
+        for( int c = 0 ; !(c == k) ; c = (c+1) ) {
+            a[ ((r*k)+c) ] = (( ( r == c ) ? (lam[idx[r]]) : (0.0) )+(alpha*gk[((r*k)+c)]))   ;
+        }  ;
+    }  ;
+    ard_pinvh(a,sigma,v,ev,e,k)  ;
+    for( int r = 0 ; !(r == k) ; r = (r+1) ) {
+        double acc = 0.0;
+        for( int c = 0 ; !(c == k) ; c = (c+1) ) {
+            acc = (acc+(sigma[((r*k)+c)]*xy[idx[c]]))  ;
+        }  ;
+        coef[ idx[r] ] = (alpha*acc)   ;
+        ;
+    }  ;
+    return 0 ;
+}
+
+
+inline int
+ard( scm2cpp::cspan<double> x,  scm2cpp::cspan<double> y,  scm2cpp::span<double> coef,  scm2cpp::span<double> sigma,  scm2cpp::span<double> lam,  scm2cpp::span<double> out,  int  maxiter,  double  tol,  double  alpha1,  double  alpha2,  double  lambda1,  double  lambda2,  double  thresh,  int  hoist,  int  n,  int  p )
+{   std::vector<double> g(( ( hoist == 1 ) ? ((p*p)) : (1) ),0.0);
+    std::vector<double> xy(p,0.0);
+    std::vector<double> xk(( ( hoist == 1 ) ? (1) : ((n*p)) ),0.0);
+    std::vector<double> gk((p*p),0.0);
+    std::vector<double> a((p*p),0.0);
+    std::vector<double> v((p*p),0.0);
+    std::vector<double> ev(p,0.0);
+    std::vector<double> e(p,0.0);
+    std::vector<double> r(n,0.0);
+    std::vector<int> idx(p,0);
+    std::vector<int> keep(p,1);
+    std::vector<double> old(p,0.0);
+    double alpha = 0.0;
+    int k = p;
+    int it = 0;
+    int done = 0;
+    if( hoist == 1 ) {
+        scm2cpp::blas_syrk(g, x, 1.0, 0.0, 1, p, n)  ;
+    } else {
+        0  ;
+    }  ;
+    scm2cpp::blas_gemv(xy, x, y, 1.0, 0.0, 1, p, n)  ;
+    double ym = 0.0;
+    double vy = 0.0;
+    for( int i = 0 ; !(i == n) ; i = (i+1) ) {
+        ym = (ym+y[i])  ;
+    }  ;
+    ym = (ym/double(n))  ;
+    for( int i = 0 ; !(i == n) ; i = (i+1) ) {
+        vy = (vy+((y[i]-ym)*(y[i]-ym)))  ;
+    }  ;
+    alpha = (1.0/((vy/double(n))+2.220446049250313e-16))  ;
+    ;
+    for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+        lam[ j ] = 1.0   ;
+        coef[ j ] = 0.0   ;
+    }  ;
+    for(  ; !((done == 1 || it >= maxiter)) ;  ) {
+        k = 0  ;
+        for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+            if( keep[j] == 1 ) {
+                idx[ k ] = j   ;
+                k = (k+1)  ;
+            } else {
+                0  ;
+            }  ;
+        }  ;
+        if( hoist == 1 ) {
+            for( int r = 0 ; !(r == k) ; r = (r+1) ) {
+                for( int c = 0 ; !(c == k) ; c = (c+1) ) {
+                    gk[ ((r*k)+c) ] = g[((idx[r]*p)+idx[c])]   ;
+                }  ;
+            }  ;
+        } else {
+            for( int i = 0 ; !(i == n) ; i = (i+1) ) {
+                for( int c = 0 ; !(c == k) ; c = (c+1) ) {
+                    xk[ ((i*k)+c) ] = x[((i*p)+idx[c])]   ;
+                }  ;
+            }  ;
+            scm2cpp::blas_syrk(gk, xk, 1.0, 0.0, 1, k, n)  ;
+        }  ;
+        ard_sigma_coef(xy,idx,lam,coef,sigma,a,v,ev,e,gk,alpha,k)  ;
+        scm2cpp::blas_gemv(r, x, coef, 1.0, 0.0, 0, n, p)  ;
+        double sse = 0.0;
+        double gsum = 0.0;
+        for( int i = 0 ; !(i == n) ; i = (i+1) ) {
+            double d = (y[i]-r[i]);
+            sse = (sse+(d*d))  ;
+            ;
+        }  ;
+        for( int c = 0 ; !(c == k) ; c = (c+1) ) {
+            int j = idx[c];
+            double gm = (1.0-(lam[j]*sigma[((c*k)+c)]));
+            gsum = (gsum+gm)  ;
+            lam[ j ] = ((gm+(2.0*lambda1))/((coef[j]*coef[j])+(2.0*lambda2)))   ;
+            ;
+            ;
+        }  ;
+        alpha = (((double(n)-gsum)+(2.0*alpha1))/(sse+(2.0*alpha2)))  ;
+        ;
+        int nk = 0;
+        double moved = 0.0;
+        for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+            if( lam[j] < thresh ) {
+                keep[ j ] = 1   ;
+                nk = (nk+1)  ;
+            } else {
+                keep[ j ] = 0   ;
+                coef[ j ] = 0.0   ;
+            }  ;
+            double dv = (old[j]-coef[j]);
+            moved = (moved+( ( dv < 0.0 ) ? ((0.0-dv)) : (dv) ))  ;
+            ;
+        }  ;
+        if( (it > 0 && moved < tol) ) {
+            done = 1  ;
+        } else {
+            for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+                old[ j ] = coef[j]   ;
+            }  ;
+            if( nk == 0 ) {
+                done = 1  ;
+            } else {
+                0  ;
+            }  ;
+        }  ;
+        ;
+        it = (it+1)  ;
+    }  ;
+    k = 0  ;
+    for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+        if( keep[j] == 1 ) {
+            idx[ k ] = j   ;
+            k = (k+1)  ;
+        } else {
+            0  ;
+        }  ;
+    }  ;
+    if( k > 0 ) {
+        if( hoist == 1 ) {
+            for( int r = 0 ; !(r == k) ; r = (r+1) ) {
+                for( int c = 0 ; !(c == k) ; c = (c+1) ) {
+                    gk[ ((r*k)+c) ] = g[((idx[r]*p)+idx[c])]   ;
+                }  ;
+            }  ;
+        } else {
+            for( int i = 0 ; !(i == n) ; i = (i+1) ) {
+                for( int c = 0 ; !(c == k) ; c = (c+1) ) {
+                    xk[ ((i*k)+c) ] = x[((i*p)+idx[c])]   ;
+                }  ;
+            }  ;
+            scm2cpp::blas_syrk(gk, xk, 1.0, 0.0, 1, k, n)  ;
+        }  ;
+        ard_sigma_coef(xy,idx,lam,coef,sigma,a,v,ev,e,gk,alpha,k)  ;
+    } else {
+        0  ;
+    }  ;
+    out[ 0 ] = alpha   ;
+    out[ 1 ] = double(k)   ;
+    return it ;
+}
+
+
+inline int
+f_regression( scm2cpp::cspan<double> x,  scm2cpp::cspan<double> y,  scm2cpp::span<double> r,  scm2cpp::span<double> f,  int  center,  int  n,  int  p,  int  q )
+{   std::vector<double> mx(p,0.0);
+    std::vector<double> sx(p,0.0);
+    std::vector<double> yc((n*q),0.0);
+    std::vector<double> yn(q,0.0);
+    double dof = double((n-( ( center == 1 ) ? (2) : (1) )));
+    for( int i = 0 ; !(i == n) ; i = (i+1) ) {
+        for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+            double v = x[((i*p)+j)];
+            mx[ j ] = (mx[j]+v)   ;
+            sx[ j ] = (sx[j]+(v*v))   ;
+            ;
+        }  ;
+    }  ;
+    for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+        mx[ j ] = (mx[j]/double(n))   ;
+    }  ;
+    for( int t = 0 ; !(t == q) ; t = (t+1) ) {
+        double m = 0.0;
+        double s = 0.0;
+        if( center == 1 ) {
+            for( int i = 0 ; !(i == n) ; i = (i+1) ) {
+                m = (m+y[((i*q)+t)])  ;
+            }  ;
+            m = (m/double(n))  ;
+        } else {
+            0  ;
+        }  ;
+        for( int i = 0 ; !(i == n) ; i = (i+1) ) {
+            double v = (y[((i*q)+t)]-m);
+            yc[ ((i*q)+t) ] = v   ;
+            s = (s+(v*v))  ;
+            ;
+        }  ;
+        yn[ t ] = sqrt(s)   ;
+        ;
+    }  ;
+    scm2cpp::blas_gemm(r, x, yc, 1.0, 0.0, 1, 0, p, q, n)  ;
+    for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+        double xn = sqrt(( ( center == 1 ) ? ((sx[j]-(double(n)*(mx[j]*mx[j])))) : (sx[j]) ));
+        for( int t = 0 ; !(t == q) ; t = (t+1) ) {
+            double rr = ((r[((j*q)+t)]/xn)/yn[t]);
+            r[ ((j*q)+t) ] = rr   ;
+            f[ ((j*q)+t) ] = (((rr*rr)/(1.0-(rr*rr)))*dof)   ;
+            ;
+        }  ;
+        ;
+    }  ;
+    return 0 ;
+}
+
+
+inline int
+f_classif( scm2cpp::cspan<double> x,  scm2cpp::cspan<int> lab,  scm2cpp::span<double> f,  int  n,  int  p,  int  k )
+{   std::vector<double> oh((n*k),0.0);
+    std::vector<double> cs((k*p),0.0);
+    std::vector<double> cnt(k,0.0);
+    std::vector<double> ss(p,0.0);
+    std::vector<double> tot(p,0.0);
+    for( int i = 0 ; !(i == n) ; i = (i+1) ) {
+        int c = lab[i];
+        oh[ ((i*k)+c) ] = 1.0   ;
+        cnt[ c ] = (cnt[c]+1.0)   ;
+        ;
+        for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+            double v = x[((i*p)+j)];
+            ss[ j ] = (ss[j]+(v*v))   ;
+            ;
+        }  ;
+    }  ;
+    scm2cpp::blas_gemm(cs, oh, x, 1.0, 0.0, 1, 0, k, p, n)  ;
+    for( int c = 0 ; !(c == k) ; c = (c+1) ) {
+        for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+            tot[ j ] = (tot[j]+cs[((c*p)+j)])   ;
+        }  ;
+    }  ;
+    for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+        double sq = ((tot[j]*tot[j])/double(n));
+        double ssbn = 0.0;
+        for( int c = 0 ; !(c == k) ; c = (c+1) ) {
+            ssbn = (ssbn+((cs[((c*p)+j)]*cs[((c*p)+j)])/cnt[c]))  ;
+        }  ;
+        ssbn = (ssbn-sq)  ;
+        double sswn = ((ss[j]-sq)-ssbn);
+        f[ j ] = ((ssbn/double((k-1)))/(sswn/double((n-k))))   ;
+        ;
+        ;
+    }  ;
+    return 0 ;
+}
+
+
+SCM2CPP_FN inline double
+min_pos( scm2cpp::cspan<double> v,  int  m )
+{   double r = 1.7976931348623157e+308;
+    for( int i = 0 ; !(i == m) ; i = (i+1) ) {
+        double x = v[i];
+        if( (x > 0.0 && x < r) ) {
+            r = x  ;
+        } else {
+            0  ;
+        }  ;
+        ;
+    }  ;
+    return r ;
+}
+
+
+SCM2CPP_FN inline double
+round_even( double  y )
+{
+    double r = std::floor(y);
+    double d = (y-r);
+    if( d > 0.5 ) {
+        return (r+1.0) ;
+    }
+    else {
+        if( d < 0.5 ) {
+            return r ;
+        }
+        else {
+            if( (2.0*std::floor((0.5*r))) == r ) {
+                return r ;
+            }
+            else {
+                return (r+1.0) ;
+            }
+        }
+    }
+}
+
+
+SCM2CPP_FN inline int
+cholesky_delete( scm2cpp::span<double> l,  int  go,  int  n,  int  k )
+{   for( int i = go ; !(i >= (n-1)) ; i = (i+1) ) {
+        for( int j = 0 ; !(j == (i+2)) ; j = (j+1) ) {
+            l[ ((i*k)+j) ] = l[(((i+1)*k)+j)]   ;
+        }  ;
+    }  ;
+    for( int i = go ; !(i >= (n-1)) ; i = (i+1) ) {
+        double a = l[((i*k)+i)];
+        double b = l[((i*k)+(i+1))];
+        double aa = ( ( a < 0.0 ) ? ((0.0-a)) : (a) );
+        double ab = ( ( b < 0.0 ) ? ((0.0-b)) : (b) );
+        double roe = ( ( aa > ab ) ? (a) : (b) );
+        double scale = (aa+ab);
+        double r = ( ( scale == 0.0 ) ? (0.0) : ((( ( roe < 0.0 ) ? (-1.0) : (1.0) )*(scale*sqrt((((a/scale)*(a/scale))+((b/scale)*(b/scale))))))) );
+        double c = ( ( scale == 0.0 ) ? (1.0) : ((a/r)) );
+        double s = ( ( scale == 0.0 ) ? (0.0) : ((b/r)) );
+        double cc = ( ( r < 0.0 ) ? ((0.0-c)) : (c) );
+        double ss = ( ( r < 0.0 ) ? ((0.0-s)) : (s) );
+        l[ ((i*k)+i) ] = ( ( r < 0.0 ) ? ((0.0-r)) : (r) )   ;
+        l[ ((i*k)+(i+1)) ] = 0.0   ;
+        for( int t = (i+1) ; !(t >= n) ; t = (t+1) ) {
+            double x = l[((t*k)+i)];
+            double y = l[((t*k)+(i+1))];
+            l[ ((t*k)+i) ] = ((cc*x)+(ss*y))   ;
+            l[ ((t*k)+(i+1)) ] = ((cc*y)-(ss*x))   ;
+            ;
+        }  ;
+        ;
+        ;
+        ;
+        ;
+        ;
+        ;
+    }  ;
+    return 0 ;
+}
+
+
+SCM2CPP_FN inline int
+swap_rc( scm2cpp::span<double> g,  int  a,  int  b,  int  p )
+{   for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+        double rv = g[((a*p)+j)];
+        g[ ((a*p)+j) ] = g[((b*p)+j)]   ;
+        g[ ((b*p)+j) ] = rv   ;
+        ;
+    }  ;
+    for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+        double cv = g[((j*p)+a)];
+        g[ ((j*p)+a) ] = g[((j*p)+b)]   ;
+        g[ ((j*p)+b) ] = cv   ;
+        ;
+    }  ;
+    return 0 ;
+}
+
+
+inline int
+lars( scm2cpp::span<double> g,  scm2cpp::cspan<double> cov,  scm2cpp::span<double> coef,  scm2cpp::span<double> path,  scm2cpp::span<double> alphas,  int  rec,  double  nsamples,  double  alphamin,  int  method,  int  maxiter,  double  eps,  int  p )
+{   std::vector<double> g0((p*p),0.0);
+    std::vector<double> cov0(p,0.0);
+    std::vector<double> cbuf(p,0.0);
+    std::vector<int> index(p,0);
+    std::vector<int> active(p,0);
+    std::vector<double> sign(p,0.0);
+    int kmax = ( ( maxiter < p ) ? (maxiter) : (p) );
+    std::vector<double> prev(p,0.0);
+    std::vector<double> dir(p,0.0);
+    std::vector<double> ls(p,0.0);
+    std::vector<double> zz(p,0.0);
+    std::vector<double> ratio(p,0.0);
+    std::vector<int> idx(p,0);
+    std::vector<int> dropv(p,0);
+    int cs = 0;
+    int na = 0;
+    int niter = 0;
+    int drop = 0;
+    int done = 0;
+    double alpha = 0.0;
+    double prevalpha = 0.0;
+    double tiny32 = 1.1754943508222875e-38;
+    double eqtol = 1.1920928955078125e-7;
+    std::vector<double> l((kmax*kmax),0.0);
+    for( int t = 0 ; !(t == (p*p)) ; t = (t+1) ) {
+        g0[ t ] = g[t]   ;
+    }  ;
+    for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+        cov0[ j ] = cov[j]   ;
+        cbuf[ j ] = cov[j]   ;
+        index[ j ] = j   ;
+        coef[ j ] = 0.0   ;
+        if( rec == 1 ) {
+            path[ j ] = 0.0   ;
+        } else {
+            0  ;
+        }  ;
+    }  ;
+    for(  ; !(done == 1) ;  ) {
+        int clen = (p-cs);
+        int cidx = 0;
+        double cval = 0.0;
+        double cabs = 0.0;
+        for( int t = 0 ; !(t == clen) ; t = (t+1) ) {
+            double v = cbuf[(cs+t)];
+            double av = ( ( v < 0.0 ) ? ((0.0-v)) : (v) );
+            if( av > cabs ) {
+                cabs = av  ;
+                cidx = t  ;
+                cval = v  ;
+            } else {
+                0  ;
+            }  ;
+            ;
+            ;
+        }  ;
+        alpha = (cabs/(1.0*nsamples))  ;
+        if( rec == 1 ) {
+            alphas[ niter ] = alpha   ;
+        } else {
+            0  ;
+        }  ;
+        if( alpha <= (alphamin+eqtol) ) {
+            if( ( ( (alpha-alphamin) < 0.0 ) ? ((alphamin-alpha)) : ((alpha-alphamin)) ) > eqtol ) {
+                if( niter > 0 ) {
+                    double ss = ((prevalpha-alphamin)/(prevalpha-alpha));
+                    for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+                        coef[ j ] = (prev[j]+(ss*(coef[j]-prev[j])))   ;
+                    }  ;
+                    ;
+                } else {
+                    0  ;
+                }  ;
+                alpha = alphamin  ;
+                if( rec == 1 ) {
+                    alphas[ niter ] = alphamin   ;
+                    for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+                        path[ ((niter*p)+j) ] = coef[j]   ;
+                    }  ;
+                } else {
+                    0  ;
+                }  ;
+            } else {
+                0  ;
+            }  ;
+            done = 1  ;
+        } else {
+            if( (niter >= maxiter || na >= p) ) {
+                done = 1  ;
+            } else {
+                int degenerate = 0;
+                if( drop == 0 ) {
+                    int m = na;
+                    int nn = (cidx+na);
+                    sign[ na ] = ( ( cval > 0.0 ) ? (1.0) : (( ( cval < 0.0 ) ? (-1.0) : (0.0) )) )   ;
+                    double tc = cbuf[(cs+cidx)];
+                    cbuf[ (cs+cidx) ] = cbuf[cs]   ;
+                    cbuf[ cs ] = tc   ;
+                    ;
+                    int ti = index[nn];
+                    index[ nn ] = index[m]   ;
+                    index[ m ] = ti   ;
+                    ;
+                    cs = (cs+1)  ;
+                    swap_rc(g,m,nn,p)  ;
+                    double c = g[((na*p)+na)];
+                    double v = 0.0;
+                    for( int i = 0 ; !(i == na) ; i = (i+1) ) {
+                        double t = g[((na*p)+i)];
+                        for( int s = 0 ; !(s == i) ; s = (s+1) ) {
+                            t = (t-(l[((i*kmax)+s)]*l[((na*kmax)+s)]))  ;
+                        }  ;
+                        double li = (t/l[((i*kmax)+i)]);
+                        l[ ((na*kmax)+i) ] = li   ;
+                        v = (v+(li*li))  ;
+                        ;
+                        ;
+                    }  ;
+                    double dd = sqrt(( ( (c-v) < 0.0 ) ? ((v-c)) : ((c-v)) ));
+                    double dg = ( ( dd > (1.0*eps) ) ? (dd) : ((1.0*eps)) );
+                    l[ ((na*kmax)+na) ] = dg   ;
+                    if( dg < 1e-7 ) {
+                        cs = (cs-1)  ;
+                        cbuf[ cs ] = 0.0   ;
+                        double tc = cbuf[(cs+cidx)];
+                        cbuf[ (cs+cidx) ] = cbuf[cs]   ;
+                        cbuf[ cs ] = tc   ;
+                        ;
+                        degenerate = 1  ;
+                    } else {
+                        active[ na ] = index[na]   ;
+                        na = (na+1)  ;
+                    }  ;
+                    ;
+                    ;
+                    ;
+                    ;
+                } else {
+                    false  ;
+                }  ;
+                if( degenerate == 0 ) {
+                    if( (method == 1 && niter > 0 && prevalpha < alpha) ) {
+                        done = 1  ;
+                    } else {
+                        double aa = 0.0;
+                        int clen2 = (p-cs);
+                        for( int i = 0 ; !(i == na) ; i = (i+1) ) {
+                            double t = sign[i];
+                            for( int s = 0 ; !(s == i) ; s = (s+1) ) {
+                                t = (t-(l[((i*kmax)+s)]*zz[s]))  ;
+                            }  ;
+                            zz[ i ] = (t/l[((i*kmax)+i)])   ;
+                            ;
+                        }  ;
+                        for( int ii = 0 ; !(ii == na) ; ii = (ii+1) ) {
+                            int i = ((na-1)-ii);
+                            double t = zz[i];
+                            for( int s = (i+1) ; !(s >= na) ; s = (s+1) ) {
+                                t = (t-(l[((s*kmax)+i)]*ls[s]))  ;
+                            }  ;
+                            ls[ i ] = (t/l[((i*kmax)+i)])   ;
+                            ;
+                            ;
+                        }  ;
+                        if( (na == 1 && ls[0] == 0.0) ) {
+                            ls[ 0 ] = 1.0   ;
+                            aa = 1.0  ;
+                        } else {
+                            double sm = 0.0;
+                            for( int i = 0 ; !(i == na) ; i = (i+1) ) {
+                                sm = (sm+(ls[i]*sign[i]))  ;
+                            }  ;
+                            aa = (1.0/sqrt(sm))  ;
+                            for( int i = 0 ; !(i == na) ; i = (i+1) ) {
+                                ls[ i ] = (aa*ls[i])   ;
+                            }  ;
+                            ;
+                        }  ;
+                        for( int t = 0 ; !(t == clen2) ; t = (t+1) ) {
+                            double acc = 0.0;
+                            for( int i = 0 ; !(i == na) ; i = (i+1) ) {
+                                acc = (acc+(g[((i*p)+(na+t))]*ls[i]))  ;
+                            }  ;
+                            dir[ t ] = (round_even(double((acc*1e+15)))/1e+15)   ;
+                            ;
+                        }  ;
+                        for( int t = 0 ; !(t == clen2) ; t = (t+1) ) {
+                            ratio[ t ] = ((cabs-cbuf[(cs+t)])/((aa-dir[t])+tiny32))   ;
+                        }  ;
+                        double gfirst = min_pos(ratio,clen2);
+                        for( int t = 0 ; !(t == clen2) ; t = (t+1) ) {
+                            ratio[ t ] = ((cabs+cbuf[(cs+t)])/((aa+dir[t])+tiny32))   ;
+                        }  ;
+                        double gsecond = min_pos(ratio,clen2);
+                        double gam = ( ( ( ( gfirst < gsecond ) ? (gfirst) : (gsecond) ) < (cabs/aa) ) ? (( ( gfirst < gsecond ) ? (gfirst) : (gsecond) )) : ((cabs/aa)) );
+                        int nidx = 0;
+                        drop = 0  ;
+                        for( int i = 0 ; !(i == na) ; i = (i+1) ) {
+                            zz[ i ] = ((0.0-coef[active[i]])/(ls[i]+tiny32))   ;
+                        }  ;
+                        double zpos = min_pos(zz,na);
+                        if( zpos < gam ) {
+                            for( int ii = 0 ; !(ii == na) ; ii = (ii+1) ) {
+                                int i = ((na-1)-ii);
+                                if( zz[i] == zpos ) {
+                                    idx[ nidx ] = i   ;
+                                    nidx = (nidx+1)  ;
+                                } else {
+                                    0  ;
+                                }  ;
+                                ;
+                            }  ;
+                            for( int r = 0 ; !(r == nidx) ; r = (r+1) ) {
+                                int i = idx[r];
+                                sign[ i ] = (0.0-sign[i])   ;
+                                ;
+                            }  ;
+                            if( method == 1 ) {
+                                gam = zpos  ;
+                            } else {
+                                0  ;
+                            }  ;
+                            drop = 1  ;
+                        } else {
+                            0  ;
+                        }  ;
+                        ;
+                        niter = (niter+1)  ;
+                        prevalpha = alpha  ;
+                        for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+                            prev[ j ] = coef[j]   ;
+                            coef[ j ] = 0.0   ;
+                        }  ;
+                        for( int i = 0 ; !(i == na) ; i = (i+1) ) {
+                            int a = active[i];
+                            coef[ a ] = (prev[a]+(gam*ls[i]))   ;
+                            ;
+                        }  ;
+                        if( rec == 1 ) {
+                            for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+                                path[ ((niter*p)+j) ] = coef[j]   ;
+                            }  ;
+                        } else {
+                            0  ;
+                        }  ;
+                        for( int t = 0 ; !(t == clen2) ; t = (t+1) ) {
+                            cbuf[ (cs+t) ] = (cbuf[(cs+t)]-(gam*dir[t]))   ;
+                        }  ;
+                        if( (drop == 1 && method == 1) ) {
+                            for( int r = 0 ; !(r == nidx) ; r = (r+1) ) {
+                                cholesky_delete(l,idx[r],na,kmax)  ;
+                            }  ;
+                            na = (na-1)  ;
+                            for( int r = 0 ; !(r == nidx) ; r = (r+1) ) {
+                                int ii = idx[r];
+                                int dropped = active[ii];
+                                for( int i = ii ; !(i >= na) ; i = (i+1) ) {
+                                    active[ i ] = active[(i+1)]   ;
+                                }  ;
+                                dropv[ r ] = dropped   ;
+                                ;
+                                ;
+                            }  ;
+                            for( int r = 0 ; !(r == nidx) ; r = (r+1) ) {
+                                int ii = idx[r];
+                                for( int i = ii ; !(i >= na) ; i = (i+1) ) {
+                                    int ti = index[i];
+                                    index[ i ] = index[(i+1)]   ;
+                                    index[ (i+1) ] = ti   ;
+                                    ;
+                                    swap_rc(g,i,int((i+1)),p)  ;
+                                }  ;
+                                ;
+                            }  ;
+                            for( int rr = 0 ; !(rr == nidx) ; rr = (rr+1) ) {
+                                int r = ((nidx-1)-rr);
+                                int d = dropv[r];
+                                double t = cov0[d];
+                                for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+                                    t = (t-(g0[((d*p)+j)]*coef[j]))  ;
+                                }  ;
+                                cs = (cs-1)  ;
+                                cbuf[ cs ] = t   ;
+                                ;
+                                ;
+                                ;
+                            }  ;
+                            for( int r = 0 ; !(r == nidx) ; r = (r+1) ) {
+                                int ii = idx[r];
+                                for( int i = ii ; !(i >= (p-1)) ; i = (i+1) ) {
+                                    sign[ i ] = sign[(i+1)]   ;
+                                }  ;
+                                sign[ (p-1) ] = 0.0   ;
+                                ;
+                            }  ;
+                        } else {
+                            0  ;
+                        }  ;
+                        ;
+                        ;
+                        ;
+                        ;
+                    }  ;
+                } else {
+                    0  ;
+                }  ;
+                ;
+            }  ;
+        }  ;
+        ;
+    }  ;
+    ;
+    return niter ;
+}
+
+
+SCM2CPP_FN inline int
+rfe_chol( scm2cpp::cspan<double> g,  scm2cpp::cspan<int> idx,  scm2cpp::span<double> l,  double  alpha,  int  k,  int  p )
+{   for( int j = 0 ; !(j == k) ; j = (j+1) ) {
+        double s = (g[((idx[j]*p)+idx[j])]+alpha);
+        for( int m = 0 ; !(m == j) ; m = (m+1) ) {
+            s = (s-(l[((j*p)+m)]*l[((j*p)+m)]))  ;
+        }  ;
+        l[ ((j*p)+j) ] = sqrt(s)   ;
+        for( int i = (j+1) ; !(i >= k) ; i = (i+1) ) {
+            double t = g[((idx[i]*p)+idx[j])];
+            for( int m = 0 ; !(m == j) ; m = (m+1) ) {
+                t = (t-(l[((i*p)+m)]*l[((j*p)+m)]))  ;
+            }  ;
+            l[ ((i*p)+j) ] = (t/l[((j*p)+j)])   ;
+            ;
+        }  ;
+        ;
+    }  ;
+    return 0 ;
+}
+
+
+SCM2CPP_FN inline int
+rfe_solve( scm2cpp::cspan<double> c,  scm2cpp::cspan<int> idx,  scm2cpp::span<double> w,  scm2cpp::cspan<double> l,  int  k,  int  p )
+{   for( int i = 0 ; !(i == k) ; i = (i+1) ) {
+        double t = c[idx[i]];
+        for( int m = 0 ; !(m == i) ; m = (m+1) ) {
+            t = (t-(l[((i*p)+m)]*w[m]))  ;
+        }  ;
+        w[ i ] = (t/l[((i*p)+i)])   ;
+        ;
+    }  ;
+    for( int ii = 0 ; !(ii == k) ; ii = (ii+1) ) {
+        int i = ((k-1)-ii);
+        double t = w[i];
+        for( int m = (i+1) ; !(m >= k) ; m = (m+1) ) {
+            t = (t-(l[((m*p)+i)]*w[m]))  ;
+        }  ;
+        w[ i ] = (t/l[((i*p)+i)])   ;
+        ;
+        ;
+    }  ;
+    return 0 ;
+}
+
+
+inline int
+rfe( scm2cpp::cspan<double> x,  scm2cpp::cspan<double> y,  scm2cpp::span<double> coef,  scm2cpp::span<int> support,  scm2cpp::span<int> ranking,  double  alpha,  int  nsel,  int  step,  int  n,  int  p )
+{   std::vector<double> xc((n*p),0.0);
+    std::vector<double> yc(n,0.0);
+    std::vector<double> mx(p,0.0);
+    std::vector<double> g((p*p),0.0);
+    std::vector<double> c(p,0.0);
+    std::vector<double> l((p*p),0.0);
+    std::vector<double> w(p,0.0);
+    std::vector<int> idx(p,0);
+    std::vector<int> ord(p,0);
+    std::vector<int> gone(p,0);
+    double ym = 0.0;
+    int rounds = 0;
+    int k = p;
+    for( int i = 0 ; !(i == n) ; i = (i+1) ) {
+        ym = (ym+y[i])  ;
+        for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+            mx[ j ] = (mx[j]+x[((i*p)+j)])   ;
+        }  ;
+    }  ;
+    ym = (ym/double(n))  ;
+    for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+        mx[ j ] = (mx[j]/double(n))   ;
+    }  ;
+    for( int i = 0 ; !(i == n) ; i = (i+1) ) {
+        yc[ i ] = (y[i]-ym)   ;
+        for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+            xc[ ((i*p)+j) ] = (x[((i*p)+j)]-mx[j])   ;
+        }  ;
+    }  ;
+    scm2cpp::blas_syrk(g, xc, 1.0, 0.0, 1, p, n)  ;
+    scm2cpp::blas_gemv(c, xc, yc, 1.0, 0.0, 1, p, n)  ;
+    for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+        support[ j ] = 1   ;
+        ranking[ j ] = 1   ;
+        idx[ j ] = j   ;
+    }  ;
+    rfe_chol(g,idx,l,alpha,p,p)  ;
+    for(  ; !(k <= nsel) ;  ) {
+        rfe_solve(c,idx,w,l,k,p)  ;
+        for( int i = 0 ; !(i == k) ; i = (i+1) ) {
+            double key = (w[i]*w[i]);
+            int jj = (i-1);
+            for(  ; !((jj < 0 || (w[ord[jj]]*w[ord[jj]]) <= key)) ;  ) {
+                ord[ (jj+1) ] = ord[jj]   ;
+                jj = (jj-1)  ;
+            }  ;
+            ord[ (jj+1) ] = i   ;
+            ;
+        }  ;
+        int thr = ( ( step < (k-nsel) ) ? (step) : ((k-nsel)) );
+        for( int r = 0 ; !(r == thr) ; r = (r+1) ) {
+            support[ idx[ord[r]] ] = 0   ;
+        }  ;
+        for( int t = 0 ; !(t == k) ; t = (t+1) ) {
+            gone[ t ] = 0   ;
+        }  ;
+        for( int r = 0 ; !(r == thr) ; r = (r+1) ) {
+            gone[ ord[r] ] = 1   ;
+        }  ;
+        int cur = k;
+        for( int tt = 0 ; !(tt == k) ; tt = (tt+1) ) {
+            int pos = ((k-1)-tt);
+            if( gone[pos] == 1 ) {
+                cholesky_delete(l,pos,cur,p)  ;
+                for( int u = pos ; !(u >= (cur-1)) ; u = (u+1) ) {
+                    idx[ u ] = idx[(u+1)]   ;
+                }  ;
+                cur = (cur-1)  ;
+            } else {
+                0  ;
+            }  ;
+            ;
+        }  ;
+        ;
+        for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+            if( support[j] == 0 ) {
+                ranking[ j ] = (ranking[j]+1)   ;
+            } else {
+                0  ;
+            }  ;
+        }  ;
+        k = (k-thr)  ;
+        rounds = (rounds+1)  ;
+        ;
+    }  ;
+    for( int j = 0 ; !(j == p) ; j = (j+1) ) {
+        coef[ j ] = 0.0   ;
+    }  ;
+    rfe_solve(c,idx,w,l,k,p)  ;
+    for( int i = 0 ; !(i == k) ; i = (i+1) ) {
+        coef[ idx[i] ] = w[i]   ;
+    }  ;
+    return rounds ;
+}
+
+
+SCM2CPP_FN inline int
+gram_xx( scm2cpp::cspan<double> x,  scm2cpp::span<double> g,  int  n,  int  p )
+{   scm2cpp::blas_syrk(g, x, 1.0, 0.0, 1, p, n)  ;
+    return 0 ;
+}
+
+
+SCM2CPP_FN inline int
+gram_xy( scm2cpp::cspan<double> x,  scm2cpp::cspan<double> y,  scm2cpp::span<double> c,  int  n,  int  p,  int  q )
+{   scm2cpp::blas_gemm(c, x, y, 1.0, 0.0, 1, 0, p, q, n)  ;
+    return 0 ;
+}
+#endif // GRAMOPS_HPP