chol_gauss.c

The kernel-ica package is a Matlab program that implements the Kernel ICA algorithm for independent

/*
 CHOL_GAUSS  - incomplete Cholesky decomposition of the Gram matrix defined
                by data x, with the Gaussian kernel with width sigma
                Symmetric pivoting is used and the algorithm stops 
                when the sum of the remaining pivots is less than TOL.
*/

#include "mex.h"
#include <math.h>
    
 void mexFunction(int nlhs,
                              mxArray *plhs[],
                              int nrhs,
                              const mxArray *prhs[])
         
             {
double *z,a,b,c,maxdiagG;
double sigma,tol,*temp,*diagG,*G, *Gbis;
int m, n,i,j,jast;
int iter;
int *pp;
int nmax;
double *x, *y, residual;

m = mxGetM(prhs[0]); /* dimension of input space might be greater than 1*/
n = mxGetN(prhs[0]); /* number of samples */
x = mxGetPr(prhs[0]); 
temp=mxGetPr(prhs[1]);
sigma=*temp;
temp=mxGetPr(prhs[2]);
tol=*temp;
if (nrhs>3)
	{
	temp=mxGetPr(prhs[3]);
	nmax=*temp;
	if (nmax==0) nmax=20*3*m/2; else nmax+=1+nmax/8;
	}
	else nmax=20*3*m/2; 



/*
mexPrintf("nmax= %d\n\n",nmax);
mexPrintf("sigma= %f\n",sigma);
mexPrintf("tol= %f\n\n",tol);
mexPrintf("n= %d\n",n);
mexPrintf("m= %d\n\n",m);*/

diagG= (double*) calloc (n,sizeof(double));
G= (double*) calloc (nmax*n,sizeof(double));
pp= (int*) calloc (n,sizeof(int));


iter=0;
residual=n;
for (i=0;i<=n-1;i++)  pp[i]=i;
for (i=0;i<=n-1;i++)  diagG[i]=1;

jast=0;

while ( residual > tol)
{
if (iter==(nmax-1))
	{
	/* need to reallocate memory to G */
	nmax+=nmax/2;
      Gbis= (double*) calloc (nmax*n,sizeof(double));
	for (i=0;i<iter*n;i++) Gbis[i]=G[i];
	free(G);
	G=Gbis;
	}


/* switches already calculated elements of G and order in pp */
if (jast!=iter)
	{
	i=pp[jast];  pp[jast]=pp[iter];  pp[iter]=i;
	for (i=0;i<=iter;i++)
		{
		a=G[jast+n*i];  G[jast+n*i]=G[iter+n*i];  G[iter+n*i]=a;
		}
	}


G[iter*(n+1)]=sqrt(diagG[jast]);
/*mexPrintf("pivot=%f\n",G[iter*n+iter]);
mexPrintf("pivot=%f\n",diagG[jast]);
*/
a=-.5/sigma/sigma;

for (i=iter+1; i<=n-1; i++) 
	{
	if (m<=1)
		b=(x[pp[iter]]-x[pp[i]])*(x[pp[iter]]-x[pp[i]]);
	else
		{
		b=0.0;
		for (j=0;j<=m-1;j++)
			{
			c=x[j+m*pp[iter]]-x[j+m*pp[i]];
			b+=c*c;
			}
		}
	G[i+n*iter]=exp(a*b);
	}

if (iter>0)
	for (j=0; j<=iter-1; j++)
		for (i=iter+1; i<=n-1; i++) G[i+n*iter]-=G[i+n*j]*G[iter+n*j];

for (i=iter+1; i<=n-1; i++) 
	{
	G[i+n*iter]/=G[iter*(n+1)];

	}
residual=0.0;
jast=iter+1;
maxdiagG=0;
for (i=iter+1; i<=n-1; i++)
	{
	b=1.0;
	for (j=0;j<=iter;j++)
		{
		 b-=G[i+j*n]*G[i+j*n];
		}
      diagG[i]=b;
	if (b>maxdiagG)
		{
		jast=i;
		maxdiagG=b;
		}
      residual+=b;
	} 

iter++;
}

plhs[0]=mxCreateDoubleMatrix(n,iter,0); 
z= mxGetPr(plhs[0]); 
for (i=0;i<=n*iter-1;i++) z[i]=G[i];


plhs[1]=mxCreateDoubleMatrix(1,n,0); 
z= mxGetPr(plhs[1]); 
for (i=0;i<=n-1;i++) z[i]=0.0+pp[i];



free(diagG);
free(G);
free(pp);

             }