File: [local] / OpenXM / src / hgm / mh / src / wmain.c (download)
Revision 1.20, Sat Mar 15 00:43:47 2014 UTC (10 years, 6 months ago) by takayama
Branch: MAIN
Changes since 1.19: +113 -27
lines
A parser of idata is added.
strategy=1 is the default.
x0value_min is changed to 1e-30.
Output several diagnostic data (estimated step size, ...)
|
/*
$OpenXM: OpenXM/src/hgm/mh/src/wmain.c,v 1.20 2014/03/15 00:43:47 takayama Exp $
License: LGPL
*/
#include <stdio.h>
#include <stdlib.h>
#include <math.h>
#include <string.h>
#include "sfile.h"
#include "mh.h"
#define SMAX 4096
#define inci(i) { i++; if (i >= argc) { fprintf(stderr,"Option argument is not given.\n"); return(NULL); }}
int MH_deallocate=0;
/*
changelog
2014.03.15 --strategy 1 is default. A new parser in setParam()
*/
extern char *MH_Gfname;
extern char *MH_Dfname;
/* global variables. They are set in setParam() */
int MH_byFile=1;
int MH_RANK;
int MH_M;
int MH_Mg; /* m */
double *MH_Beta; /* beta[0], ..., beta[m-1] */
double *MH_Ng; /* freedom n. c=(m+1)/2+n/2; Note that it is a pointer */
double MH_X0g; /* initial point */
static double *Iv; /* Initial values of mhg sorted by mhbase() in rd.rr at beta*x0 */
static double Ef; /* exponential factor at beta*x0 */
extern double MH_Hg; /* step size of rk defined in rk.c */
int MH_Dp; /* Data sampling period */
static double Xng=0.0; /* the last point */
int MH_RawName = 0;
static int Testrank=0;
/* If MH_success is set to 1, then strategy, MH_abserr, MH_relerr seem to
be properly set.
*/
int MH_success=0;
/*
Estimation of the maximal coeff of A in y'=Ay.
This might be too rough estimate.
*/
double MH_coeff_max;
/*
Estimation of h by MH_coeff_max;
*/
double MH_estimated_start_step;
extern int MH_Verbose;
extern int MH_strategy;
extern double MH_abserr;
extern double MH_relerr;
extern int MH_P95; /* 95 % points */
int mh_gopen_file(void);
static int setParamTest(void);
static int setParamDefault(void);
static int setParam(char *fname);
static int showParam(void);
static int next(struct SFILE *fp,char *s, char *msg);
static double estimateHg(int m, double beta[],double x0);
/* #define DEBUG */
#ifdef DEBUG
char *MH_Dfname; char *MH_Gfname; double MH_Hg;
int mh_gopen_file(void) { }
struct MH_RESULT mh_rkmain(double x0,double y0[],double xn) { }
#endif
void mh_freeWorkArea(void) {
extern int MH_deallocate;
MH_deallocate=1; /* switch to deallocation mode. */
mh_main(0,NULL);
setParam(NULL);
mh_rkmain(0.0, NULL, 0.0);
mh_rf(0.0, NULL, 0, NULL, 0);
MH_deallocate=0; /* switch to the normal mode. */
}
static int mypower(int x,int n) {
int a,i;
a = 1;
for (i=0; i<n; i++) a = a*x;
return(a);
}
#ifdef STANDALONE2
main(int argc,char *argv[]) {
int strategy=STRATEGY_DEFAULT;
double err[2]={-1.0,-1.0};
int i;
mh_exit(MH_RESET_EXIT); /* standalone mode */
/* mh_main(argc,argv);
mh_freeWorkArea(); */
mh_main(argc,argv);
/* showParam(); */
return(0);
}
#endif
struct MH_RESULT *mh_main(int argc,char *argv[]) {
static double *y0=NULL;
double x0,xn;
double ef;
int i,rank;
struct MH_RESULT *rp=NULL;
extern int MH_deallocate;
extern int MH_byFile;
MH_byFile=1;
if (MH_deallocate) { if (y0) mh_free(y0); return(rp); }
setParam(NULL); MH_Gfname = MH_Dfname = NULL; MH_Verbose=1;
for (i=1; i<argc; i++) {
if (strcmp(argv[i],"--idata")==0) {
inci(i);
setParam(argv[i]); MH_Verbose=0;
}else if (strcmp(argv[i],"--gnuplotf")==0) {
inci(i);
MH_Gfname = (char *)mh_malloc(SMAX);
strcpy(MH_Gfname,argv[i]);
}else if (strcmp(argv[i],"--dataf")==0) {
inci(i);
MH_Dfname = (char *)mh_malloc(SMAX);
strcpy(MH_Dfname,argv[i]);
}else if (strcmp(argv[i],"--xmax")==0) {
inci(i);
sscanf(argv[i],"%lf",&Xng);
}else if (strcmp(argv[i],"--step")==0) {
inci(i);
sscanf(argv[i],"%lg",&MH_Hg);
}else if (strcmp(argv[i],"--help")==0) {
mh_usage(); return(rp);
}else if (strcmp(argv[i],"--raw")==0) {
MH_RawName = 1;
}else if (strcmp(argv[i],"--test")==0) {
inci(i);
sscanf(argv[i],"%d",&Testrank);
setParamTest();
}else if (strcmp(argv[i],"--95")==0) {
MH_P95=1;
}else if (strcmp(argv[i],"--verbose")==0) {
MH_Verbose=1;
}else if (strcmp(argv[i],"--bystring")==0) {
MH_byFile = 0;
}else if (strcmp(argv[i],"--strategy")==0) {
i++; sscanf(argv[i],"%d",&MH_strategy);
}else if (strcmp(argv[i],"--abserr")==0) {
i++; sscanf(argv[i],"%lg",&MH_abserr);
}else if (strcmp(argv[i],"--relerr")==0) {
i++; sscanf(argv[i],"%lg",&MH_relerr);
}else {
fprintf(stderr,"Unknown option %s\n",argv[i]);
mh_usage();
return(rp);
}
}
x0 = MH_X0g;
xn = Xng;
ef = Ef;
rank = mypower(2,MH_Mg);
y0 = (double *) mh_malloc(sizeof(double)*rank);
for (i=0; i<rank; i++) y0[i] = ef*Iv[i];
mh_gopen_file();
rp = (struct MH_RESULT*) mh_malloc(sizeof(struct MH_RESULT));
if (MH_strategy) {
if (MH_abserr > SIGDIGIT_DEFAULT*myabs(y0[0])) {
MH_success = 0;
fprintf(stderr,"%%%%Warning, abserr seems not to be small enough, abserr=%lg, y[0]=%lg\n",MH_abserr,y0[0]);
}else{
MH_success = 1;
}
}else{
MH_success = 0;
}
MH_estimated_start_step = estimateHg(MH_Mg,MH_Beta,MH_X0g);
if (MH_Verbose) showParam();
if (MH_Verbose) {for (i=0; i<rank; i++) printf("%lf\n",y0[i]); }
*rp=mh_rkmain(x0,y0,xn);
return(rp);
}
int mh_usage() {
fprintf(stderr,"Usages:\n");
fprintf(stderr,"hgm_w-n [--idata input_data_file --gnuplotf gnuplot_file_name\n");
fprintf(stderr," --dataf output_data_file --raw --xmax xmax --test m --step h]\n");
fprintf(stderr,"[ --95 --verbose] \n");
fprintf(stderr,"[ --strategy s --abserr ae --relerr re] \n");
fprintf(stderr,"s:0 rk, s:1 adaptive, s:2 adaptive&multiply, see rk.c for the default value of ae and re.\n");
fprintf(stderr,"strategy default = %d\n",MH_strategy);
fprintf(stderr,"--raw does not add data parameters to the output_data_file.\n");
fprintf(stderr,"\nThe command hgm_w-n [options] evaluates Pr({y | y<xmax}), which is the cumulative distribution function of the largest root of the m by m Wishart matrix with n degrees of freedom and the covariantce matrix sigma.\n");
fprintf(stderr,"All the eigenvalues of sigma must be simple.\n");
fprintf(stderr,"Parameters are specified by the input_data_file.\n");
fprintf(stderr,"Parameters are redefined when they appear more than once in the idata file and the command line options.\n");
fprintf(stderr,"The format of the input_data_file, which should be generated by the command hgm_jack-n.\n");
fprintf(stderr," MH_Mg: m, MH_Beta: beta=sigma^(-1)/2 (diagonized), MH_Ng: n, MH_X0g: starting value of x,\n");
fprintf(stderr," Iv: initial values at MH_X0g*MH_Beta (see our paper how to order them), \n");
fprintf(stderr," Ef: a scalar factor to the initial value. It may set to 1.\n");
fprintf(stderr," MH_Hg: h (step size),\n");
fprintf(stderr," MH_Dp: output data is stored in every MH_Dp steps when output_data_file is specified.\n");
fprintf(stderr," Xng: terminating value of x.\n");
fprintf(stderr," --95: output the 95%% point. --verbose: verbose mode.\n");
fprintf(stderr," The line started with %% is a comment line.\n");
fprintf(stderr," An example format of the input_data_file can be obtained by executing hgm_jack-n with no option.\n");
fprintf(stderr,"When --idata option is used, this command is quiet. Use --verbose option if you want to see some messages.\n");
fprintf(stderr,"\nExamples:\n");
fprintf(stderr,"[1] ./hgm_w-n \n");
fprintf(stderr,"[2] ./hgm_w-n --xmax 20\n");
fprintf(stderr,"[3] ./hgm_w-n --test 6\n");
fprintf(stderr," A test run in Mg=6.\n");
fprintf(stderr,"[4] ./hgm_jack-n --idata Testdata/tmp-idata3.txt --degree 15 >t.txt\n");
fprintf(stderr," ./hgm_w-n --idata t.txt --gnuplotf test-g --verbose\n");
fprintf(stderr," gnuplot -persist <test-g-gp.txt\n");
fprintf(stderr," tmp-idata3.txt is a sample input data distributed with this file.\n");
fprintf(stderr," test-g-gp.txt is an input file of the gnuplot\n");
fprintf(stderr," test-g is the table of x and the values of Pr({y | y<x}).\n");
}
static int setParamTest() {
int rank;
int i;
extern int Testrank;
extern int MH_Verbose;
MH_Verbose=1;
MH_M= MH_Mg = Testrank ;
MH_RANK = rank = mypower(2,MH_Mg);
MH_Beta = (double *)mh_malloc(sizeof(double)*MH_Mg);
for (i=0; i<MH_Mg; i++) MH_Beta[i] = 1.0+0.1*i;
MH_Ng = (double *)mh_malloc(sizeof(double)); *MH_Ng = 3.0;
Iv = (double *)mh_malloc(sizeof(double)*rank);
for (i=0; i<rank; i++) Iv[i] = 0;
Iv[0] = 0.001;
Ef = 1;
MH_X0g = 0.3;
MH_Hg = 0.001;
MH_Dp = 1;
Xng = 10.0;
}
static int setParamDefault() {
int rank;
MH_M=MH_Mg = 2 ;
MH_RANK=rank = mypower(2,MH_Mg);
MH_Beta = (double *)mh_malloc(sizeof(double)*MH_Mg);
MH_Beta[0] = 1.0; MH_Beta[1] = 2.0;
MH_Ng = (double *)mh_malloc(sizeof(double)); *MH_Ng = 3.0;
Iv = (double *)mh_malloc(sizeof(double)*rank);
Iv[0] = 1.58693;
Iv[1] = 0.811369;
Iv[2] = 0.846874;
Iv[3] = 0.413438;
Ef = 0.01034957388338225707;
MH_X0g = 0.3;
MH_Hg = 0.001;
MH_Dp = 1;
Xng = 10.0;
}
static int next(struct SFILE *sfp,char *s,char *msg) {
s[0] = '%';
while (s[0] == '%') {
if (!mh_fgets(s,SMAX,sfp)) {
fprintf(stderr,"Data format error at %s\n",msg);
mh_exit(-1);
}
if (s[0] != '%') return(0);
}
}
static int setParam(char *fname) {
int rank;
char s[SMAX];
struct SFILE *fp;
int i;
struct mh_token tk;
extern int MH_deallocate;
extern int MH_byFile;
if (MH_deallocate) {
if (MH_Beta) mh_free(MH_Beta);
if (MH_Ng) mh_free(MH_Ng);
if (Iv) mh_free(Iv);
return(0);
}
if (fname == NULL) return(setParamDefault());
if ((fp=mh_fopen(fname,"r",MH_byFile)) == NULL) {
fprintf(stderr,"File %s is not found.\n",fname);
mh_exit(-1);
}
next(fp,s,"MH_Mg(m)");
sscanf(s,"%d",&MH_Mg); MH_M=MH_Mg;
MH_RANK=rank = mypower(2,MH_Mg);
MH_Beta = (double *)mh_malloc(sizeof(double)*MH_Mg);
for (i=0; i<MH_Mg; i++) {
next(fp,s,"MH_Beta");
sscanf(s,"%lf",&(MH_Beta[i]));
}
MH_Ng = (double *)mh_malloc(sizeof(double));
next(fp,s,"MH_Ng(freedom parameter n)");
sscanf(s,"%lf",MH_Ng);
next(fp,s,"MH_X0g(initial point)");
sscanf(s,"%lf",&MH_X0g);
Iv = (double *)mh_malloc(sizeof(double)*rank);
for (i=0; i<rank; i++) {
next(fp,s,"Iv(initial values)");
sscanf(s,"%lg",&(Iv[i]));
}
next(fp,s,"Ef(exponential factor)");
sscanf(s,"%lg",&Ef);
next(fp,s,"MH_Hg (step size of rk)");
sscanf(s,"%lg",&MH_Hg);
next(fp,s,"MH_Dp (data sampling period)");
sscanf(s,"%d",&MH_Dp);
next(fp,s,"Xng (the last point, cf. --xmax)");
sscanf(s,"%lf",&Xng);
/* Reading the optional parameters */
while ((tk = mh_getoken(s,SMAX-1,fp)).type != MH_TOKEN_EOF) {
/* expect ID */
if (tk.type != MH_TOKEN_ID) {
fprintf(stderr,"Syntax error at %s\n",s); mh_exit(-1);
}
if ((strcmp(s,"abserr")==0) || (strcmp(s,"abserror")==0)) {
if (mh_getoken(s,SMAX-1,fp).type != MH_TOKEN_EQ) {
fprintf(stderr,"Syntax error at %s\n",s); mh_exit(-1);
}
if ((tk=mh_getoken(s,SMAX-1,fp)).type != MH_TOKEN_DOUBLE) {
fprintf(stderr,"Syntax error at %s\n",s); mh_exit(-1);
}
MH_abserr = tk.dval;
continue;
}
if ((strcmp(s,"relerr")==0) || (strcmp(s,"relerror")==0)) {
if (mh_getoken(s,SMAX-1,fp).type != MH_TOKEN_EQ) {
fprintf(stderr,"Syntax error at %s\n",s); mh_exit(-1);
}
if ((tk=mh_getoken(s,SMAX-1,fp)).type != MH_TOKEN_DOUBLE) {
fprintf(stderr,"Syntax error at %s\n",s); mh_exit(-1);
}
MH_relerr = tk.dval;
continue;
}
if (strcmp(s,"strategy")==0) {
if (mh_getoken(s,SMAX-1,fp).type != MH_TOKEN_EQ) {
fprintf(stderr,"Syntax error at %s\n",s); mh_exit(-1);
}
if ((tk=mh_getoken(s,SMAX-1,fp)).type != MH_TOKEN_INT) {
fprintf(stderr,"Syntax error at %s\n",s); mh_exit(-1);
}
MH_strategy = tk.ival;
continue;
}
fprintf(stderr,"Unknown ID at %s\n",s); mh_exit(-1);
}
mh_fclose(fp);
}
static int showParam() {
int rank,i;
extern int MH_strategy;
extern double MH_abserr;
extern double MH_relerr;
rank = mypower(2,MH_Mg);
printf("%%MH_Mg=\n%d\n",MH_Mg);
for (i=0; i<MH_Mg; i++) {
printf("%%MH_Beta[%d]=\n%lf\n",i,MH_Beta[i]);
}
printf("%%MH_Ng=\n%lf\n",*MH_Ng);
printf("%%MH_X0g=\n%lf\n",MH_X0g);
for (i=0; i<rank; i++) {
printf("%%Iv[%d]=\n%lg\n",i,Iv[i]);
}
printf("%%Ef=\n%lf\n",Ef);
printf("%%MH_Hg=\n%lf\n",MH_Hg);
printf("%%MH_Dp=\n%d\n",MH_Dp);
printf("%%Xng=\n%lf\n",Xng);
printf("%%strategy=%d\n",MH_strategy);
printf("%%abserr=%lg, %%relerr=%lg\n",MH_abserr,MH_relerr);
printf("#MH_success=%d\n",MH_success);
printf("#MH_coeff_max=%lg\n",MH_coeff_max);
printf("#MH_estimated_start_step=%lg\n",MH_estimated_start_step);
}
static double estimateHg(int m, double beta[],double x0) {
int i,j;
double dmin;
double cmax;
double h;
/* mynote on 2014.03.15 */
if (m>1) dmin = myabs(beta[1]-beta[0]);
else dmin=myabs(beta[0]);
for (i=0; i<m; i++) {
for (j=i+1; j<m; j++) {
if (myabs(beta[i]-beta[j]) < dmin) dmin = myabs(beta[i]-beta[j]);
}
}
dmin = dmin*x0*2;
cmax = 1.0;
for (i=0; i<m; i++) cmax = cmax*dmin;
cmax = 1.0/cmax;
MH_coeff_max=cmax;
h = exp(log(MH_abserr/cmax)/5.0);
MH_estimated_start_step = h;
return h;
}