/********************************************************************
* *
- * THIS FILE IS PART OF THE Ogg Vorbis SOFTWARE CODEC SOURCE CODE. *
- * USE, DISTRIBUTION AND REPRODUCTION OF THIS SOURCE IS GOVERNED BY *
- * THE GNU PUBLIC LICENSE 2, WHICH IS INCLUDED WITH THIS SOURCE. *
- * PLEASE READ THESE TERMS DISTRIBUTING. *
+ * THIS FILE IS PART OF THE OggVorbis SOFTWARE CODEC SOURCE CODE. *
+ * USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
+ * GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
+ * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
- * THE OggSQUISH SOURCE CODE IS (C) COPYRIGHT 1994-2000 *
- * by Monty <monty@xiph.org> and The XIPHOPHORUS Company *
- * http://www.xiph.org/ *
+ * THE OggVorbis SOURCE CODE IS (C) COPYRIGHT 1994-2001 *
+ * by the Xiph.Org Foundation http://www.xiph.org/ *
* *
********************************************************************
function: function calls to collect codebook metrics
- last mod: $Id: metrics.c,v 1.5 2000/01/28 09:05:20 xiphmont Exp $
********************************************************************/
#include <math.h>
#include "bookutil.h"
-/* set up metrics */
+/* collect the following metrics:
-double meanamplitude_acc=0.;
-double meanamplitudesq_acc=0.;
-double meanerror_acc=0.;
-double meanerrorsq_acc=0.;
-double meandev_acc=0.;
+ mean and mean squared amplitude
+ mean and mean squared error
+ mean and mean squared error (per sample) by entry
+ worst case fit by entry
+ entry cell size
+ hits by entry
+ total bits
+ total samples
+ (average bits per sample)*/
-double *histogram=NULL;
-double *histogram_error=NULL;
-double *histogram_errorsq=NULL;
-double *histogram_distance=NULL;
-double *histogram_hi=NULL;
-double *histogram_lo=NULL;
-double count=0.;
+/* set up metrics */
-int books=0;
-int dim=-1;
-double *work;
-
-int histerrsort(const void *a, const void *b){
- double av=histogram_distance[*((long *)a)];
- double bv=histogram_distance[*((long *)b)];
- if(av<bv)return(-1);
- return(1);
+float meanamplitude_acc=0.f;
+float meanamplitudesq_acc=0.f;
+float meanerror_acc=0.f;
+float meanerrorsq_acc=0.f;
+
+float **histogram=NULL;
+float **histogram_error=NULL;
+float **histogram_errorsq=NULL;
+float **histogram_hi=NULL;
+float **histogram_lo=NULL;
+float bits=0.f;
+float count=0.f;
+
+static float *_now(codebook *c, int i){
+ return c->valuelist+i*c->c->dim;
}
+int books=0;
+
void process_preprocess(codebook **bs,char *basename){
- while(bs[books]){
- codebook *b=bs[books];
- if(dim==-1){
- dim=b->c->dim;
- work=malloc(sizeof(double)*dim);
- }else{
- if(dim!=b->c->dim){
- fprintf(stderr,"Each codebook in a cascade must have the same dimensional order\n");
- exit(1);
- }
- }
- books++;
- }
+ int i;
+ while(bs[books])books++;
if(books){
- const static_codebook *b=bs[books-1]->c;
- histogram=calloc(b->entries,sizeof(double));
- histogram_distance=calloc(b->entries,sizeof(double));
- histogram_errorsq=calloc(b->entries*dim,sizeof(double));
- histogram_error=calloc(b->entries*dim,sizeof(double));
- histogram_hi=calloc(b->entries*dim,sizeof(double));
- histogram_lo=calloc(b->entries*dim,sizeof(double));
+ histogram=_ogg_calloc(books,sizeof(float *));
+ histogram_error=_ogg_calloc(books,sizeof(float *));
+ histogram_errorsq=_ogg_calloc(books,sizeof(float *));
+ histogram_hi=_ogg_calloc(books,sizeof(float *));
+ histogram_lo=_ogg_calloc(books,sizeof(float *));
}else{
fprintf(stderr,"Specify at least one codebook\n");
exit(1);
}
+
+ for(i=0;i<books;i++){
+ codebook *b=bs[i];
+ histogram[i]=_ogg_calloc(b->entries,sizeof(float));
+ histogram_error[i]=_ogg_calloc(b->entries*b->dim,sizeof(float));
+ histogram_errorsq[i]=_ogg_calloc(b->entries*b->dim,sizeof(float));
+ histogram_hi[i]=_ogg_calloc(b->entries*b->dim,sizeof(float));
+ histogram_lo[i]=_ogg_calloc(b->entries*b->dim,sizeof(float));
+ }
+}
+
+static float _dist(int el,float *a, float *b){
+ int i;
+ float acc=0.f;
+ for(i=0;i<el;i++){
+ float val=(a[i]-b[i]);
+ acc+=val*val;
+ }
+ return acc;
}
-void process_postprocess(codebook **b,char *basename){
- int i,j,k;
+void cell_spacing(codebook *c){
+ int j,k;
+ float min=-1.f,max=-1.f,mean=0.f,meansq=0.f;
+ long total=0;
+
+ /* minimum, maximum, mean, ms cell spacing */
+ for(j=0;j<c->c->entries;j++){
+ if(c->c->lengthlist[j]>0){
+ float localmin=-1.;
+ for(k=0;k<c->c->entries;k++){
+ if(c->c->lengthlist[k]>0){
+ float this=_dist(c->c->dim,_now(c,j),_now(c,k));
+ if(j!=k &&
+ (localmin==-1 || this<localmin))
+ localmin=this;
+ }
+ }
+
+ if(min==-1 || localmin<min)min=localmin;
+ if(max==-1 || localmin>max)max=localmin;
+ mean+=sqrt(localmin);
+ meansq+=localmin;
+ total++;
+ }
+ }
+
+ fprintf(stderr,"\tminimum cell spacing (closest side): %g\n",sqrt(min));
+ fprintf(stderr,"\tmaximum cell spacing (closest side): %g\n",sqrt(max));
+ fprintf(stderr,"\tmean closest side spacing: %g\n",mean/total);
+ fprintf(stderr,"\tmean sq closest side spacing: %g\n",sqrt(meansq/total));
+}
+
+void process_postprocess(codebook **bs,char *basename){
+ int i,k,book;
char *buffer=alloca(strlen(basename)+80);
- codebook *bb=b[books-1];
-
- fprintf(stderr,"Done. Processed %ld data points for %ld entries:\n",
- (long)count,bb->c->entries);
- fprintf(stderr,"\tglobal mean amplitude: %g\n",
- meanamplitude_acc/(count*dim));
- fprintf(stderr,"\tglobal mean squared amplitude: %g\n",
- sqrt(meanamplitudesq_acc/(count*dim)));
-
- fprintf(stderr,"\tglobal mean error: %g\n",
- meanerror_acc/(count*dim));
- fprintf(stderr,"\tglobal mean squared error: %g\n",
- sqrt(meanerrorsq_acc/(count*dim)));
- fprintf(stderr,"\tglobal mean deviation: %g\n",
- meandev_acc/(count*dim));
- {
+
+ fprintf(stderr,"Done. Processed %ld data points:\n\n",
+ (long)count);
+
+ fprintf(stderr,"Global statistics:******************\n\n");
+
+ fprintf(stderr,"\ttotal samples: %ld\n",(long)count);
+ fprintf(stderr,"\ttotal bits required to code: %ld\n",(long)bits);
+ fprintf(stderr,"\taverage bits per sample: %g\n\n",bits/count);
+
+ fprintf(stderr,"\tmean sample amplitude: %g\n",
+ meanamplitude_acc/count);
+ fprintf(stderr,"\tmean squared sample amplitude: %g\n\n",
+ sqrt(meanamplitudesq_acc/count));
+
+ fprintf(stderr,"\tmean code error: %g\n",
+ meanerror_acc/count);
+ fprintf(stderr,"\tmean squared code error: %g\n\n",
+ sqrt(meanerrorsq_acc/count));
+
+ for(book=0;book<books;book++){
FILE *out;
+ codebook *b=bs[book];
+ int n=b->c->entries;
+ int dim=b->c->dim;
+
+ fprintf(stderr,"Book %d statistics:------------------\n",book);
- sprintf(buffer,"%s-mse.m",basename);
+ cell_spacing(b);
+
+ sprintf(buffer,"%s-%d-mse.m",basename,book);
out=fopen(buffer,"w");
if(!out){
fprintf(stderr,"Could not open file %s for writing\n",buffer);
exit(1);
}
- for(i=0;i<bb->c->entries;i++){
- for(k=0;k<bb->c->dim;k++){
- fprintf(out,"%ld, %g, %g\n",
- i*bb->c->dim+k,(bb->valuelist+i*bb->c->dim)[k],
- sqrt((histogram_errorsq+i*bb->c->dim)[k]/histogram[i]));
+ for(i=0;i<n;i++){
+ for(k=0;k<dim;k++){
+ fprintf(out,"%d, %g, %g\n",
+ i*dim+k,(b->valuelist+i*dim)[k],
+ sqrt((histogram_errorsq[book]+i*dim)[k]/histogram[book][i]));
}
}
fclose(out);
- sprintf(buffer,"%s-me.m",basename);
+ sprintf(buffer,"%s-%d-me.m",basename,book);
out=fopen(buffer,"w");
if(!out){
fprintf(stderr,"Could not open file %s for writing\n",buffer);
exit(1);
}
- for(i=0;i<bb->c->entries;i++){
- for(k=0;k<bb->c->dim;k++){
- fprintf(out,"%ld, %g, %g\n",
- i*bb->c->dim+k,(bb->valuelist+i*bb->c->dim)[k],
- (histogram_error+i*bb->c->dim)[k]/histogram[i]);
+ for(i=0;i<n;i++){
+ for(k=0;k<dim;k++){
+ fprintf(out,"%d, %g, %g\n",
+ i*dim+k,(b->valuelist+i*dim)[k],
+ (histogram_error[book]+i*dim)[k]/histogram[book][i]);
}
}
fclose(out);
- sprintf(buffer,"%s-worst.m",basename);
+ sprintf(buffer,"%s-%d-worst.m",basename,book);
out=fopen(buffer,"w");
if(!out){
fprintf(stderr,"Could not open file %s for writing\n",buffer);
exit(1);
}
- for(i=0;i<bb->c->entries;i++){
- for(k=0;k<bb->c->dim;k++){
- fprintf(out,"%ld, %g, %g, %g\n",
- i*bb->c->dim+k,(bb->valuelist+i*bb->c->dim)[k],
- (bb->valuelist+i*bb->c->dim)[k]+(histogram_lo+i*bb->c->dim)[k],
- (bb->valuelist+i*bb->c->dim)[k]+(histogram_hi+i*bb->c->dim)[k]);
+ for(i=0;i<n;i++){
+ for(k=0;k<dim;k++){
+ fprintf(out,"%d, %g, %g, %g\n",
+ i*dim+k,(b->valuelist+i*dim)[k],
+ (b->valuelist+i*dim)[k]+(histogram_lo[book]+i*dim)[k],
+ (b->valuelist+i*dim)[k]+(histogram_hi[book]+i*dim)[k]);
}
}
fclose(out);
}
+}
- {
- FILE *out;
- long *index=alloca(sizeof(long)*bb->c->entries);
- sprintf(buffer,"%s-distance.m",basename);
- out=fopen(buffer,"w");
- if(!out){
- fprintf(stderr,"Could not open file %s for writing\n",buffer);
- exit(1);
+float process_one(codebook *b,int book,float *a,int dim,int step,int addmul,
+ float base){
+ int j,entry;
+ float amplitude=0.f;
+
+ if(book==0){
+ float last=base;
+ for(j=0;j<dim;j++){
+ amplitude=a[j*step]-(b->c->q_sequencep?last:0);
+ meanamplitude_acc+=fabs(amplitude);
+ meanamplitudesq_acc+=amplitude*amplitude;
+ count++;
+ last=a[j*step];
}
- for(j=0;j<bb->c->entries;j++){
- if(histogram[j])histogram_distance[j]/=histogram[j];
- index[j]=j;
+ }
+
+ if(b->c->q_sequencep){
+ float temp;
+ for(j=0;j<dim;j++){
+ temp=a[j*step];
+ a[j*step]-=base;
}
+ base=temp;
+ }
- qsort(index,bb->c->entries,sizeof(long),histerrsort);
+ entry=vorbis_book_besterror(b,a,step,addmul);
- for(j=0;j<bb->c->entries;j++)
- for(k=0;k<histogram[index[j]];k++)
- fprintf(out,"%g,\n",histogram_distance[index[j]]);
- fclose(out);
-
+ if(entry==-1){
+ fprintf(stderr,"Internal error: _best returned -1.\n");
+ exit(1);
}
+ histogram[book][entry]++;
+ bits+=vorbis_book_codelen(b,entry);
+
+ for(j=0;j<dim;j++){
+ float error=a[j*step];
+
+ if(book==books-1){
+ meanerror_acc+=fabs(error);
+ meanerrorsq_acc+=error*error;
+ }
+ histogram_errorsq[book][entry*dim+j]+=error*error;
+ histogram_error[book][entry*dim+j]+=fabs(error);
+ if(histogram[book][entry]==0 || histogram_hi[book][entry*dim+j]<error)
+ histogram_hi[book][entry*dim+j]=error;
+ if(histogram[book][entry]==0 || histogram_lo[book][entry*dim+j]>error)
+ histogram_lo[book][entry*dim+j]=error;
+ }
+ return base;
}
-void process_vector(codebook **bs,double *a){
+
+void process_vector(codebook **bs,int *addmul,int inter,float *a,int n){
int bi;
int i;
- double amplitude=0.;
- double distance=0.;
- double base=0.;
- int entry;
- double *e;
- memcpy(work,a,sizeof(double)*dim);
for(bi=0;bi<books;bi++){
codebook *b=bs[bi];
- entry=codebook_entry(b,work);
- e=b->valuelist+b->c->dim*entry;
- for(i=0;i<b->c->dim;i++)work[i]-=e[i];
- }
+ int dim=b->dim;
+ float base=0.f;
- for(i=0;i<dim;i++){
- double error=work[i];
- if(bs[0]->c->q_sequencep){
- amplitude=a[i]-base;
- base=a[i];
- }else
- amplitude=a[i];
-
- meanamplitude_acc+=fabs(amplitude);
- meanamplitudesq_acc+=amplitude*amplitude;
- meanerror_acc+=fabs(error);
- meanerrorsq_acc+=error*error;
-
- if(amplitude)
- meandev_acc+=fabs(error/amplitude);
- else
- meandev_acc+=fabs(error); /* yeah, yeah */
-
- histogram_errorsq[entry*dim+i]+=error*error;
- histogram_error[entry*dim+i]+=fabs(error);
- if(histogram[entry]==0 || histogram_hi[entry*dim+i]<error)
- histogram_hi[entry*dim+i]=error;
- if(histogram[entry]==0 || histogram_lo[entry*dim+i]>error)
- histogram_lo[entry*dim+i]=error;
- distance+=error*error;
+ if(inter){
+ for(i=0;i<n/dim;i++)
+ base=process_one(b,bi,a+i,dim,n/dim,addmul[bi],base);
+ }else{
+ for(i=0;i<=n-dim;i+=dim)
+ base=process_one(b,bi,a+i,dim,1,addmul[bi],base);
+ }
}
- histogram[entry]++;
- histogram_distance[entry]+=sqrt(distance);
-
- if((long)(count++)%100)spinnit("working.... lines: ",count);
+ if((long)(count)%100)spinnit("working.... samples: ",count);
}
void process_usage(void){
fprintf(stderr,
- "usage: vqmetrics <codebook>.vqh datafile.vqd [datafile.vqd]...\n\n"
- " data can be taken on stdin. Output goes to output files:\n"
- " basename-me.m: gnuplot: mean error by entry value\n"
- " basename-mse.m: gnuplot: mean square error by entry value\n"
- " basename-worst.m: gnuplot: worst error by entry value\n"
- " basename-distance.m: gnuplot file showing distance probability\n"
- "\n");
+ "usage: vqmetrics [-i] +|*<codebook>.vqh [ +|*<codebook.vqh> ]... \n"
+ " datafile.vqd [datafile.vqd]...\n\n"
+ " data can be taken on stdin. -i indicates interleaved coding.\n"
+ " Output goes to output files:\n"
+ " basename-me.m: gnuplot: mean error by entry value\n"
+ " basename-mse.m: gnuplot: mean square error by entry value\n"
+ " basename-worst.m: gnuplot: worst error by entry value\n"
+ " basename-distance.m: gnuplot file showing distance probability\n"
+ "\n");
}