lorenzo.volpi
/
QuaPy
forked from moreo/QuaPy
1
0
Fork 0
Code Pull Requests Activity
QuaPy/svm-perf-quantification-ext...

541 lines
20 KiB
Diff
Raw Blame History

diff -ruN svm_perf/svm_struct/svm_struct_main.c svm_perf_quantification/svm_struct/svm_struct_main.c
--- svm_perf/svm_struct/svm_struct_main.c 2020-10-28 12:23:19.000000000 +0100
+++ svm_perf_quantification/svm_struct/svm_struct_main.c 2020-10-28 12:23:53.000000000 +0100
@@ -128,7 +128,8 @@
struct_parm->newconstretrain=100;
struct_parm->ccache_size=5;
struct_parm->batch_size=100;
-
+ struct_parm->loss_parm=1.0;
+ struct_parm->beta=1.0; // AIC-QBETA
strcpy (modelfile, "svm_struct_model");
strcpy (learn_parm->predfile, "trans_predictions");
strcpy (learn_parm->alphafile, "");
@@ -170,6 +171,7 @@
case 'p': i++; struct_parm->slack_norm=atol(argv[i]); break;
case 'e': i++; struct_parm->epsilon=atof(argv[i]); break;
case 'k': i++; struct_parm->newconstretrain=atol(argv[i]); break;
+ case 'j': i++; struct_parm->beta=atof(argv[i]); break; // AIC-QBETA
case 'h': i++; learn_parm->svm_iter_to_shrink=atol(argv[i]); break;
case '#': i++; learn_parm->maxiter=atol(argv[i]); break;
case 'm': i++; learn_parm->kernel_cache_size=atol(argv[i]); break;
@@ -189,6 +191,7 @@
case '-': strcpy(struct_parm->custom_argv[struct_parm->custom_argc++],argv[i]);i++; strcpy(struct_parm->custom_argv[struct_parm->custom_argc++],argv[i]);break;
case 'v': i++; (*struct_verbosity)=atol(argv[i]); break;
case 'y': i++; (*verbosity)=atol(argv[i]); break;
+ case '!': i++; struct_parm->loss_parm=atof(argv[i]); break;
default: printf("\nUnrecognized option %s!\n\n",argv[i]);
print_help();
exit(0);
@@ -396,6 +399,9 @@
printf(" (in the same order as in the training set)\n");
printf("Application-Specific Options:\n");
print_struct_help();
+ printf("*************************************************\n"); // AIC-QBETA
+ printf(" -j float -> parameter beta for qbeta-based loss functions (default: 1.0)\n");
+ printf("*************************************************\n");
wait_any_key();
printf("\nMore details in:\n");
diff -ruN svm_perf/svm_struct_api.c svm_perf_quantification/svm_struct_api.c
--- svm_perf/svm_struct_api.c 2020-10-28 12:23:19.000000000 +0100
+++ svm_perf_quantification/svm_struct_api.c 2020-10-28 12:23:53.000000000 +0100
@@ -20,6 +20,7 @@
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
+#include <math.h>
#include "svm_struct_api.h"
#include "svm_light/svm_common.h"
#include "svm_struct/svm_struct_common.h"
@@ -27,7 +28,9 @@
#define MAX(x,y) ((x) < (y) ? (y) : (x))
#define MIN(x,y) ((x) > (y) ? (y) : (x))
+#define ABS(x) ((x) < (0) ? (-(x)) : (x))
#define SIGN(x) ((x) > (0) ? (1) : (((x) < (0) ? (-1) : (0))))
+#define PI (3.141592653589793)
int compareup(const void *a, const void *b)
{
@@ -72,6 +75,16 @@
double rocarea(LABEL y, LABEL ybar);
double prbep(LABEL y, LABEL ybar);
double avgprec(LABEL y, LABEL ybar);
+/* AIC-QBETA */
+double gm(int a, int b, int c, int d);
+double nae(int a, int b, int c, int d);
+double ae(int a, int b, int c, int d);
+double rae(int a, int b, int c, int d);
+double Qbeta(int a, int b, int c, int d, double beta);
+double Qbeta_acc(int a, int b, int c, int d, double beta);
+double Qbeta_f1(int a, int b, int c, int d, double beta);
+double Qbeta_gm(int a, int b, int c, int d, double beta);
+/* AIC-QBETA */
double zeroone_loss(int a, int b, int c, int d);
double fone_loss(int a, int b, int c, int d);
@@ -82,6 +95,23 @@
double swappedpairs_loss(LABEL y, LABEL ybar);
double avgprec_loss(LABEL y, LABEL ybar);
+double kldiv(int a, int b, int c, int d); // KLD
+double kldiv_loss(int a, int b, int c, int d); // KLD
+double nkldiv_loss(int a, int b, int c, int d); // KLD
+
+double milli_loss(int a, int b, int c, int d); //MILL
+
+/* AIC-QBETA */
+double gm_loss(int a, int b, int c, int d);
+double nae_loss(int a, int b, int c, int d);
+double ae_loss(int a, int b, int c, int d);
+double rae_loss(int a, int b, int c, int d);
+double Qbeta_loss(int a, int b, int c, int d, double beta);
+double Qbeta_acc_loss(int a, int b, int c, int d, double beta);
+double Qbeta_f1_loss(int a, int b, int c, int d, double beta);
+double Qbeta_gm_loss(int a, int b, int c, int d, double beta);
+/* AIC-QBETA */
+
void svm_struct_learn_api_init(int argc, char* argv[])
{
/* Called in learning part before anything else is done to allow
@@ -181,10 +211,22 @@
/* change label value for better scaling using thresholdmetrics */
if((sparm->loss_function == ZEROONE)
|| (sparm->loss_function == FONE)
+ || (sparm->loss_function == GM) // AIC-QBETA
+ || (sparm->loss_function == NAE) // AIC-QBETA
+ || (sparm->loss_function == AE) // AIC-QBETA
+ || (sparm->loss_function == RAE) // AIC-QBETA
+ || (sparm->loss_function == QBETA) // AIC-QBETA
+ || (sparm->loss_function == QBETA_ACC) // AIC-QBETA
+ || (sparm->loss_function == QBETA_F1) // AIC-QBETA
+ || (sparm->loss_function == QBETA_GM) // AIC-QBETA
|| (sparm->loss_function == ERRORRATE)
|| (sparm->loss_function == PRBEP)
|| (sparm->loss_function == PREC_K)
- || (sparm->loss_function == REC_K)) {
+ || (sparm->loss_function == REC_K)
+ || (sparm->loss_function == KLD)
+ || (sparm->loss_function == NKLD)
+ || (sparm->loss_function == MILLI)
+ ) {
for(i=0;i<sample.examples[0].x.totdoc;i++) {
if(sample.examples[0].y.class[i]>0)
sample.examples[0].y.class[i]=0.5*100.0/(numn+nump);
@@ -520,10 +562,22 @@
LABEL ybar;
if((sparm->loss_function == ZEROONE)
|| (sparm->loss_function == FONE)
+ || (sparm->loss_function == GM) // AIC-QBETA
+ || (sparm->loss_function == NAE) // AIC-QBETA
+ || (sparm->loss_function == AE) // AIC-QBETA
+ || (sparm->loss_function == RAE) // AIC-QBETA
+ || (sparm->loss_function == QBETA) // AIC-QBETA
+ || (sparm->loss_function == QBETA_ACC) // AIC-QBETA
+ || (sparm->loss_function == QBETA_F1) // AIC-QBETA
+ || (sparm->loss_function == QBETA_GM) // AIC-QBETA
|| (sparm->loss_function == ERRORRATE)
|| (sparm->loss_function == PRBEP)
|| (sparm->loss_function == PREC_K)
- || (sparm->loss_function == REC_K)) {
+ || (sparm->loss_function == REC_K)
+ || (sparm->loss_function == KLD)
+ || (sparm->loss_function == NKLD)
+ || (sparm->loss_function == MILLI)
+ ) {
ybar=find_most_violated_constraint_thresholdmetric(x,y,sm,sparm,
sparm->loss_type);
}
@@ -562,9 +616,21 @@
sparm->loss_type); */
else if((sparm->loss_function == ZEROONE)
|| (sparm->loss_function == FONE)
+ || (sparm->loss_function == GM) // AIC-QBETA
+ || (sparm->loss_function == NAE) // AIC-QBETA
+ || (sparm->loss_function == AE) // AIC-QBETA
+ || (sparm->loss_function == RAE) // AIC-QBETA
+ || (sparm->loss_function == QBETA) // AIC-QBETA
+ || (sparm->loss_function == QBETA_ACC) // AIC-QBETA
+ || (sparm->loss_function == QBETA_F1) // AIC-QBETA
+ || (sparm->loss_function == QBETA_GM) // AIC-QBETA
|| (sparm->loss_function == PRBEP)
|| (sparm->loss_function == PREC_K)
- || (sparm->loss_function == REC_K))
+ || (sparm->loss_function == REC_K)
+ || (sparm->loss_function == KLD)
+ || (sparm->loss_function == NKLD)
+ || (sparm->loss_function == MILLI)
+ )
ybar=find_most_violated_constraint_thresholdmetric(x,y,sm,sparm,
sparm->loss_type);
else if((sparm->loss_function == SWAPPEDPAIRS))
@@ -741,7 +807,23 @@
if(sparm->loss_function == ZEROONE)
loss=zeroone_loss(a,numn-d,nump-a,d);
else if(sparm->loss_function == FONE)
- loss=fone_loss(a,numn-d,nump-a,d);
+ loss=fone_loss(a,numn-d,nump-a,d);
+ else if(sparm->loss_function == GM) // AIC-QBETA
+ loss=gm_loss(a,numn-d,nump-a,d);
+ else if(sparm->loss_function == NAE) // AIC-QBETA
+ loss=nae_loss(a,numn-d,nump-a,d);
+ else if(sparm->loss_function == AE) // AIC-QBETA
+ loss=ae_loss(a,numn-d,nump-a,d);
+ else if(sparm->loss_function == RAE) // AIC-QBETA
+ loss=rae_loss(a,numn-d,nump-a,d);
+ else if(sparm->loss_function == QBETA) // AIC-QBETA
+ loss=Qbeta_loss(a,numn-d,nump-a,d,sparm->beta);
+ else if(sparm->loss_function == QBETA_ACC) // AIC-QBETA
+ loss=Qbeta_acc_loss(a,numn-d,nump-a,d,sparm->beta);
+ else if(sparm->loss_function == QBETA_F1) // AIC-QBETA
+ loss=Qbeta_f1_loss(a,numn-d,nump-a,d,sparm->beta);
+ else if(sparm->loss_function == QBETA_GM) // AIC-QBETA
+ loss=Qbeta_gm_loss(a,numn-d,nump-a,d,sparm->beta);
else if(sparm->loss_function == ERRORRATE)
loss=errorrate_loss(a,numn-d,nump-a,d);
else if((sparm->loss_function == PRBEP) && (a+numn-d == nump))
@@ -750,6 +832,12 @@
loss=prec_k_loss(a,numn-d,nump-a,d);
else if((sparm->loss_function == REC_K) && (a+numn-d <= prec_rec_k))
loss=rec_k_loss(a,numn-d,nump-a,d);
+ else if(sparm->loss_function == KLD) //KLD
+ loss=kldiv_loss(a,numn-d,nump-a,d); //KLD
+ else if(sparm->loss_function == NKLD) //KLD
+ loss=nkldiv_loss(a,numn-d,nump-a,d); //KLD
+ else if(sparm->loss_function == MILLI) //MILLI
+ loss=milli_loss(a,numn-d,nump-a,d); //MILLI
else {
loss=0;
}
@@ -1213,6 +1301,7 @@
}
/* printf("%f %f\n",y.class[i],ybar.class[i]); */
}
+ //printf("********** loss %d (a,b,c,d) (%d,%d,%d,%d) beta=%f\n",sparm->loss_function,a,b,c,d,sparm->beta);
/* Return the loss according to the selected loss function. */
if(sparm->loss_function == ZEROONE) { /* type 0 loss: 0/1 loss */
/* return 0, if y==ybar. return 1 else */
@@ -1221,6 +1310,30 @@
else if(sparm->loss_function == FONE) {
loss=fone_loss(a,b,c,d);
}
+ else if(sparm->loss_function == GM) { // AIC-QBETA
+ loss=gm_loss(a,b,c,d);
+ }
+ else if(sparm->loss_function == NAE) { // AIC-QBETA
+ loss=nae_loss(a,b,c,d);
+ }
+ else if(sparm->loss_function == AE) { // AIC-QBETA
+ loss=ae_loss(a,b,c,d);
+ }
+ else if(sparm->loss_function == RAE) { // AIC-QBETA
+ loss=rae_loss(a,b,c,d);
+ }
+ else if(sparm->loss_function == QBETA) { // AIC-QBETA
+ loss=Qbeta_loss(a,b,c,d,sparm->beta);
+ }
+ else if(sparm->loss_function == QBETA_ACC) { // AIC-QBETA
+ loss=Qbeta_acc_loss(a,b,c,d,sparm->beta);
+ }
+ else if(sparm->loss_function == QBETA_F1) { // AIC-QBETA
+ loss=Qbeta_f1_loss(a,b,c,d,sparm->beta);
+ }
+ else if(sparm->loss_function == QBETA_GM) { // AIC-QBETA
+ loss=Qbeta_gm_loss(a,b,c,d,sparm->beta);
+ }
else if(sparm->loss_function == ERRORRATE) {
loss=errorrate_loss(a,b,c,d);
}
@@ -1242,6 +1355,15 @@
else if(sparm->loss_function == AVGPREC) {
loss=avgprec_loss(y,ybar);
}
+ else if(sparm->loss_function == KLD) { //KLD
+ loss=kldiv_loss(a,b,c,d); //KLD
+ } //KLD
+ else if(sparm->loss_function == NKLD) { //KLD
+ loss=nkldiv_loss(a,b,c,d); //KLD
+ } //KLD
+ else if(sparm->loss_function == MILLI) { //MILLI
+ loss=milli_loss(a,b,c,d); //MILLI
+ } //MILLI
else {
/* Put your code for different loss functions here. But then
find_most_violated_constraint_???(x, y, sm) has to return the
@@ -1320,6 +1442,16 @@
printf("PRBEP : %5.2f\n",teststats->prbep);
printf("ROCArea : %5.2f\n",teststats->rocarea);
printf("AvgPrec : %5.2f\n",teststats->avgprec);
+ printf("Qb : %5.2f\n",teststats->Qbeta);
+ printf("Qb (Acc) : %5.2f\n",teststats->Qbeta_acc);
+ printf("Qb (F1) : %5.2f\n",teststats->Qbeta_f1);
+ printf("Qb (GM) : %5.2f\n",teststats->Qbeta_gm);
+ printf("NAE : %5.2f\n",teststats->nae);
+ printf("AE : %5.2f\n",teststats->ae);
+ printf("RAE : %5.2f\n",teststats->rae);
+ printf("GM : %5.2f\n",teststats->gm);
+ printf("KLD : %5.2f\n",teststats->kld);
+ printf("NKLD : %5.2f\n",teststats->nkld);
}
else {
printf("NOTE: %ld test examples are unlabeled, so performance cannot be computed. The\n",teststats->test_data_unlabeled);
@@ -1352,6 +1484,29 @@
teststats->recall=100.0-loss(ex.y,ypred,sparm);
sparm->loss_function=FONE;
teststats->fone=100.0-loss(ex.y,ypred,sparm);
+
+ sparm->loss_function=GM; // AIC-QBETA
+ teststats->gm=100.0-loss(ex.y,ypred,sparm);
+ sparm->loss_function=NAE; // AIC-QBETA
+ teststats->nae=100.0-loss(ex.y,ypred,sparm);
+ sparm->loss_function=AE; // AIC-QBETA
+ teststats->ae=100.0-loss(ex.y,ypred,sparm);
+ sparm->loss_function=RAE; // AIC-QBETA
+ teststats->rae=100.0-loss(ex.y,ypred,sparm);
+ sparm->loss_function=QBETA; // AIC-QBETA
+ teststats->Qbeta=100.0-loss(ex.y,ypred,sparm);
+ sparm->loss_function=QBETA_ACC; // AIC-QBETA
+ teststats->Qbeta_acc=100.0-loss(ex.y,ypred,sparm);
+ sparm->loss_function=QBETA_F1; // AIC-QBETA
+ teststats->Qbeta_f1=100.0-loss(ex.y,ypred,sparm);
+ sparm->loss_function=QBETA_GM; // AIC-QBETA
+ teststats->Qbeta_gm=100.0-loss(ex.y,ypred,sparm);
+
+ sparm->loss_function=KLD; // KLD
+ teststats->kld=100-loss(ex.y,ypred,sparm);
+ sparm->loss_function=NKLD; // KLD
+ teststats->nkld=100.0-loss(ex.y,ypred,sparm);
+
teststats->prbep=prbep(ex.y,ypred);
teststats->rocarea=rocarea(ex.y,ypred);
teststats->avgprec=avgprec(ex.y,ypred);
@@ -1403,6 +1558,7 @@
STRUCTMODEL sm;
sm.svm_model=read_model(file);
+ sparm->beta = 1; // AIC-QBETA *****************************
sparm->loss_function=ERRORRATE;
sparm->bias=0;
sparm->bias_featurenum=0;
@@ -1514,6 +1670,18 @@
printf(" %2d Prec@k: 100 minus precision at k in percent.\n",PREC_K);
printf(" %2d Rec@k: 100 minus recall at k in percent.\n",REC_K);
printf(" %2d ROCArea: Percentage of swapped pos/neg pairs (i.e. 100 - ROCArea).\n\n",SWAPPEDPAIRS);
+ printf(" %2d Kullback-Leibler divergence.\n",KLD); //KLD
+ printf(" %2d Normalized Kullback-Leibler divergence.\n",NKLD); //KLD
+ printf(" %2d MILLI.\n",MILLI); //MILLI
+ printf(" %2d GM: geometric mean of tpr and tnr\n",GM); // AIC-QBETA
+ printf(" %2d NAE: normalized absolute error (Esuli & Sebastiani)\n",NAE); // AIC-QBETA
+ printf(" %2d AE: absolute error (Esuli & Sebastiani)\n",AE); // AIC-QBETA
+ printf(" %2d RAE: relative absolute error (Esuli & Sebastiani)\n",RAE); // AIC-QBETA
+ printf(" %2d Qbeta: 100 minus the Qbeta-score in percent (with recall)\n",QBETA); // AIC-QBETA
+ printf(" %2d Qbeta_acc: 100 minus the Qbeta-score in percent (with acc)\n",QBETA_ACC); // AIC-QBETA
+ printf(" %2d Qbeta_f1: 100 minus the Qbeta-score in percent (with F1)\n",QBETA_F1); // AIC-QBETA
+ printf(" %2d Qbeta_gm: 100 minus the Qbeta-score in percent (with GM)\n",QBETA_GM); // AIC-QBETA
+
printf("NOTE: The '-c' parameters in SVM-light and SVM-perf are related as\n");
printf(" c_light = c_perf*100/n for the 'Errorrate' loss function, where n is the\n");
printf(" number of training examples.\n\n");
@@ -1785,7 +1953,65 @@
free(predset);
return(100.0*(apr/(double)(nump)));
}
-
+/* AIC-QBETA */
+double tnr(int a, int b, int c, int d)
+{
+ /* Returns tnr as fractional value. */
+ if((b+d) == 0) return(0.0);
+ return((double)d/(double)(b+d));
+}
+double gm(int a, int b, int c, int d)
+{
+ double tprate = rec(a,b,c,d);
+ double tnrate = tnr(a,b,c,d);
+ return sqrt( tprate * tnrate );
+}
+double nae(int a, int b, int c, int d)
+{
+ double maximo = (a+c > b+d? a+c: b+d);
+ return 1.0 - ( (double)abs(c-b) / maximo);
+ //return 1.0 - ( (double)abs(c-b) / (double)(a+b+c+d)); // ABSERR
+}
+double ae(int a, int b, int c, int d)
+{
+ return (double)abs(c-b) / (double)(a+b+c+d) ; // ABSERR
+}
+double rae(int a, int b, int c, int d)
+{
+ double absdif = (double)abs(c-b) ;
+ double smooth_rae_pos = absdif / ((double)(a+c+0.5)) ;
+ double smooth_rae_neg = absdif / ((double)(b+d+0.5)) ;
+ return 0.5*(smooth_rae_pos + smooth_rae_neg) ;
+}
+double Qbeta(int a, int b, int c, int d, double beta)
+{
+ if(a+c == 0) return(0.0);
+ double qperf=nae(a,b,c,d);
+ double cperf=rec(a,b,c,d);
+ return((1+beta*beta)*qperf*cperf/((beta*beta*cperf)+qperf));
+}
+double Qbeta_acc(int a, int b, int c, int d, double beta)
+{
+ if(a+c == 0) return(0.0);
+ double qperf=nae(a,b,c,d);
+ double cperf=1.0-errorrate(a,b,c,d);
+ return((1+beta*beta)*qperf*cperf/((beta*beta*cperf)+qperf));
+}
+double Qbeta_f1(int a, int b, int c, int d, double beta)
+{
+ if(a+c == 0) return(0.0);
+ double qperf=nae(a,b,c,d);
+ double cperf=fone(a,b,c,d);
+ return((1+beta*beta)*qperf*cperf/((beta*beta*cperf)+qperf));
+}
+double Qbeta_gm(int a, int b, int c, int d, double beta)
+{
+ if(a+c == 0) return(0.0);
+ double qperf=nae(a,b,c,d);
+ double cperf=gm(a,b,c,d);
+ return((1+beta*beta)*qperf*cperf/((beta*beta*cperf)+qperf));
+}
+/* AIC-QBETA */
/*------- Loss functions based on performance measures --------*/
double zeroone_loss(int a, int b, int c, int d)
@@ -1846,4 +2072,70 @@
}
+//KLD
+double kldiv(int a, int b, int c, int d)
+{
+ double sum = a+b+c+d+1.0;
+ double pab = (a+b+0.5)/sum;
+ double pac = (a+c+0.5)/sum;
+ double pbd = (b+d+0.5)/sum;
+ double pcd = (c+d+0.5)/sum;
+
+ double kl = pac*log(pac/pab)+pbd*log(pbd/pcd);
+
+ return kl;
+}
+
+//KLD
+double kldiv_loss(int a, int b, int c, int d)
+{
+ return kldiv(a,b,c,d);
+}
+
+//KLD
+double nkldiv_loss(int a, int b, int c, int d)
+{
+ return 100.0-(100.0*2.0/(1.0+exp(kldiv(a,b,c,d))));
+}
+
+//MILLI
+double milli_loss(int a, int b, int c, int d)
+{
+ int sum = a+b+c+d;
+ return 100.0*(b+c)*ABS(b-c);
+}
+/* AIC-QBETA */
+double gm_loss(int a, int b, int c, int d)
+{
+ return 100.0 * (1.0-gm(a,b,c,d));
+}
+double nae_loss(int a, int b, int c, int d)
+{
+ return 100.0 * (1.0-nae(a,b,c,d));
+}
+double ae_loss(int a, int b, int c, int d)
+{
+ return 100.0 * ae(a,b,c,d);
+}
+double rae_loss(int a, int b, int c, int d)
+{
+ return 100.0 * rae(a,b,c,d);
+}
+double Qbeta_loss(int a, int b, int c, int d,double beta)
+{
+ return(100.0*(1.0-Qbeta(a,b,c,d,beta)));
+}
+double Qbeta_acc_loss(int a, int b, int c, int d,double beta)
+{
+ return(100.0*(1.0-Qbeta_acc(a,b,c,d,beta)));
+}
+double Qbeta_f1_loss(int a, int b, int c, int d,double beta)
+{
+ return(100.0*(1.0-Qbeta_f1(a,b,c,d,beta)));
+}
+double Qbeta_gm_loss(int a, int b, int c, int d,double beta)
+{
+ return(100.0*(1.0-Qbeta_gm(a,b,c,d,beta)));
+}
+/* AIC-QBETA */
diff -ruN svm_perf/svm_struct_api_types.h svm_perf_quantification/svm_struct_api_types.h
--- svm_perf/svm_struct_api_types.h 2020-10-28 12:23:19.000000000 +0100
+++ svm_perf_quantification/svm_struct_api_types.h 2020-10-28 12:23:53.000000000 +0100
@@ -28,14 +28,25 @@
# define INST_VERSION_DATE "15.07.2009"
/* Identifiers for loss functions */
-#define ZEROONE 0
-#define FONE 1
-#define ERRORRATE 2
-#define PRBEP 3
-#define PREC_K 4
-#define REC_K 5
-#define SWAPPEDPAIRS 10
-#define AVGPREC 11
+#define ZEROONE 0
+#define FONE 1
+#define ERRORRATE 2
+#define PRBEP 3
+#define PREC_K 4
+#define REC_K 5
+#define SWAPPEDPAIRS 10
+#define AVGPREC 11
+#define KLD 12 //KLD
+#define NKLD 13 //KLD
+#define MILLI 16 //MILLI
+#define GM 20 // AIC-QBETA
+#define NAE 21 // AIC-QBETA
+#define QBETA 22 // AIC-QBETA
+#define QBETA_ACC 23 // AIC-QBETA
+#define QBETA_F1 24 // AIC-QBETA
+#define QBETA_GM 25 // AIC-QBETA
+#define AE 26 // AIC-QBETA
+#define RAE 27 // AIC-QBETA
/* default precision for solving the optimization problem */
# define DEFAULT_EPS 0.1
@@ -169,6 +180,8 @@
svm_perf_classify. This uses more
memory, but is faster if the support
vectors in the model are dense. */
+ double loss_parm;
+ double beta; /* AIC-QBETA */
} STRUCT_LEARN_PARM;
typedef struct struct_test_stats {
@@ -183,6 +196,16 @@
double prbep;
double rocarea;
double avgprec;
+ double kld; //KLD
+ double nkld; //KLD
+ double gm; // AIC-QBETA
+ double nae; // AIC-QBETA
+ double ae; // AIC-QBETA
+ double rae; // AIC-QBETA
+ double Qbeta; // AIC-QBETA
+ double Qbeta_acc; // AIC-QBETA
+ double Qbeta_f1; // AIC-QBETA
+ double Qbeta_gm; // AIC-QBETA
} STRUCT_TEST_STATS;
typedef struct struct_id_score {
View Git Blame Copy Permalink