int main(int argc, char **argv){
fflush(stdout);
fprintf(stdout, "\nProgram that generates training, evaluation and test sets for the OPF classifier\n");
fprintf(stdout, "\nIf you have any problem, please contact: ");
fprintf(stdout, "\n- [email protected]");
fprintf(stdout, "\n- [email protected]\n");
fprintf(stdout, "\nLibOPF version 2.0 (2009)\n");
fprintf(stdout, "\n"); fflush(stdout);
if(argc != 6){
fprintf(stderr, "\nusage opf_split <P1> <P2> <P3> <P4> <P5>");
fprintf(stderr, "\nP1: input dataset in the OPF file format");
fprintf(stderr, "\nP2: percentage for the training set size [0,1]");
fprintf(stderr, "\nP3: percentage for the evaluation set size [0,1] (leave 0 in the case of no learning)");
fprintf(stderr, "\nP4: percentage for the test set size [0,1]");
fprintf(stderr, "\nP5: normalize features? 1 - Yes 0 - No\n\n");
exit(-1);
}
Subgraph *g = NULL, *gAux = NULL, *gTraining = NULL, *gEvaluating = NULL, *gTesting = NULL;
float training_p = atof(argv[2]), evaluating_p = atof(argv[3]), testing_p = atof(argv[4]);
int normalize = atoi(argv[5]);
CheckInputData(training_p, evaluating_p, testing_p);
fprintf(stdout, "\nReading data set ..."); fflush(stdout);
g = ReadSubgraph(argv[1]);
fprintf(stdout, " OK"); fflush(stdout);
if(normalize) opf_NormalizeFeatures(g);
fprintf(stdout, "\nSplitting data set ..."); fflush(stdout);
opf_SplitSubgraph(g, &gAux, &gTesting, training_p+evaluating_p);
if (evaluating_p > 0)
opf_SplitSubgraph(gAux, &gTraining, &gEvaluating, training_p/(training_p+evaluating_p));
else gTraining = CopySubgraph(gAux);
fprintf(stdout, " OK"); fflush(stdout);
fprintf(stdout, "\nWriting data sets to disk ..."); fflush(stdout);
WriteSubgraph(gTraining, "training.dat");
WriteSubgraphPrototype(gTraining, "trainingp.dat");
if (evaluating_p > 0) {
WriteSubgraph(gEvaluating, "evaluating.dat");
WriteSubgraphPrototype(gEvaluating, "evaluatingp.dat");
}
WriteSubgraph(gTesting, "testing.dat");
WriteSubgraphPrototype(gTesting, "testingp.dat");
fprintf(stdout, " OK"); fflush(stdout);
fprintf(stdout, "\nDeallocating memory ...");
DestroySubgraph(&g);
DestroySubgraph(&gAux);
DestroySubgraph(&gTraining);
DestroySubgraph(&gEvaluating);
DestroySubgraph(&gTesting);
fprintf(stdout, " OK\n");
return 0;
}
int main(int argc, char **argv)
{
if (argc != 5)
{
fprintf(stderr, "\nusage FeatureSelection <training set> <evaluating set> <testing set> <search space configuration file>\n");
exit(-1);
}
SearchSpace *s = NULL;
int i;
double time_opt, time_classify, classification_error;
timer tic, toc;
FILE *f = NULL;
TransferFunc optTransfer = NULL;
Subgraph *Train = NULL, *Evaluate = NULL, *Merge = NULL, *Test = NULL, *newTrain = NULL, *newTest = NULL;
Train = ReadSubgraph(argv[1]);
Evaluate = ReadSubgraph(argv[2]);
Test = ReadSubgraph(argv[3]);
s = ReadSearchSpaceFromFile(argv[4], _WCA_);
optTransfer = S2TransferFunction;
for (i = 0; i < Train->nfeats; i++)
{
s->LB[i] = -20;
s->LB[i] = 20;
}
fprintf(stderr, "\nInitializing search space ... ");
InitializeSearchSpace(s, _WCA_);
fprintf(stderr, "\nOk\n");
fflush(stderr);
fprintf(stderr, "\nRunning WCA ... ");
gettimeofday(&tic, NULL);
runWCA(s, FeatureSelectionOPF, Train, Evaluate, optTransfer);
gettimeofday(&toc, NULL);
fflush(stderr);
fprintf(stderr, "\nOK\n");
time_opt = ((toc.tv_sec - tic.tv_sec) * 1000.0 + (toc.tv_usec - tic.tv_usec) * 0.001) / 1000.0;
fprintf(stdout, "\nOptimization time: %f seconds\n", time_opt);
fflush(stderr);
Merge = opf_MergeSubgraph(Train, Evaluate);
fflush(stderr);
fprintf(stderr, "\nWriting new training and testing sets ...\n");
newTrain = CreateSubgraphFromSelectedFeatures(Merge, s->g);
newTest = CreateSubgraphFromSelectedFeatures(Test, s->g);
fprintf(stderr, "\nTraining set\n");
WriteSubgraph(newTrain, "training.wca.dat");
fprintf(stderr, "\n\nTesting set\n");
WriteSubgraph(newTest, "testing.wca.dat");
fflush(stderr);
fprintf(stderr, "\nOK\n");
opf_OPFTraining(newTrain);
gettimeofday(&tic, NULL);
opf_OPFClassifying(newTrain, newTest);
gettimeofday(&toc, NULL);
classification_error = opf_Accuracy(newTest);
time_classify = ((toc.tv_sec - tic.tv_sec) * 1000.0 + (toc.tv_usec - tic.tv_usec) * 0.001) / 1000.0;
fprintf(stdout, "\nClassification time: %f seconds\n", time_classify);
fflush(stderr);
f = fopen("best_feats.txt", "a");
fprintf(f, "%d %d", newTrain->nfeats, (int)s->g[0]);
for (i = 1; i < Train->nfeats; i++)
{
fprintf(f, " %d", (int)s->g[i]);
}
fprintf(f, "\n");
fclose(f);
fprintf(stderr, "\nAccuracy: %.2lf%%\n", 100 * classification_error);
f = fopen("final_accuracy.txt", "a");
fprintf(f, "%lf\n", classification_error);
fclose(f);
fflush(stderr);
fprintf(stderr, "\nDeallocating memory ...");
DestroySubgraph(&Train);
DestroySubgraph(&Evaluate);
DestroySubgraph(&Merge);
DestroySubgraph(&Test);
DestroySubgraph(&newTrain);
DestroySubgraph(&newTest);
fflush(stderr);
fprintf(stderr, "\nOK\n");
f = fopen("optimization.time", "a");
fprintf(f, "%f %f\n", time_opt, time_classify);
fclose(f);
DestroySearchSpace(&s, _WCA_);
return 0;
}
