/**
* Train the SVM using several cross validation iterations
* @param pNorm Normalization object
*/
void CrossValidation::train(Normalizer * pNorm) {
if (VERB > 0) {
cerr << "---Training with Cpos";
if (selectedCpos_ > 0) {
cerr << "=" << selectedCpos_;
} else {
cerr << " selected by cross validation";
}
cerr << ", Cneg";
if (selectedCneg_ > 0) {
cerr << "=" << selectedCneg_;
} else {
cerr << " selected by cross validation";
}
cerr << ", fdr=" << selectionFdr_ << endl;
}
// iterate
int foundPositivesOldOld = 0, foundPositivesOld = 0, foundPositives = 0;
for (unsigned int i = 0; i < niter_; i++) {
if (VERB > 1) {
cerr << "Iteration " << i + 1 << " :\t";
}
bool updateDOC = true;
foundPositives = doStep(updateDOC);
if (VERB > 1) {
cerr << "After the iteration step, " << foundPositives
<< " target PSMs with q<" << testFdr_
<< " were estimated by cross validation" << endl;
}
if (VERB > 2) {
cerr << "Obtained weights" << endl;
printAllWeights(cerr, pNorm);
}
if (foundPositives > 0 && foundPositivesOldOld > 0 && quickValidation_ &&
(static_cast<double>(foundPositives - foundPositivesOldOld) <=
foundPositivesOldOld * requiredIncreaseOver2Iterations_)) {
if (VERB > 1) {
std::cerr << "Performance increase over the last two iterations " <<
"indicate that the algorithm has converged\n" <<
"(" << foundPositives << " vs " << foundPositivesOldOld << ")" <<
std::endl;
}
break;
}
foundPositivesOldOld = foundPositivesOld;
foundPositivesOld = foundPositives;
}
if (VERB == 2) {
cerr
<< "Obtained weights (only showing weights of first cross validation set)"
<< endl;
printSetWeights(cerr, 0, pNorm);
}
foundPositives = 0;
for (size_t set = 0; set < numFolds_; ++set) {
if (DataSet::getCalcDoc()) {
testScores_[set].getDOC().copyDOCparameters(trainScores_[set].getDOC());
testScores_[set].setDOCFeatures();
}
foundPositives += testScores_[set].calcScores(w_[set], testFdr_);
}
if (VERB > 0) {
std::cerr << "After all training done, " << foundPositives <<
" target PSMs with q<" << testFdr_ <<
" were found when measuring on the test set" << std::endl;
}
}
/**
* Train the SVM using several cross validation iterations
* @param w list of normal vectors
*/
void CrossValidation::train(Normalizer * pNorm) {
if (VERB > 0) {
cerr << "---Training with Cpos";
if (selectedCpos > 0) {
cerr << "=" << selectedCpos;
} else {
cerr << " selected by cross validation";
}
cerr << ", Cneg";
if (selectedCneg > 0) {
cerr << "=" << selectedCneg;
} else {
cerr << " selected by cross validation";
}
cerr << ", fdr=" << selectionFdr << endl;
}
// iterate
int foundPositivesOldOld=0, foundPositivesOld=0, foundPositives=0;
for (unsigned int i = 0; i < niter; i++) {
if (VERB > 1) {
cerr << "Iteration " << i + 1 << " :\t";
}
foundPositives = xv_step(true);
if (VERB > 1) {
cerr << "After the iteration step, " << foundPositives
<< " target PSMs with q<" << selectionFdr
<< " were estimated by cross validation" << endl;
}
if (VERB > 2) {
cerr << "Obtained weights" << endl;
printAllWeights(cerr, pNorm);
}
if (foundPositives>0 && foundPositivesOldOld>0 && quickValidation) {
if ((double)(foundPositives-foundPositivesOldOld)<=(foundPositivesOldOld*requiredIncreaseOver2Iterations)) {
if (VERB > 1) {
cerr << "Performance increase over the last two iterations indicate that the algorithm has converged" << endl;
cerr << "(" << foundPositives << " vs " << foundPositivesOldOld << ")" << endl;
}
break;
}
}
foundPositivesOldOld=foundPositivesOld;
foundPositivesOld=foundPositives;
}
if (VERB == 2) {
cerr
<< "Obtained weights (only showing weights of first cross validation set)"
<< endl;
printSetWeights(cerr, 0, pNorm);
}
foundPositives = 0;
for (size_t set = 0; set < xval_fold; ++set) {
if (DataSet::getCalcDoc()) {
xv_test[set].getDOC().copyDOCparameters(xv_train[set].getDOC());
xv_test[set].setDOCFeatures();
}
foundPositives += xv_test[set].calcScores(w[set], testFdr);
}
if (VERB > 0) {
cerr << "After all training done, " << foundPositives << " target PSMs with q<"
<< testFdr << " were found when measuring on the test set"
<< endl;
}
}