double parzenWindowEstimator::getHist(const int i, const int j)
{
if ( posHist(i,j)+negHist(i,j) < 5 )
return 0;
return posHist(i,j)/(posHist(i,j)+negHist(i,j));
}
double parzenWindowEstimator1D::getF_X(const std::vector<double> x)
{
double f_x = 0;
for (size_t i = 0; i < posHist.rows(); i++)
{
if ( posHist(i,0)>=0 )
{
double factor=(posHist(i,0)+negHist(i,0))>0?posHist(i,0)/(posHist(i,0)+negHist(i,0)):0;
f_x += factor * gauss(ext(0,0)+i*binWidth[0],sigm[0],x[0]);
}
}
return f_x;
}
double parzenWindowEstimator2D::getF_X(const std::vector<double> x)
{
double f_x = 0;
for (size_t i = 0; i < posHist.rows(); i++)
{
for (size_t j = 0; j < posHist.cols(); j++)
{
if ( posHist(i,j)>=0 )
{
double factor=(posHist(i,j)+negHist(i,j))>0?posHist(i,j)/(posHist(i,j)+negHist(i,j)):0;
f_x += factor * gauss2D(ext(0,0)+i*binWidth[0],ext(1,0)+j*binWidth[1],sigm[0],sigm[1],x[0],x[1]);
}
}
}
return f_x;
}
bool parzenWindowEstimator::addSample(const std::vector<double> x)
{
std::vector<int> b;
if (getIndexes(x,b))
{
printf("adding sample %i %i from %g %g\n",b[0],b[1],x[0],x[1]);
posHist(b[0],b[1]) += 1;
return true;
}
else
return false;
}
int Application::main(int argc,char *argv[])
{
// Process command line
BOOM::CommandLine cmd(argc,argv,"");
if(cmd.numArgs()!=5)
throw string(
"\ncompute-signal-likelihoods <*.model> <pos-examples.fasta> \n\
<neg-examples.fasta> <consensuses> <#bins>\n\
\n\
example:\n\
compute-signal-likelihoods tag.model tag.fasta contigs.fasta TAG,TGA,TAA\n\
\n\
\n");
BOOM::String modelFilename=cmd.arg(0);
BOOM::String posExamples=cmd.arg(1);
BOOM::String negExamples=cmd.arg(2);
BOOM::String consensusArg=cmd.arg(3);
int numBins=cmd.arg(4).asInt();
if(!filestemRegex.search(modelFilename))
throw BOOM::String("Couldn't parse filestem from filename: ")+
modelFilename;
BOOM::String filestem=filestemRegex[1];
// Load the model
alphabet=DnaAlphabet::global();
GarbageIgnorer GC;
SignalSensor *model=SignalSensor::load(modelFilename,GC);
BOOM::Vector<BOOM::String> *consensuses=consensusArg.getFields(",");
int numConsensuses=consensuses->size();
for(int i=0 ; i<numConsensuses ; ++i)
model->addConsensus((*consensuses)[i]);
// Load the examples
BOOM::Vector<double> *posScores=loadPosExamples(posExamples,*model);
BOOM::Vector<double> *negScores=loadPosExamples(negExamples,*model);
// Report summary stats
BOOM::SummaryStats posStats(*posScores);
BOOM::SummaryStats backgroundStats(*negScores);
cout<<"positives: "<<posStats.getMean()<<"+/-"<<posStats.getStdDev()<<
" ("<<posStats.getMin()<<"-"<<posStats.getMax()<<")"<<endl;
cout<<"background: "<<backgroundStats.getMean()<<"+/-"
<<backgroundStats.getStdDev()<<" ("<<backgroundStats.getMin()
<<"-"<<backgroundStats.getMax()<<")"<<endl;
writeHistogramFile(*posScores,filestem+".pos-hist");
writeHistogramFile(*negScores,filestem+".neg-hist");
// Construct histograms
double minScore=POSITIVE_INFINITY, maxScore=NEGATIVE_INFINITY;
getExtrema(*posScores,minScore,maxScore);
getExtrema(*negScores,minScore,maxScore);
Histogram<double> posHist(minScore,maxScore,numBins,0.01);
Histogram<double> backgroundHist(minScore,maxScore,numBins,1);
posHist.addCounts(*posScores);
backgroundHist.addCounts(*negScores);
backgroundHist.addCounts(*posScores);
// Compute log-likelihood ratios and write output file
computeRatios(filestem,posHist,backgroundHist,minScore,maxScore,
numBins);
/*
posHist.divideBy(backgroundHist);
posHist.useLogs();
BOOM::String outfile=filestem+".isp";
posHist.save(outfile);
*/
return 0;
}