void testPredictor(Predictor p){
//Load sample structure
SampleStructure sampleStructure = getTestSampleStructure();
//Load test data
Array<Sample> testData = getTestSampleData();
unsigned errors = 0;
for(unsigned s = 0; s < testData.length; s++){
Sample predictedSample = p.predict(testData[s]);
std::cout << "SAMPLE " << ((double)s) << ": ";
for(unsigned f = 0; f < sampleStructure.contvarCount; f++){
std::cout << testData[s].contvars[f] << ", ";
}
for(unsigned f = 0; f < sampleStructure.catvarCount; f++){
std::cout << ((double)testData[s].catvars[f]) << ", ";
}
std::cout << "\nPREDICTIONS: ";
for(unsigned f = 0; f < sampleStructure.contvarCount; f++){
contvar_t prediction = predictedSample.contvars[f]; //p.classifiers[f]->predict(testData[s]);
std::cout << ((double)prediction) << ", ";
//if(prediction != testData[s].catvars[f]){
// errors++;
//}
}for(unsigned f = 0; f < sampleStructure.catvarCount; f++){
catvar_t prediction = predictedSample.catvars[f]; //p.classifiers[f]->predict(testData[s]);
std::cout << ((double)prediction) << ", ";
if(prediction != testData[s].catvars[f]){
errors++;
}
}
std::cout << "\n";
}
std::cout << "Catvar Accuracy: " << (1.0 - ((double)errors) / (testData.length * sampleStructure.catvarCount)) << ".\n";
}
void PredictorRegistryTest::testNext()
{
ContextTracker* pointer = static_cast<ContextTracker*>((void*)0xdeadbeef);
registry->setContextTracker(pointer);
PredictorRegistry::Iterator it = registry->iterator();
Predictor* predictor = 0;
while (it.hasNext()) {
predictor = it.next();
}
// since we've iterated till the end of the predictors list, predictor
// is now pointing to the DummyPredictor, so let's test we got the
// dummy prediction back
Prediction prediction = predictor->predict(20, 0);
CPPUNIT_ASSERT(predictor != 0);
size_t expected_size = 18;
CPPUNIT_ASSERT_EQUAL(expected_size, prediction.size());
CPPUNIT_ASSERT_EQUAL(Suggestion("foo1", 0.99), prediction.getSuggestion(0));
CPPUNIT_ASSERT_EQUAL(Suggestion("foobar6", 0.74), prediction.getSuggestion(17));
}
void testFrac(Predictor p){
std::cout << "Testing FRaC.\n";
SampleStructure st = getTestSampleStructure();
Array<Sample> evaluationData = getEvaluationSampleData();
ErrorModelCreators emc;
emc.rc = [](SampleStructure* st, unsigned index, contvar_t* trueData, contvar_t* predictedData, unsigned length){
return (ContinuousErrorModel*) new Gaussian(trueData, predictedData, length);
};
emc.cc = [](SampleStructure* st, unsigned index, catvar_t* trueData, catvar_t* predictedData, unsigned length){
return (CategoricalErrorModel*) new SurprisalMatrix(st->catvarSizes[index], trueData, predictedData, length);
};
emc.bcc = [](SampleStructure* st, unsigned index, binvar_t* trueData, binvar_t* predictedData, unsigned length){
Array<catvar_t> trueConv = Array<binvar_t>(trueData, length).map<catvar_t>([](binvar_t bv){return (catvar_t)bv;});
Array<catvar_t> predConv = Array<binvar_t>(predictedData, length).map<catvar_t>([](binvar_t bv){return (catvar_t)bv;});
SurprisalMatrix catmodel = SurprisalMatrix(2, trueConv.data, predConv.data, length);
trueConv.freeMemory();
predConv.freeMemory();
return (BinaryErrorModel*) new BinaryErrorModelFromCategoricalErrorModel<SurprisalMatrix>(SurprisalMatrix());
};
//Note: Here we build error models on training data: we should be using new validation data.
ErrorModelCollection e = ErrorModelCollection(st, emc, evaluationData, p);
/*
gaussian* rem = new gaussian[st.contvarCount];
NormalizedSurprisalMatrix* cem = new NormalizedSurprisalMatrix[st.catvarCount];
for(unsigned i = 0; i < st.catvarCount; i++){
unsigned length = trainingData.length;
catvar_t* trueData = new catvar_t[length];
catvar_t* predictedData = new catvar_t[length];
for(unsigned j = 0; j < length; j++){
trueData[j] = trainingData[j].catvars[i];
predictedData[j] = p.classifiers[i]->predict(trainingData[j]);
}
cem[i] = NormalizedSurprisalMatrix(st.catvarSizes[i], trueData, predictedData, length);
}
char* bcem = new char[st.binvarCount];
ErrorModelCollection e = ErrorModelCollection(rem, cem, bcem);
*/
TraditionalFRaC frac(st, p, e);
Array<Sample> testData = getTestSampleData();
for(unsigned i = 0; i < testData.length; i++){
double ns = frac.calculateNS(testData[i], p.predict(testData[i]));
std::cout << "Normal Sample " << i << ": NS = " << ns << "\n";
std::cout << StructuredSample(st, testData[i]) << "\n";
}
Array<Sample> anomTestData = getAnomalousSampleData();
for(unsigned i = 0; i < anomTestData.length; i++){
double ns = frac.calculateNS(anomTestData[i], p.predict(testData[i]));
std::cout << "Anomalous Sample " << i << ": NS = " << ns << "\n";
std::cout << StructuredSample(st, anomTestData[i]) << "\n";
}
}
/**
* program main entry
*
* @param argc number of program arguments
* @param argv array of program arguments of length argc
*/
int main(int argc, char **argv){
try {
// set overall logging style
el::Loggers::reconfigureAllLoggers(el::ConfigurationType::Format, std::string("# %level : %msg"));
// TODO setup log file
el::Loggers::reconfigureAllLoggers(el::ConfigurationType::ToFile, std::string("false"));
el::Loggers::reconfigureAllLoggers(el::ConfigurationType::ToStandardOutput, std::string("true"));
// set additional logging flags
el::Loggers::addFlag(el::LoggingFlag::DisableApplicationAbortOnFatalLog);
el::Loggers::addFlag(el::LoggingFlag::LogDetailedCrashReason);
el::Loggers::addFlag(el::LoggingFlag::AllowVerboseIfModuleNotSpecified);
#if INTARNA_LOG_COLORING
el::Loggers::addFlag(el::LoggingFlag::ColoredTerminalOutput);
#endif
// setup logging with given parameters
START_EASYLOGGINGPP(argc, argv);
// check if log file set and update all loggers before going on
if (el::Helpers::commandLineArgs() != NULL && el::Helpers::commandLineArgs()->hasParamWithValue(el::base::consts::kDefaultLogFileParam))
{
// default all to file
el::Loggers::reconfigureAllLoggers(el::ConfigurationType::ToStandardOutput, std::string("false"));
el::Loggers::reconfigureAllLoggers(el::ConfigurationType::ToFile, std::string("true"));
// enforec error out to standard output
el::Loggers::reconfigureAllLoggers(el::Level::Error, el::ConfigurationType::ToStandardOutput, std::string("true"));
el::Loggers::reconfigureAllLoggers(el::Level::Error, el::ConfigurationType::ToFile, std::string("false"));
}
// parse command line parameters
CommandLineParsing parameters;
{
VLOG(1) <<"parsing arguments"<<"...";
int retCode = parameters.parse( argc, argv );
if (retCode != CommandLineParsing::ReturnCode::KEEP_GOING) {
return retCode;
}
}
#if INTARNA_MULITHREADING
// OMP shared variables to enable exception forwarding from within OMP parallelized for loop
bool threadAborted = false;
std::exception_ptr exceptionPtrDuringOmp = NULL;
std::stringstream exceptionInfoDuringOmp;
#endif
// number of already reported interactions to enable IntaRNA v1 separator output
size_t reportedInteractions = 0;
// storage to avoid accessibility recomputation (init NULL)
std::vector< ReverseAccessibility * > queryAcc(parameters.getQuerySequences().size(), NULL);
// compute all query accessibilities to enable parallelization
#if INTARNA_MULITHREADING
// parallelize this loop if possible; if not -> parallelize the query-loop
# pragma omp parallel for schedule(dynamic) num_threads( parameters.getThreads() ) shared(queryAcc,reportedInteractions,exceptionPtrDuringOmp,exceptionInfoDuringOmp)
#endif
for (size_t qi=0; qi<queryAcc.size(); qi++) {
// get accessibility handler
#if INTARNA_MULITHREADING
#pragma omp flush (threadAborted)
// explicit try-catch-block due to missing OMP exception forwarding
if (!threadAborted) {
try {
// get query accessibility handler
#pragma omp critical(intarna_omp_logOutput)
#endif
VLOG(1) <<"computing accessibility for query '"<<parameters.getQuerySequences().at(qi).getId()<<"'...";
Accessibility * queryAccOrig = parameters.getQueryAccessibility(qi);
INTARNA_CHECK_NOT_NULL(queryAccOrig,"query initialization failed");
// reverse indexing of target sequence for the computation
queryAcc[qi] = new ReverseAccessibility(*queryAccOrig);
// check if we have to warn about ambiguity
if (queryAccOrig->getSequence().isAmbiguous()) {
#if INTARNA_MULITHREADING
#pragma omp critical(intarna_omp_logOutput)
#endif
LOG(INFO) <<"Sequence '"<<queryAccOrig->getSequence().getId()
<<"' contains ambiguous nucleotide encodings. These positions are ignored for interaction computation.";
}
#if INTARNA_MULITHREADING
////////////////////// exception handling ///////////////////////////
} catch (std::exception & e) {
// ensure exception handling for first failed thread only
#pragma omp critical(intarna_omp_exception)
{
if (!threadAborted) {
// store exception information
exceptionPtrDuringOmp = std::make_exception_ptr(e);
exceptionInfoDuringOmp <<" #thread "<<omp_get_thread_num() <<" #query "<<qi <<" : "<<e.what();
// trigger abortion of all threads
threadAborted = true;
#pragma omp flush (threadAborted)
}
} // omp critical(intarna_omp_exception)
} catch (...) {
// ensure exception handling for first failed thread only
#pragma omp critical(intarna_omp_exception)
{
if (!threadAborted) {
// store exception information
exceptionPtrDuringOmp = std::current_exception();
exceptionInfoDuringOmp <<" #thread "<<omp_get_thread_num() <<" #query "<<qi;
// trigger abortion of all threads
threadAborted = true;
#pragma omp flush (threadAborted)
}
} // omp critical(intarna_omp_exception)
}
} // if not threadAborted
#endif
}
// check which loop to parallelize
const bool parallelizeTargetLoop = parameters.getTargetSequences().size() > 1;
const bool parallelizeQueryLoop = !parallelizeTargetLoop && parameters.getQuerySequences().size() > 1;
const bool parallelizeWindowsLoop = !parallelizeTargetLoop && !parallelizeQueryLoop;
// run prediction for all pairs of sequences
// first: iterate over all target sequences
#if INTARNA_MULITHREADING
// parallelize this loop if possible; if not -> parallelize the query-loop
# pragma omp parallel for schedule(dynamic) num_threads( parameters.getThreads() ) shared(queryAcc,reportedInteractions,exceptionPtrDuringOmp,exceptionInfoDuringOmp) if(parallelizeTargetLoop)
#endif
for ( size_t targetNumber = 0; targetNumber < parameters.getTargetSequences().size(); ++targetNumber )
{
#if INTARNA_MULITHREADING
#pragma omp flush (threadAborted)
// explicit try-catch-block due to missing OMP exception forwarding
if (!threadAborted) {
try {
// get target accessibility handler
#pragma omp critical(intarna_omp_logOutput)
#endif
{ VLOG(1) <<"computing accessibility for target '"<<parameters.getTargetSequences().at(targetNumber).getId()<<"'..."; }
// VRNA not completely threadsafe ...
Accessibility * targetAcc = parameters.getTargetAccessibility(targetNumber);
INTARNA_CHECK_NOT_NULL(targetAcc,"target initialization failed");
// check if we have to warn about ambiguity
if (targetAcc->getSequence().isAmbiguous()) {
#if INTARNA_MULITHREADING
#pragma omp critical(intarna_omp_logOutput)
#endif
{ LOG(INFO) <<"Sequence '"<<targetAcc->getSequence().getId()
<<"' contains ambiguous IUPAC nucleotide encodings. These positions are ignored for interaction computation and replaced by 'N'.";}
}
// second: iterate over all query sequences
#if INTARNA_MULITHREADING
// this parallelization should only be enabled if the outer target-loop is not parallelized
# pragma omp parallel for schedule(dynamic) num_threads( parameters.getThreads() ) shared(queryAcc,reportedInteractions,exceptionPtrDuringOmp,exceptionInfoDuringOmp,targetAcc,targetNumber) if(parallelizeQueryLoop)
#endif
for ( size_t queryNumber = 0; queryNumber < parameters.getQuerySequences().size(); ++queryNumber )
{
#if INTARNA_MULITHREADING
#pragma omp flush (threadAborted)
// explicit try-catch-block due to missing OMP exception forwarding
if (!threadAborted) {
try {
#endif
// sanity check
assert( queryAcc.at(queryNumber) != NULL );
// get energy computation handler for both sequences
InteractionEnergy* energy = parameters.getEnergyHandler( *targetAcc, *(queryAcc.at(queryNumber)) );
INTARNA_CHECK_NOT_NULL(energy,"energy initialization failed");
// get output/storage handler
OutputHandler * output = parameters.getOutputHandler( *energy );
INTARNA_CHECK_NOT_NULL(output,"output handler initialization failed");
// check if we have to add separator for IntaRNA v1 output
if (reportedInteractions > 0 && dynamic_cast<OutputHandlerIntaRNA1*>(output) != NULL) {
dynamic_cast<OutputHandlerIntaRNA1*>(output)->addSeparator( true );
}
// setup collecting output handler to ensure
// k-best output per query-target combination
// and not per region combination if not requested
OutputHandlerInteractionList bestInteractions(
(parameters.reportBestPerRegion() ? std::numeric_limits<size_t>::max() : 1 )
* parameters.getOutputConstraint().reportMax );
// run prediction for all range combinations
BOOST_FOREACH(const IndexRange & tRange, parameters.getTargetRanges(*energy, targetNumber)) {
BOOST_FOREACH(const IndexRange & qRange, parameters.getQueryRanges(*energy, queryNumber)) {
// get windows for both ranges
std::vector<IndexRange> queryWindows = qRange.overlappingWindows(parameters.getWindowWidth(), parameters.getWindowOverlap());
std::vector<IndexRange> targetWindows = tRange.overlappingWindows(parameters.getWindowWidth(), parameters.getWindowOverlap());
// iterate over all window combinations
#if INTARNA_MULITHREADING
// this parallelization should only be enabled if neither the outer target-loop nor the inner query-loop are parallelized
# pragma omp parallel for schedule(dynamic) collapse(2) num_threads( parameters.getThreads() ) shared(queryAcc,reportedInteractions,exceptionPtrDuringOmp,exceptionInfoDuringOmp,targetAcc,targetNumber,queryNumber,queryWindows,targetWindows, bestInteractions, energy) if(parallelizeWindowsLoop)
#endif
for (int qNumWindow = 0; qNumWindow < queryWindows.size(); ++qNumWindow) {
for (int tNumWindow = 0; tNumWindow < targetWindows.size(); ++tNumWindow) {
#if INTARNA_MULITHREADING
#pragma omp flush (threadAborted)
// explicit try-catch-block due to missing OMP exception forwarding
if (!threadAborted) {
try {
#endif
IndexRange qWindow = queryWindows.at(qNumWindow);
IndexRange tWindow = targetWindows.at(tNumWindow);
#if INTARNA_MULITHREADING
#pragma omp critical(intarna_omp_logOutput)
#endif
{ VLOG(1) <<"predicting interactions for"
<<" target "<<targetAcc->getSequence().getId()
<<" (range " <<(tWindow+1)<<")"
<<" and"
<<" query "<<queryAcc.at(queryNumber)->getSequence().getId()
<<" (range " <<(qWindow+1)<<")"
#if INTARNA_MULITHREADING
#if INTARNA_IN_DEBUG_MODE
<<" in thread "<<omp_get_thread_num()
#endif
#endif
<<" ..."; }
// get interaction prediction handler
Predictor * predictor = parameters.getPredictor( *energy, bestInteractions );
INTARNA_CHECK_NOT_NULL(predictor,"predictor initialization failed");
// run prediction for this window combination
predictor->predict( tWindow
, queryAcc.at(queryNumber)->getReversedIndexRange(qWindow)
, parameters.getOutputConstraint()
);
// garbage collection
INTARNA_CLEANUP(predictor);
#if INTARNA_MULITHREADING
////////////////////// exception handling ///////////////////////////
} catch (std::exception & e) {
// ensure exception handling for first failed thread only
#pragma omp critical(intarna_omp_exception)
{
if (!threadAborted) {
// store exception information
exceptionPtrDuringOmp = std::make_exception_ptr(e);
exceptionInfoDuringOmp <<" #thread "<<omp_get_thread_num() <<" #target "<<targetNumber <<" #query " <<queryNumber <<" : "<<e.what();
// trigger abortion of all threads
threadAborted = true;
#pragma omp flush (threadAborted)
}
} // omp critical(intarna_omp_exception)
} catch (...) {
// ensure exception handling for first failed thread only
#pragma omp critical(intarna_omp_exception)
{
if (!threadAborted) {
// store exception information
exceptionPtrDuringOmp = std::current_exception();
exceptionInfoDuringOmp <<" #thread "<<omp_get_thread_num() <<" #target "<<targetNumber <<" #query " <<queryNumber;
// trigger abortion of all threads
threadAborted = true;
#pragma omp flush (threadAborted)
}
} // omp critical(intarna_omp_exception)
}
} // if not threadAborted
#endif
}} // window combinations
} // target ranges
} // query ranges
#if INTARNA_MULITHREADING
#pragma omp critical(intarna_omp_outputHandlerUpdate)
#endif
{// update final output handler
BOOST_FOREACH( const Interaction * inter, bestInteractions) {
// forward all reported interactions for all regions to final output handler
output->add(*inter);
}}
#if INTARNA_MULITHREADING
#pragma omp atomic update
#endif
reportedInteractions += output->reported();
// garbage collection
INTARNA_CLEANUP(output);
INTARNA_CLEANUP(energy);
#if INTARNA_MULITHREADING
////////////////////// exception handling ///////////////////////////
} catch (std::exception & e) {
// ensure exception handling for first failed thread only
#pragma omp critical(intarna_omp_exception)
{
if (!threadAborted) {
// store exception information
exceptionPtrDuringOmp = std::make_exception_ptr(e);
exceptionInfoDuringOmp <<" #thread "<<omp_get_thread_num() <<" #target "<<targetNumber <<" #query " <<queryNumber <<" : "<<e.what();
// trigger abortion of all threads
threadAborted = true;
#pragma omp flush (threadAborted)
}
} // omp critical(intarna_omp_exception)
} catch (...) {
// ensure exception handling for first failed thread only
#pragma omp critical(intarna_omp_exception)
{
if (!threadAborted) {
// store exception information
exceptionPtrDuringOmp = std::current_exception();
exceptionInfoDuringOmp <<" #thread "<<omp_get_thread_num() <<" #target "<<targetNumber <<" #query " <<queryNumber;
// trigger abortion of all threads
threadAborted = true;
#pragma omp flush (threadAborted)
}
} // omp critical(intarna_omp_exception)
}
} // if not threadAborted
#endif
} // for queries
// write accessibility to file if needed
parameters.writeTargetAccessibility( *targetAcc );
// garbage collection
INTARNA_CLEANUP(targetAcc);
#if INTARNA_MULITHREADING
////////////////////// exception handling ///////////////////////////
} catch (std::exception & e) {
// ensure exception handling for first failed thread only
#pragma omp critical(intarna_omp_exception)
{
if (!threadAborted) {
// store exception information
exceptionPtrDuringOmp = std::make_exception_ptr(e);
exceptionInfoDuringOmp <<" #thread "<<omp_get_thread_num() <<" #target "<<targetNumber <<" : "<<e.what();
// trigger abortion of all threads
threadAborted = true;
#pragma omp flush (threadAborted)
}
} // omp critical(intarna_omp_exception)
} catch (...) {
// ensure exception handling for first failed thread only
#pragma omp critical(intarna_omp_exception)
{
if (!threadAborted) {
// store exception information
exceptionPtrDuringOmp = std::current_exception();
exceptionInfoDuringOmp <<" #thread "<<omp_get_thread_num() <<" #target "<<targetNumber;
// trigger abortion of all threads
threadAborted = true;
#pragma omp flush (threadAborted)
}
} // omp critical(intarna_omp_exception)
}
} // if not threadAborted
void DejavuPredictorTest::testPredict()
{
*stream << "polly wants a cracker ";
ct->update();
// get pointer to dejavu predictor
Predictor* predictor = predictorRegistry->iterator().next();
{
*stream << "polly ";
Prediction expected;
CPPUNIT_ASSERT_EQUAL(expected, predictor->predict(SIZE, 0));
ct->update();
}
{
*stream << "wants ";
Prediction expected;
CPPUNIT_ASSERT_EQUAL(expected, predictor->predict(SIZE, 0));
ct->update();
}
{
*stream << "a ";
Prediction expected;
expected.addSuggestion(Suggestion("cracker", 1.0));
CPPUNIT_ASSERT_EQUAL(expected, predictor->predict(SIZE, 0));
ct->update();
}
*stream << "soda ";
ct->update();
{
*stream << "polly ";
Prediction expected;
CPPUNIT_ASSERT_EQUAL(expected, predictor->predict(SIZE, 0));
ct->update();
}
{
*stream << "wants ";
Prediction expected;
CPPUNIT_ASSERT_EQUAL(expected, predictor->predict(SIZE, 0));
ct->update();
}
{
*stream << "a ";
Prediction expected;
expected.addSuggestion(Suggestion("cracker", 1.0));
expected.addSuggestion(Suggestion("soda", 1.0));
CPPUNIT_ASSERT_EQUAL(expected, predictor->predict(SIZE, 0));
ct->update();
}
*stream << "cake ";
ct->update();
{
*stream << "polly ";
Prediction expected;
CPPUNIT_ASSERT_EQUAL(expected, predictor->predict(SIZE, 0));
ct->update();
}
{
*stream << "wants ";
Prediction expected;
CPPUNIT_ASSERT_EQUAL(expected, predictor->predict(SIZE, 0));
ct->update();
}
{
*stream << "a ";
Prediction expected;
expected.addSuggestion(Suggestion("cake", 1.0));
expected.addSuggestion(Suggestion("cracker", 1.0));
expected.addSuggestion(Suggestion("soda", 1.0));
CPPUNIT_ASSERT_EQUAL(expected, predictor->predict(SIZE, 0));
ct->update();
}
*stream << "crumble ";
ct->update();
{
// test filter
const char* filter[] = { "cra", "so", 0 };
*stream << "polly wants a ";
Prediction expected;
expected.addSuggestion(Suggestion("cracker", 1.0));
expected.addSuggestion(Suggestion("soda", 1.0));
CPPUNIT_ASSERT_EQUAL(expected, predictor->predict(SIZE, filter));
ct->update();
}
*stream << "break ";
ct->update();
{
// test filter
const char* filter[] = { "r", 0 };
*stream << "polly wants a c";
Prediction expected;
expected.addSuggestion(Suggestion("cracker", 1.0));
expected.addSuggestion(Suggestion("crumble", 1.0));
CPPUNIT_ASSERT_EQUAL(expected, predictor->predict(SIZE, filter));
ct->update();
}
*stream << "uddle ";
ct->update();
}