Chris@439: /* -*- c-basic-offset: 4 indent-tabs-mode: nil -*-  vi:set ts=8 sts=4 sw=4: */
Chris@439: 
Chris@439: /*
Chris@439:     Sonic Visualiser
Chris@439:     An audio file viewer and annotation editor.
Chris@439:     Centre for Digital Music, Queen Mary, University of London.
Chris@439:     This file copyright 2008 QMUL.
Chris@439:    
Chris@439:     This program is free software; you can redistribute it and/or
Chris@439:     modify it under the terms of the GNU General Public License as
Chris@439:     published by the Free Software Foundation; either version 2 of the
Chris@439:     License, or (at your option) any later version.  See the file
Chris@439:     COPYING included with this distribution for more information.
Chris@439: */
Chris@439: 
Chris@439: #include "RDFImporter.h"
Chris@439: 
Chris@439: #include <map>
Chris@439: #include <vector>
Chris@439: 
Chris@439: #include <iostream>
Chris@439: #include <cmath>
Chris@439: 
Chris@439: #include "SimpleSPARQLQuery.h"
Chris@439: 
Chris@439: #include "base/ProgressReporter.h"
Chris@439: #include "base/RealTime.h"
Chris@439: 
Chris@439: #include "data/model/SparseOneDimensionalModel.h"
Chris@439: #include "data/model/SparseTimeValueModel.h"
Chris@439: #include "data/model/EditableDenseThreeDimensionalModel.h"
Chris@439: 
Chris@439: using std::cerr;
Chris@439: using std::endl;
Chris@439: 
Chris@439: class RDFImporterImpl
Chris@439: {
Chris@439: public:
Chris@439:     RDFImporterImpl(QString url, int sampleRate);
Chris@439:     virtual ~RDFImporterImpl();
Chris@439:     
Chris@439:     bool isOK();
Chris@439:     QString getErrorString() const;
Chris@439: 
Chris@439:     std::vector<Model *> getDataModels(ProgressReporter *);
Chris@439: 
Chris@439: protected:
Chris@439:     QString m_uristring;
Chris@439:     QString m_errorString;
Chris@439:     int m_sampleRate;
Chris@439: 
Chris@439:     typedef std::vector<float> ValueList;
Chris@439:     typedef std::map<RealTime, ValueList> TimeValueMap;
Chris@439:     typedef std::map<QString, TimeValueMap> TypeTimeValueMap;
Chris@439:     typedef std::map<QString, TypeTimeValueMap> SourceTypeTimeValueMap;
Chris@439: 
Chris@439:     void extractStructure(const TimeValueMap &map, bool &sparse,
Chris@439:                           int &minValueCount, int &maxValueCount);
Chris@439: 
Chris@439:     void fillModel(SparseOneDimensionalModel *, const TimeValueMap &);
Chris@439:     void fillModel(SparseTimeValueModel *, const TimeValueMap &);
Chris@439:     void fillModel(EditableDenseThreeDimensionalModel *, const TimeValueMap &);
Chris@439: };
Chris@439: 
Chris@439: 
Chris@439: QString
Chris@439: RDFImporter::getKnownExtensions()
Chris@439: {
Chris@439:     return "*.rdf *.n3 *.ttl";
Chris@439: }
Chris@439: 
Chris@439: RDFImporter::RDFImporter(QString url, int sampleRate) :
Chris@439:     m_d(new RDFImporterImpl(url, sampleRate)) 
Chris@439: {
Chris@439: }
Chris@439: 
Chris@439: RDFImporter::~RDFImporter()
Chris@439: {
Chris@439:     delete m_d;
Chris@439: }
Chris@439: 
Chris@439: bool
Chris@439: RDFImporter::isOK()
Chris@439: {
Chris@439:     return m_d->isOK();
Chris@439: }
Chris@439: 
Chris@439: QString
Chris@439: RDFImporter::getErrorString() const
Chris@439: {
Chris@439:     return m_d->getErrorString();
Chris@439: }
Chris@439: 
Chris@439: std::vector<Model *>
Chris@439: RDFImporter::getDataModels(ProgressReporter *r)
Chris@439: {
Chris@439:     return m_d->getDataModels(r);
Chris@439: }
Chris@439: 
Chris@439: RDFImporterImpl::RDFImporterImpl(QString uri, int sampleRate) :
Chris@439:     m_uristring(uri),
Chris@439:     m_sampleRate(sampleRate)
Chris@439: {
Chris@439: }
Chris@439: 
Chris@439: RDFImporterImpl::~RDFImporterImpl()
Chris@439: {
Chris@439: }
Chris@439: 
Chris@439: bool
Chris@439: RDFImporterImpl::isOK()
Chris@439: {
Chris@439:     return (m_errorString == "");
Chris@439: }
Chris@439: 
Chris@439: QString
Chris@439: RDFImporterImpl::getErrorString() const
Chris@439: {
Chris@439:     return m_errorString;
Chris@439: }
Chris@439: 
Chris@439: std::vector<Model *>
Chris@439: RDFImporterImpl::getDataModels(ProgressReporter *reporter)
Chris@439: {
Chris@439:     std::vector<Model *> models;
Chris@439: 
Chris@439:     // Our query is intended to retrieve every thing that has a time,
Chris@439:     // and every feature type and value associated with a thing that
Chris@439:     // has a time.
Chris@439: 
Chris@439:     // We will then need to refine this big bag of results into a set
Chris@439:     // of data models.
Chris@439: 
Chris@439:     // Results that have different source signals should go into
Chris@439:     // different models.
Chris@439: 
Chris@439:     // Results that have different feature types should go into
Chris@439:     // different models.
Chris@439: 
Chris@439:     // Results that are sparse should go into different models from
Chris@439:     // those that are dense (we need to examine the timestamps to
Chris@439:     // establish this -- if the timestamps are regular, the results
Chris@439:     // are dense -- so we can't do it as we go along, only after
Chris@439:     // collecting all results).
Chris@439: 
Chris@439:     // Timed things that have features associated with them should not
Chris@439:     // appear directly in any model -- their features should appear
Chris@439:     // instead -- and these should be different models from those used
Chris@439:     // for timed things that do not have features.
Chris@439: 
Chris@439:     // As we load the results, we'll push them into a partially
Chris@439:     // structured container that maps from source signal (URI as
Chris@439:     // string) -> feature type (likewise) -> time -> list of values.
Chris@439:     // If the source signal or feature type is unavailable, the empty
Chris@439:     // string will do.
Chris@439: 
Chris@439:     SourceTypeTimeValueMap m;
Chris@439: 
Chris@439:     QString queryString = QString(
Chris@439: 
Chris@439:         " PREFIX event: <http://purl.org/NET/c4dm/event.owl#>"
Chris@439:         " PREFIX time: <http://purl.org/NET/c4dm/timeline.owl#>"
Chris@439:         " PREFIX mo: <http://purl.org/ontology/mo/>"
Chris@439:         " PREFIX af: <http://purl.org/ontology/af/>"
Chris@439: 
Chris@439:         " SELECT ?signalSource ?time ?eventType ?value"
Chris@439:         " FROM <%1>"
Chris@439: 
Chris@439:         " WHERE {"
Chris@439:         "   ?signal mo:available_as ?signalSource ."
Chris@439:         "   ?signal mo:time ?interval ."
Chris@439:         "   ?interval time:onTimeLine ?tl ."
Chris@439:         "   ?t time:onTimeLine ?tl ."
Chris@439:         "   ?t time:at ?time ."
Chris@439:         "   ?timedThing event:time ?t ."
Chris@439:         "   ?timedThing a ?eventType ."
Chris@439:         "   OPTIONAL {"
Chris@439:         "     ?timedThing af:hasFeature ?feature ."
Chris@439:         "     ?feature af:value ?value"
Chris@439:         "   }"
Chris@439:         " }"
Chris@439: 
Chris@439:         ).arg(m_uristring);
Chris@439: 
Chris@439:     SimpleSPARQLQuery query(queryString);
Chris@439:     query.setProgressReporter(reporter);
Chris@439: 
Chris@439:     cerr << "Query will be: " << queryString.toStdString() << endl;
Chris@439: 
Chris@439:     SimpleSPARQLQuery::ResultList results = query.execute();
Chris@439: 
Chris@439:     if (!query.isOK()) {
Chris@439:         m_errorString = query.getErrorString();
Chris@439:         return models;
Chris@439:     }
Chris@439: 
Chris@439:     if (query.wasCancelled()) {
Chris@439:         m_errorString = "Query cancelled";
Chris@439:         return models;
Chris@439:     }        
Chris@439: 
Chris@439:     for (int i = 0; i < results.size(); ++i) {
Chris@439: 
Chris@439:         QString source = results[i]["signalSource"].value;
Chris@439: 
Chris@439:         QString timestring = results[i]["time"].value;
Chris@439:         RealTime time;
Chris@439:         time = RealTime::fromXsdDuration(timestring.toStdString());
Chris@439:         cerr << "time = " << time.toString() << " (from xsd:duration \""
Chris@439:              << timestring.toStdString() << "\")" << endl;
Chris@439: 
Chris@439:         QString type = results[i]["eventType"].value;
Chris@439: 
Chris@439:         QString valuestring = results[i]["value"].value;
Chris@439:         float value = 0.f;
Chris@439:         bool haveValue = false;
Chris@439:         if (valuestring != "") {
Chris@439:             value = valuestring.toFloat(&haveValue);
Chris@439:             cerr << "value = " << value << endl;
Chris@439:         }
Chris@439: 
Chris@439:         if (haveValue) {
Chris@439:             m[source][type][time].push_back(value);
Chris@439:         } else if (m[source][type].find(time) == m[source][type].end()) {
Chris@439:             m[source][type][time] = ValueList();
Chris@439:         }
Chris@439:     }
Chris@439: 
Chris@439:     for (SourceTypeTimeValueMap::const_iterator mi = m.begin();
Chris@439:          mi != m.end(); ++mi) {
Chris@439:         
Chris@439:         QString source = mi->first;
Chris@439: 
Chris@439:         for (TypeTimeValueMap::const_iterator ttvi = mi->second.begin();
Chris@439:              ttvi != mi->second.end(); ++ttvi) {
Chris@439:             
Chris@439:             QString type = ttvi->first;
Chris@439: 
Chris@439:             // Now we need to work out what sort of model to use for
Chris@439:             // this source/type combination.  Ultimately we'll
Chris@439:             // hopefully be able to map directly from the type to the
Chris@439:             // model on the basis of known structures for the types,
Chris@439:             // but we also want to be able to handle untyped data
Chris@439:             // according to its apparent structure so let's do that
Chris@439:             // first.
Chris@439: 
Chris@439:             bool sparse = false;
Chris@439:             int minValueCount = 0, maxValueCount = 0;
Chris@439: 
Chris@439:             extractStructure(ttvi->second, sparse, minValueCount, maxValueCount);
Chris@439:     
Chris@439:             cerr << "For source \"" << source.toStdString() << "\", type \""
Chris@439:                  << type.toStdString() << "\" we have sparse = " << sparse
Chris@439:                  << ", min value count = " << minValueCount << ", max = "
Chris@439:                  << maxValueCount << endl;
Chris@439: 
Chris@439:             // Model allocations:
Chris@439:             //
Chris@439:             // Sparse, no values: SparseOneDimensionalModel
Chris@439:             //
Chris@439:             // Sparse, always 1 value: SparseTimeValueModel
Chris@439:             //
Chris@439:             // Sparse, > 1 value: No standard model for this.  If
Chris@439:             // there are always 2 values, perhaps hack it into
Chris@439:             // NoteModel for now?  Or always use SparseTimeValueModel
Chris@439:             // and discard all but the first value.
Chris@439:             //
Chris@439:             // Dense, no values: Meaningless; no suitable model
Chris@439:             //
Chris@439:             // Dense, > 0 values: EditableDenseThreeDimensionalModel
Chris@439:             //
Chris@439:             // These should just be our fallback positions; we want to
Chris@439:             // be reading semantic data from the RDF in order to pick
Chris@439:             // the right model directly
Chris@439: 
Chris@439:             enum { SODM, STVM, EDTDM } modelType = SODM;
Chris@439: 
Chris@439:             if (sparse) {
Chris@439:                 if (maxValueCount == 0) {
Chris@439:                     modelType = SODM;
Chris@439:                 } else if (minValueCount == 1 && maxValueCount == 1) {
Chris@439:                     modelType = STVM;
Chris@439:                 } else {
Chris@439:                     cerr << "WARNING: No suitable model available for sparse data with between " << minValueCount << " and " << maxValueCount << " values" << endl;
Chris@439:                     modelType = STVM;
Chris@439:                 }
Chris@439:             } else {
Chris@439:                 if (maxValueCount == 0) {
Chris@439:                     cerr << "WARNING: Dense data set with no values is not meaningful, skipping" << endl;
Chris@439:                     continue;
Chris@439:                 } else {
Chris@439:                     modelType = EDTDM;
Chris@439:                 }
Chris@439:             }
Chris@439: 
Chris@439:             //!!! set model name &c
Chris@439: 
Chris@439:             if (modelType == SODM) {
Chris@439: 
Chris@439:                 SparseOneDimensionalModel *model = 
Chris@439:                     new SparseOneDimensionalModel(m_sampleRate, 1, false);
Chris@439:                 
Chris@439:                 fillModel(model, ttvi->second);
Chris@439:                 models.push_back(model);
Chris@439: 
Chris@439:             } else if (modelType == STVM) {
Chris@439: 
Chris@439:                 SparseTimeValueModel *model = 
Chris@439:                     new SparseTimeValueModel(m_sampleRate, 1, false);
Chris@439:                 
Chris@439:                 fillModel(model, ttvi->second);
Chris@439:                 models.push_back(model);
Chris@439: 
Chris@439:             } else {
Chris@439:                 
Chris@439:                 EditableDenseThreeDimensionalModel *model =
Chris@439:                     new EditableDenseThreeDimensionalModel(m_sampleRate, 1, 0,
Chris@439:                                                            false);
Chris@439: 
Chris@439:                 fillModel(model, ttvi->second);
Chris@439:                 models.push_back(model);
Chris@439:             }
Chris@439:         }
Chris@439:     }
Chris@439: 
Chris@439: 
Chris@439:     return models;
Chris@439: }
Chris@439: 
Chris@439: void
Chris@439: RDFImporterImpl::extractStructure(const TimeValueMap &tvm,
Chris@439:                                   bool &sparse,
Chris@439:                                   int &minValueCount,
Chris@439:                                   int &maxValueCount)
Chris@439: {
Chris@439:     // These are floats intentionally rather than RealTime --
Chris@439:     // see logic for handling rounding error below
Chris@439:     float firstTime = 0.f;
Chris@439:     float timeStep = 0.f;
Chris@439:     bool haveTimeStep = false;
Chris@439:     
Chris@439:     for (TimeValueMap::const_iterator tvi = tvm.begin(); tvi != tvm.end(); ++tvi) {
Chris@439:         
Chris@439:         RealTime time = tvi->first;
Chris@439:         int valueCount = tvi->second.size();
Chris@439:         
Chris@439:         if (tvi == tvm.begin()) {
Chris@439:             
Chris@439:             minValueCount = valueCount;
Chris@439:             maxValueCount = valueCount;
Chris@439:             
Chris@439:             firstTime = time.toDouble();
Chris@439:             
Chris@439:         } else {
Chris@439:             
Chris@439:             if (valueCount < minValueCount) minValueCount = valueCount;
Chris@439:             if (valueCount > maxValueCount) maxValueCount = valueCount;
Chris@439:             
Chris@439:             if (!haveTimeStep) {
Chris@439:                 timeStep = time.toDouble() - firstTime;
Chris@439:                 if (timeStep == 0.f) sparse = true;
Chris@439:                 haveTimeStep = true;
Chris@439:             } else if (!sparse) {
Chris@439:                 // test whether this time is within
Chris@439:                 // rounding-error range of being an integer
Chris@439:                 // multiple of some constant away from the
Chris@439:                 // first time
Chris@439:                 float timeAsFloat = time.toDouble();
Chris@439:                 int count = int((timeAsFloat - firstTime) / timeStep + 0.5);
Chris@439:                 float expected = firstTime + (timeStep * count);
Chris@439:                 if (fabsf(expected - timeAsFloat) > 1e-6) {
Chris@439:                     cerr << "Event at " << timeAsFloat << " is not evenly spaced -- would expect it to be " << expected << " for a spacing of " << count << " * " << timeStep << endl;
Chris@439:                     sparse = true;
Chris@439:                 }
Chris@439:             }
Chris@439:         }
Chris@439:     }
Chris@439: }
Chris@439: 
Chris@439: void
Chris@439: RDFImporterImpl::fillModel(SparseOneDimensionalModel *model,
Chris@439:                            const TimeValueMap &tvm)
Chris@439: {
Chris@439:     //!!! labels &c not yet handled
Chris@439: 
Chris@439:     for (TimeValueMap::const_iterator tvi = tvm.begin();
Chris@439:          tvi != tvm.end(); ++tvi) {
Chris@439:         
Chris@439:         RealTime time = tvi->first;
Chris@439:         long frame = RealTime::realTime2Frame(time, m_sampleRate);
Chris@439: 
Chris@439:         SparseOneDimensionalModel::Point point(frame);
Chris@439: 
Chris@439:         model->addPoint(point);
Chris@439:     }
Chris@439: }
Chris@439: 
Chris@439: void
Chris@439: RDFImporterImpl::fillModel(SparseTimeValueModel *model,
Chris@439:                            const TimeValueMap &tvm)
Chris@439: {
Chris@439:     //!!! labels &c not yet handled
Chris@439: 
Chris@439:     for (TimeValueMap::const_iterator tvi = tvm.begin();
Chris@439:          tvi != tvm.end(); ++tvi) {
Chris@439:         
Chris@439:         RealTime time = tvi->first;
Chris@439:         long frame = RealTime::realTime2Frame(time, m_sampleRate);
Chris@439: 
Chris@439:         float value = 0.f;
Chris@439:         if (!tvi->second.empty()) value = *tvi->second.begin();
Chris@439:         
Chris@439:         SparseTimeValueModel::Point point(frame, value, "");
Chris@439: 
Chris@439:         model->addPoint(point);
Chris@439:     }
Chris@439: }
Chris@439: 
Chris@439: void
Chris@439: RDFImporterImpl::fillModel(EditableDenseThreeDimensionalModel *model,
Chris@439:                            const TimeValueMap &tvm)
Chris@439: {
Chris@439:     //!!! labels &c not yet handled
Chris@439: 
Chris@439:     //!!! start time offset not yet handled
Chris@439: 
Chris@439:     size_t col = 0;
Chris@439: 
Chris@439:     for (TimeValueMap::const_iterator tvi = tvm.begin();
Chris@439:          tvi != tvm.end(); ++tvi) {
Chris@439:         
Chris@439:         model->setColumn(col++, tvi->second);
Chris@439:     }
Chris@439: }
Chris@439: