Mercurial > hg > svcore
view data/fileio/CSVFileReader.cpp @ 875:3e6ed8a8577b tonioni
Use a sparse time-value model only for outputs with fixed bin count of 1, not for those with unknown bin count.
(Precursor to using more than one model for outputs with unknown bin count)
author | Chris Cannam |
---|---|
date | Tue, 28 Jan 2014 18:52:22 +0000 |
parents | f5cd33909744 |
children | 69cc0454ed72 |
line wrap: on
line source
/* -*- c-basic-offset: 4 indent-tabs-mode: nil -*- vi:set ts=8 sts=4 sw=4: */ /* Sonic Visualiser An audio file viewer and annotation editor. Centre for Digital Music, Queen Mary, University of London. This file copyright 2006 Chris Cannam. This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version. See the file COPYING included with this distribution for more information. */ #include "CSVFileReader.h" #include "model/Model.h" #include "base/RealTime.h" #include "base/StringBits.h" #include "model/SparseOneDimensionalModel.h" #include "model/SparseTimeValueModel.h" #include "model/EditableDenseThreeDimensionalModel.h" #include "model/RegionModel.h" #include "DataFileReaderFactory.h" #include <QFile> #include <QString> #include <QRegExp> #include <QStringList> #include <QTextStream> #include <iostream> #include <map> CSVFileReader::CSVFileReader(QString path, CSVFormat format, size_t mainModelSampleRate) : m_format(format), m_file(0), m_warnings(0), m_mainModelSampleRate(mainModelSampleRate) { m_file = new QFile(path); bool good = false; if (!m_file->exists()) { m_error = QFile::tr("File \"%1\" does not exist").arg(path); } else if (!m_file->open(QIODevice::ReadOnly | QIODevice::Text)) { m_error = QFile::tr("Failed to open file \"%1\"").arg(path); } else { good = true; } if (!good) { delete m_file; m_file = 0; } } CSVFileReader::~CSVFileReader() { SVDEBUG << "CSVFileReader::~CSVFileReader: file is " << m_file << endl; if (m_file) { SVDEBUG << "CSVFileReader::CSVFileReader: Closing file" << endl; m_file->close(); } delete m_file; } bool CSVFileReader::isOK() const { return (m_file != 0); } QString CSVFileReader::getError() const { return m_error; } size_t CSVFileReader::convertTimeValue(QString s, int lineno, size_t sampleRate, size_t windowSize) const { QRegExp nonNumericRx("[^0-9eE.,+-]"); unsigned int warnLimit = 10; CSVFormat::TimeUnits timeUnits = m_format.getTimeUnits(); size_t calculatedFrame = 0; bool ok = false; QString numeric = s; numeric.remove(nonNumericRx); if (timeUnits == CSVFormat::TimeSeconds) { double time = numeric.toDouble(&ok); if (!ok) time = StringBits::stringToDoubleLocaleFree(numeric, &ok); calculatedFrame = int(time * sampleRate + 0.5); } else { long n = numeric.toLong(&ok); if (n >= 0) calculatedFrame = n; if (timeUnits == CSVFormat::TimeWindows) { calculatedFrame *= windowSize; } } if (!ok) { if (m_warnings < warnLimit) { cerr << "WARNING: CSVFileReader::load: " << "Bad time format (\"" << s << "\") in data line " << lineno+1 << endl; } else if (m_warnings == warnLimit) { cerr << "WARNING: Too many warnings" << endl; } ++m_warnings; } return calculatedFrame; } Model * CSVFileReader::load() const { if (!m_file) return 0; CSVFormat::ModelType modelType = m_format.getModelType(); CSVFormat::TimingType timingType = m_format.getTimingType(); CSVFormat::TimeUnits timeUnits = m_format.getTimeUnits(); size_t sampleRate = m_format.getSampleRate(); size_t windowSize = m_format.getWindowSize(); QChar separator = m_format.getSeparator(); bool allowQuoting = m_format.getAllowQuoting(); if (timingType == CSVFormat::ExplicitTiming) { if (modelType == CSVFormat::ThreeDimensionalModel) { // This will be overridden later if more than one line // appears in our file, but we want to choose a default // that's likely to be visible windowSize = 1024; } else { windowSize = 1; } if (timeUnits == CSVFormat::TimeSeconds) { sampleRate = m_mainModelSampleRate; } } SparseOneDimensionalModel *model1 = 0; SparseTimeValueModel *model2 = 0; RegionModel *model2a = 0; EditableDenseThreeDimensionalModel *model3 = 0; Model *model = 0; QTextStream in(m_file); in.seek(0); unsigned int warnings = 0, warnLimit = 10; unsigned int lineno = 0; float min = 0.0, max = 0.0; size_t frameNo = 0; size_t duration = 0; size_t endFrame = 0; bool haveAnyValue = false; bool haveEndTime = false; size_t startFrame = 0; // for calculation of dense model resolution bool firstEverValue = true; std::map<QString, int> labelCountMap; int valueColumns = 0; for (int i = 0; i < m_format.getColumnCount(); ++i) { if (m_format.getColumnPurpose(i) == CSVFormat::ColumnValue) { ++valueColumns; } } while (!in.atEnd()) { // QTextStream's readLine doesn't cope with old-style Mac // CR-only line endings. Why did they bother making the class // cope with more than one sort of line ending, if it still // can't be configured to cope with all the common sorts? // For the time being we'll deal with this case (which is // relatively uncommon for us, but still necessary to handle) // by reading the entire file using a single readLine, and // splitting it. For CR and CR/LF line endings this will just // read a line at a time, and that's obviously OK. QString chunk = in.readLine(); QStringList lines = chunk.split('\r', QString::SkipEmptyParts); for (size_t li = 0; li < lines.size(); ++li) { QString line = lines[li]; if (line.startsWith("#")) continue; QStringList list = StringBits::split(line, separator, allowQuoting); if (!model) { switch (modelType) { case CSVFormat::OneDimensionalModel: model1 = new SparseOneDimensionalModel(sampleRate, windowSize); model = model1; break; case CSVFormat::TwoDimensionalModel: model2 = new SparseTimeValueModel(sampleRate, windowSize, false); model = model2; break; case CSVFormat::TwoDimensionalModelWithDuration: model2a = new RegionModel(sampleRate, windowSize, false); model = model2a; break; case CSVFormat::ThreeDimensionalModel: model3 = new EditableDenseThreeDimensionalModel (sampleRate, windowSize, valueColumns, EditableDenseThreeDimensionalModel::NoCompression); model = model3; break; } } float value = 0.f; QString label = ""; duration = 0.f; haveEndTime = false; for (int i = 0; i < list.size(); ++i) { QString s = list[i]; CSVFormat::ColumnPurpose purpose = m_format.getColumnPurpose(i); switch (purpose) { case CSVFormat::ColumnUnknown: break; case CSVFormat::ColumnStartTime: frameNo = convertTimeValue(s, lineno, sampleRate, windowSize); break; case CSVFormat::ColumnEndTime: endFrame = convertTimeValue(s, lineno, sampleRate, windowSize); haveEndTime = true; break; case CSVFormat::ColumnDuration: duration = convertTimeValue(s, lineno, sampleRate, windowSize); break; case CSVFormat::ColumnValue: value = s.toFloat(); haveAnyValue = true; break; case CSVFormat::ColumnLabel: label = s; ++labelCountMap[label]; break; } } if (haveEndTime) { // ... calculate duration now all cols read if (endFrame > frameNo) { duration = endFrame - frameNo; } } if (modelType == CSVFormat::OneDimensionalModel) { SparseOneDimensionalModel::Point point(frameNo, label); model1->addPoint(point); } else if (modelType == CSVFormat::TwoDimensionalModel) { SparseTimeValueModel::Point point(frameNo, value, label); model2->addPoint(point); } else if (modelType == CSVFormat::TwoDimensionalModelWithDuration) { RegionModel::Point point(frameNo, value, duration, label); model2a->addPoint(point); } else if (modelType == CSVFormat::ThreeDimensionalModel) { DenseThreeDimensionalModel::Column values; for (int i = 0; i < list.size(); ++i) { if (m_format.getColumnPurpose(i) != CSVFormat::ColumnValue) { continue; } bool ok = false; float value = list[i].toFloat(&ok); values.push_back(value); if (firstEverValue || value < min) min = value; if (firstEverValue || value > max) max = value; if (firstEverValue) { startFrame = frameNo; model3->setStartFrame(startFrame); } else if (lineno == 1 && timingType == CSVFormat::ExplicitTiming) { model3->setResolution(frameNo - startFrame); } firstEverValue = false; if (!ok) { if (warnings < warnLimit) { cerr << "WARNING: CSVFileReader::load: " << "Non-numeric value \"" << list[i] << "\" in data line " << lineno+1 << ":" << endl; cerr << line << endl; ++warnings; } else if (warnings == warnLimit) { // cerr << "WARNING: Too many warnings" << endl; } } } // SVDEBUG << "Setting bin values for count " << lineno << ", frame " // << frameNo << ", time " << RealTime::frame2RealTime(frameNo, sampleRate) << endl; model3->setColumn(lineno, values); } ++lineno; if (timingType == CSVFormat::ImplicitTiming || list.size() == 0) { frameNo += windowSize; } } } if (!haveAnyValue) { if (model2a) { // assign values for regions based on label frequency; we // have this in our labelCountMap, sort of std::map<int, std::map<QString, float> > countLabelValueMap; for (std::map<QString, int>::iterator i = labelCountMap.begin(); i != labelCountMap.end(); ++i) { countLabelValueMap[i->second][i->first] = 0.f; } float v = 0.f; for (std::map<int, std::map<QString, float> >::iterator i = countLabelValueMap.end(); i != countLabelValueMap.begin(); ) { --i; for (std::map<QString, float>::iterator j = i->second.begin(); j != i->second.end(); ++j) { j->second = v; v = v + 1.f; } } std::map<RegionModel::Point, RegionModel::Point, RegionModel::Point::Comparator> pointMap; for (RegionModel::PointList::const_iterator i = model2a->getPoints().begin(); i != model2a->getPoints().end(); ++i) { RegionModel::Point p(*i); v = countLabelValueMap[labelCountMap[p.label]][p.label]; RegionModel::Point pp(p.frame, v, p.duration, p.label); pointMap[p] = pp; } for (std::map<RegionModel::Point, RegionModel::Point>::iterator i = pointMap.begin(); i != pointMap.end(); ++i) { model2a->deletePoint(i->first); model2a->addPoint(i->second); } } } if (modelType == CSVFormat::ThreeDimensionalModel) { model3->setMinimumLevel(min); model3->setMaximumLevel(max); } return model; }