private String getMaxHiddenVariable(String frame, int[] intTokNums, String[][] data) {
THashSet<String> hiddenUnits = mFrameMap.get(frame);
String maxHV = null;
double maxVal = -Double.MAX_VALUE;
DependencyParse parse = DependencyParse.processFN(data, 0.0);
for (String unit : hiddenUnits) {
FeatureExtractor featex = new FeatureExtractor();
IntCounter<String> valMap =
featex.extractFeaturesLessMemory(
frame,
intTokNums,
unit,
data,
"test",
mRelatedWordsForWord,
mRevisedRelationsMap,
mHVLemmas,
parse);
Set<String> features = valMap.keySet();
double featSum = 0.0;
for (String feat : features) {
double val = valMap.getT(feat);
int ind = localA.get(feat);
double paramVal = V[ind].exponentiate();
double prod = val * paramVal;
featSum += prod;
}
double expVal = Math.exp(featSum);
if (expVal > maxVal) {
maxVal = expVal;
maxHV = unit;
}
}
return maxHV;
}
public Combine() {
this.context = FeatureExtractorFactory.getInstance(FeatureExtractorType.CONTEXT);
this.single = FeatureExtractorFactory.getInstance(FeatureExtractorType.SINGLE_CONCEPT);
this.vicinity = FeatureExtractorFactory.getInstance(FeatureExtractorType.CONCEPT_VICINITY);
this.dimensions = context.getDimension() + single.getDimension() + vicinity.getDimension();
}
public static ArrayList<Interview> processInterviews(
ArrayList<Interview> interviews, FeatureExtractor extractor) {
ArrayList<Interview> new_interviews = new ArrayList<>();
for (Iterator<Interview> i = interviews.iterator(); i.hasNext(); )
new_interviews.add(extractor.update(i.next()));
return new_interviews;
}
public List<String> predictArgumentLines(Sentence sentence, List<String> idResult, int kBest)
throws IOException {
final List<FrameFeatures> frameFeaturesList = Lists.newArrayList();
final FeatureExtractor featureExtractor = new FeatureExtractor();
for (String feLine : idResult) {
final DataPointWithFrameElements dataPoint = new DataPointWithFrameElements(sentence, feLine);
final String frame = dataPoint.getFrameName();
final DependencyParses parses = dataPoint.getParses();
final int targetStartTokenIdx = dataPoint.getTargetTokenIdxs()[0];
final int targetEndTokenIdx =
dataPoint.getTargetTokenIdxs()[dataPoint.getTargetTokenIdxs().length - 1];
final List<SpanAndParseIdx> spans = DataPrep.findSpans(dataPoint, 1);
final List<String> frameElements =
Lists.newArrayList(frameElementsForFrame.lookupFrameElements(frame));
final List<SpanAndCorrespondingFeatures[]> featuresAndSpanByArgument = Lists.newArrayList();
for (String frameElement : frameElements) {
final List<SpanAndCorrespondingFeatures> spansAndFeatures = Lists.newArrayList();
for (SpanAndParseIdx candidateSpanAndParseIdx : spans) {
final Range0Based span = candidateSpanAndParseIdx.span;
final DependencyParse parse = parses.get(candidateSpanAndParseIdx.parseIdx);
final Set<String> featureSet =
featureExtractor
.extractFeatures(dataPoint, frame, frameElement, span, parse)
.elementSet();
final int[] featArray = convertToIdxs(featureSet);
spansAndFeatures.add(
new SpanAndCorrespondingFeatures(new int[] {span.start, span.end}, featArray));
}
featuresAndSpanByArgument.add(
spansAndFeatures.toArray(new SpanAndCorrespondingFeatures[spansAndFeatures.size()]));
}
frameFeaturesList.add(
new FrameFeatures(
frame,
targetStartTokenIdx,
targetEndTokenIdx,
frameElements,
featuresAndSpanByArgument));
}
return decoder.decodeAll(frameFeaturesList, idResult, 0, kBest);
}
private double getValueForFrame(String frame, int[] intTokNums, String[][] data) {
THashSet<String> hiddenUnits = mFrameMap.get(frame);
double result = 0.0;
DependencyParse parse = DependencyParse.processFN(data, 0.0);
for (String unit : hiddenUnits) {
IntCounter<String> valMap = null;
FeatureExtractor featex = new FeatureExtractor();
if (clusterMap == null) {
valMap =
featex.extractFeaturesLessMemory(
frame,
intTokNums,
unit,
data,
"test",
mRelatedWordsForWord,
mRevisedRelationsMap,
mHVLemmas,
parse);
} else { // not supported
valMap =
featex.extractFeaturesWithClusters(
frame, intTokNums, unit, data, mWNR, "test", null, null, parse, clusterMap, K);
}
Set<String> features = valMap.keySet();
double featSum = 0.0;
for (String feat : features) {
double val = valMap.getT(feat);
int ind = localA.get(feat);
double paramVal = V[ind].exponentiate();
double prod = val * paramVal;
featSum += prod;
}
double expVal = Math.exp(featSum);
result += expVal;
}
return result;
}
public static double[] feature(AudioSamples audio, AudioFeature i) throws Exception {
/**
* 1. Spectral Centroid 2. Spectral Rolloff Point 3. Spectral Flux 4. Compactness 5. Spectral
* Variability 6. Root Mean Square 7. Fration of Low Energy Windows 8. Zero Crossings 9.
* Strongest Beat 10. Beat Sum 11. MFCC 12. ConstantQ 13. LPC 14. Method of Moments 15. Peak
* Detection 16. Area Method of MFCCs
*/
FeatureExtractor featureExt;
double[] samples = audio.getSamplesMixedDown();
double sampleRate = audio.getSamplingRateAsDouble();
double[][] otherFeatures = audio.getSamplesChannelSegregated();
double[][] windowSample;
switch (i.getValue()) {
case 1:
featureExt = new SpectralCentroid();
return featureExt.extractFeature(samples, sampleRate, otherFeatures);
case 2:
featureExt = new PowerSpectrum();
otherFeatures[0] = featureExt.extractFeature(samples, sampleRate, otherFeatures);
featureExt = new SpectralRolloffPoint();
return featureExt.extractFeature(samples, sampleRate, otherFeatures);
case 3:
windowSample = audio.getSampleWindowsMixedDown(2);
featureExt = new MagnitudeSpectrum();
otherFeatures[0] = featureExt.extractFeature(windowSample[0], sampleRate, otherFeatures);
otherFeatures[1] = featureExt.extractFeature(windowSample[1], sampleRate, otherFeatures);
featureExt = new SpectralFlux();
return featureExt.extractFeature(samples, sampleRate, otherFeatures);
case 4:
featureExt = new MagnitudeSpectrum();
otherFeatures[0] = featureExt.extractFeature(samples, sampleRate, otherFeatures);
featureExt = new Compactness();
return featureExt.extractFeature(samples, sampleRate, otherFeatures);
case 5:
featureExt = new MagnitudeSpectrum();
otherFeatures[0] = featureExt.extractFeature(samples, sampleRate, otherFeatures);
featureExt = new SpectralVariability();
return featureExt.extractFeature(samples, sampleRate, otherFeatures);
case 6:
featureExt = new RMS();
return featureExt.extractFeature(samples, sampleRate, otherFeatures);
case 7:
featureExt = new RMS();
windowSample = audio.getSampleWindowsMixedDown(100);
for (int j = 0; j < 100; j++) {
otherFeatures[j] = featureExt.extractFeature(windowSample[j], sampleRate, null);
}
featureExt = new FractionOfLowEnergyWindows();
return featureExt.extractFeature(samples, sampleRate, otherFeatures);
case 8:
featureExt = new ZeroCrossings();
return featureExt.extractFeature(samples, sampleRate, otherFeatures);
case 9:
double[][] otherFeature = new double[3][];
featureExt = new RMS();
/*String [] dependecies3 = featureExt.getDepenedencies();
if(dependecies3!=null)
for(int k=0;k<dependecies3.length;k++)
System.out.println(dependecies3[k]);*/
otherFeature[0] = featureExt.extractFeature(samples, sampleRate, null);
featureExt = new BeatHistogram();
/*String [] dependecies1 = featureExt.getDepenedencies();
for(int k=0;k<dependecies1.length;k++)
System.out.println(dependecies1[k]);*/
double[][] RmsFeat = new double[1][];
RmsFeat[0] = otherFeature[0];
otherFeature[1] = featureExt.extractFeature(samples, sampleRate, RmsFeat);
featureExt = new BeatHistogramLabels();
/*String [] dependecies2 = featureExt.getDepenedencies();
for(int k=0;k<de pendecies2.length;k++)
System.out.println(dependecies2[k]);*/
double[][] beatHistogramFeat = new double[1][];
beatHistogramFeat[0] = otherFeature[1];
otherFeature[2] = featureExt.extractFeature(samples, sampleRate, beatHistogramFeat);
featureExt = new StrongestBeat();
double[][] StrongestBeatFeat = new double[2][];
StrongestBeatFeat[0] = otherFeature[1];
StrongestBeatFeat[1] = otherFeature[2];
/*String[] dependencies = featureExt.getDepenedencies();
for(int k=0;k<dependencies.length;k++)
System.out.println(dependencies[k]);*/
return featureExt.extractFeature(samples, sampleRate, StrongestBeatFeat);
/* featureExt = new BeatHistogram();
double[][] otherFeature = new double[2][];
otherFeature[0] = featureExt.extractFeature(samples, sampleRate, otherFeatures);
featureExt = new BeatHistogramLabels();
otherFeature[1] = featureExt.extractFeature(samples, sampleRate, otherFeatures);
featureExt = new StrongestBeat();
return featureExt.extractFeature(samples, sampleRate, otherFeature); */
case 10:
featureExt = new BeatHistogram();
otherFeatures[0] = featureExt.extractFeature(samples, sampleRate, otherFeatures);
featureExt = new BeatSum();
return featureExt.extractFeature(samples, sampleRate, otherFeatures);
case 11:
featureExt = new MagnitudeSpectrum();
otherFeatures[0] = featureExt.extractFeature(samples, sampleRate, otherFeatures);
featureExt = new MFCC();
return featureExt.extractFeature(samples, sampleRate, otherFeatures);
case 12:
featureExt = new ConstantQ();
return featureExt.extractFeature(samples, sampleRate, otherFeatures);
case 13:
featureExt = new LPC();
return featureExt.extractFeature(samples, sampleRate, otherFeatures);
case 14:
featureExt = new MagnitudeSpectrum();
otherFeatures[0] = featureExt.extractFeature(samples, sampleRate, otherFeatures);
featureExt = new Moments();
return featureExt.extractFeature(samples, sampleRate, otherFeatures);
case 15:
featureExt = new MagnitudeSpectrum();
otherFeatures[0] = featureExt.extractFeature(samples, sampleRate, otherFeatures);
featureExt = new PeakFinder();
return featureExt.extractFeature(samples, sampleRate, otherFeatures);
case 16:
featureExt = new MagnitudeSpectrum();
windowSample = audio.getSampleWindowsMixedDown(100);
for (int j = 0; j < 100; j++) {
otherFeatures[j] = featureExt.extractFeature(windowSample[j], sampleRate, null);
}
featureExt = new AreaMoments();
return featureExt.extractFeature(samples, sampleRate, otherFeatures);
default:
return null;
}
}
