public IList <FeatureValue> GetValues(RawWordEncounter encounter, FeatureSelectionContext context)
{
if (context.DataSetName == DataSetName.Train ||
context.DataSetName == DataSetName.Validation)
{
var embedding = context.Project.MeaningEmbeddings.GetVectorOrDefault(encounter.Meaning);
return(FeatureValue.NewArray(embedding));
}
var meanings = context.ReorderedDictionary
.GetByName(encounter.Word)?
.Meanings.Values
.Where(x => x.PartOfSpeech == encounter.Pos)
.ToArray();
if (meanings == null || meanings.Length == 0)
{
return(FeatureValue.NewArray(0, context.Project.MeaningEmbeddings.VectorLength));
}
var averageEmbedding = new float[context.Project.MeaningEmbeddings.VectorLength];
var divisor = 0;
for (var m = 0; m < meanings.Length; m++)
{
var embedding = context.Project.MeaningEmbeddings.GetVectorOrDefault(meanings[m].Meaning);
for (var i = 0; i < averageEmbedding.Length; i++)
{
averageEmbedding[i] += embedding[i] * (meanings[m].Encounters + 1);
divisor += meanings[m].Encounters + 1;
}
}
for (var i = 0; i < averageEmbedding.Length; i++)
{
averageEmbedding[i] /= divisor;
}
return(FeatureValue.NewArray(averageEmbedding));
}
public IList <FeatureValue> GetValues(RawWordEncounter encounter, FeatureSelectionContext context)
{
if (context.DataSetName == DataSetName.Train ||
context.DataSetName == DataSetName.Validation)
{
var embedding = context.Project.MeaningEmbeddings.GetVectorOrDefault(encounter.Meaning);
return(FeatureValue.NewArray(embedding));
}
var mostFrequentMeaning = context.ReorderedDictionary
.GetByName(encounter.Word)?
.Meanings.Values
.SingleOrDefault(x => x.Id == 1 && x.PartOfSpeech == encounter.Pos)?
.Meaning;
var mostFrequentEmbedding = context.Project.MeaningEmbeddings
.GetVectorOrDefault(mostFrequentMeaning);
return(FeatureValue.NewArray(mostFrequentEmbedding));
}
public IList <FeatureValue> GetValues(RawWordEncounter encounter, FeatureSelectionContext context)
{
var posVector = context.FilteredPosList.GetVector(encounter.Pos);
return(FeatureValue.NewArray(posVector));
}
public IList <FeatureValue> Aggregate(
IList <EncounterValues> values, RawRecord record,
FeatureGroup featureGroup, FeatureSelectionContext context)
{
var wordAnalyses = new WordAnalysis[values.Count];
var hasNullWordAnalysis = false;
for (var i = 0; i < values.Count; i++)
{
if (values[i].Encounter == RawWordEncounter.EmptyWordEncounter)
{
return(FeatureValue.NewArray(0, values[i].Values.Count));
}
wordAnalyses[i] = context.Project.DataAnalysis.GetByName(values[i].Encounter.Word);
if (wordAnalyses[i] == null)
{
hasNullWordAnalysis = true;
}
}
var weights = new double[wordAnalyses.Length];
if (hasNullWordAnalysis)
{
var weight = 1d / wordAnalyses.Length;
for (var i = 0; i < weights.Length; i++)
{
weights[i] = weight;
}
}
else
{
var trainEncounters = new double[wordAnalyses.Length];
var trainEncountersSum = 0;
for (var i = 0; i < weights.Length; i++)
{
var currentTrainingEncounters = wordAnalyses[i].TrainEncounters.Values
.Sum(x => x.Encounters);
trainEncounters[i] = currentTrainingEncounters;
trainEncountersSum += currentTrainingEncounters;
}
if (trainEncountersSum == 0)
{
return(FeatureValue.NewArray(0, values[0].Values.Count));
}
for (var i = 0; i < weights.Length; i++)
{
weights[i] = trainEncounters[i] / trainEncountersSum;
}
}
var result = new float[values[0].Values.Count];
for (var vectorIndex = 0; vectorIndex < result.Length; vectorIndex++)
{
for (var encounterIndex = 0; encounterIndex < values.Count; encounterIndex++)
{
result[vectorIndex] += (float)(weights[encounterIndex] *
values[encounterIndex].Values[vectorIndex].NumericValue);
}
}
return(FeatureValue.NewArray(result));
}
public IList <FeatureValue> GetValues(RawWordEncounter encounter, FeatureSelectionContext context)
{
var embedding = context.Project.MeaningEmbeddings.GetVectorOrDefault(encounter.Meaning);
return(FeatureValue.NewArray(embedding));
}