public override sealed void Load(BinarySerializer reader)
{
BinWidth = reader.ReadDouble();
TagDistrTable = new TagDistrTable <SentimentLabel>(reader);
mBinModel = reader.ReadObject <IModel <SentimentLabel, SparseVector <double> > >();
IsTrained = reader.ReadBool();
}
protected virtual Prediction <LblT> GetPrediction(double[] scores, TagDistrTable <LblT> tagDistrTable)
{
IEnumerable <KeyDat <double, LblT> > labelProbs = mTagDistrTable.GetDistrValues(scores)
.OrderByDescending(kv => Math.Abs(kv.Value.GetValueOrDefault()))
.Select(kv => new KeyDat <double, LblT>(kv.Value.GetValueOrDefault(), kv.Key));
return(new Prediction <LblT>(labelProbs));
}
public void Load(BinarySerializer reader)
{
mTagDistrTable = new TagDistrTable <LblT>(reader);
mInnerModels = new IModel <LblT, ExT> [reader.ReadInt()];
for (int i = 0; i < mInnerModels.Length; i++)
{
mInnerModels[i] = reader.ReadObject <IModel <LblT, ExT> >();
}
IsTrained = reader.ReadBool();
}
protected BinVotingClassifier(IModel <LblT, ExT>[] innerModels, double binWidth = 0.05, params LblT[] excludedTags)
{
Preconditions.CheckArgument(typeof(LblT).IsEnum);
Preconditions.CheckNotNull(innerModels);
Preconditions.CheckArgument(innerModels.Length > 0);
Preconditions.CheckArgumentRange(binWidth > 0);
mInnerModels = innerModels;
mTagDistrTable = new EnumTagDistrTable <LblT>(innerModels.Length, binWidth, -5, 5, excludedTags)
{
CalcDistrFunc = (tagCounts, values, tag) => CalcLabelScore(tagCounts, values, tag)
};
}
private void UpdateDistrTable()
{
if (mExampleScores == null || mBinWidth == 0)
{
mTagDistrTable = null;
return;
}
mTagDistrTable = new EnumTagDistrTable <SentimentLabel>(2, mBinWidth, -5, 5, SentimentLabel.Exclude)
{
CalcDistrFunc = (tagCounts, values, tag) => ((double)tagCounts[tag] + 1) / (tagCounts.Values.Sum() + tagCounts.Count)
};
foreach (ExampleScore es in mExampleScores)
{
mTagDistrTable.AddCount(es.Label, es.PosScore, es.NegScore);
}
mTagDistrTable.Calculate();
}
public override void Train(ILabeledExampleCollection <SentimentLabel, SparseVector <double> > dataset)
{
Preconditions.CheckNotNull(dataset);
Preconditions.CheckArgumentRange(TagDistrTable == null || TagDistrTable.NumOfDimensions == 2);
mBinModel = CreateModel();
mBinModel.Train(new LabeledDataset <SentimentLabel, SparseVector <double> >(dataset.Where(le => le.Label != SentimentLabel.Neutral)));
TagDistrTable = new EnumTagDistrTable <SentimentLabel>(1, BinWidth, -5, 5, SentimentLabel.Exclude)
{
CalcDistrFunc = (tagCounts, values, tag) => ((double)tagCounts[tag] + 1) / (tagCounts.Values.Sum() + tagCounts.Count) // use Laplace formula
};
foreach (LabeledExample <SentimentLabel, SparseVector <double> > le in dataset)
{
Prediction <SentimentLabel> prediction = mBinModel.Predict(le.Example);
TagDistrTable.AddCount(le.Label, prediction.BestClassLabel == SentimentLabel.Positive ? prediction.BestScore : -prediction.BestScore);
}
TagDistrTable.Calculate();
IsTrained = true;
}
