private protected override OneVersusAllModelParameters CreatePredictor()
{
Host.Check(TrainedEnsemble != null, "The predictor cannot be created before training is complete.");
Host.Assert(_numClass > 1, "Must know the number of classes before creating a predictor.");
Host.Assert(TrainedEnsemble.NumTrees % _numClass == 0, "Number of trees should be a multiple of number of classes.");
var innerArgs = LightGbmInterfaceUtils.JoinParameters(GbmOptions);
IPredictorProducing <float>[] predictors = new IPredictorProducing <float> [_tlcNumClass];
for (int i = 0; i < _tlcNumClass; ++i)
{
var pred = CreateBinaryPredictor(i, innerArgs);
var cali = new PlattCalibrator(Host, -0.5, 0);
predictors[i] = new FeatureWeightsCalibratedModelParameters <LightGbmBinaryModelParameters, PlattCalibrator>(Host, pred, cali);
}
string obj = (string)GetGbmParameters()["objective"];
if (obj == "multiclass")
{
return(OneVersusAllModelParameters.Create(Host, OneVersusAllModelParameters.OutputFormula.Softmax, predictors));
}
else
{
return(OneVersusAllModelParameters.Create(Host, predictors));
}
}
public static PredictorModelOutput CombineOvaModels(IHostEnvironment env, CombineOvaPredictorModelsInput input)
{
Contracts.CheckValue(env, nameof(env));
var host = env.Register("CombineOvaModels");
host.CheckValue(input, nameof(input));
EntryPointUtils.CheckInputArgs(host, input);
host.CheckNonEmpty(input.ModelArray, nameof(input.ModelArray));
// Something tells me we should put normalization as part of macro expansion, but since i get
// subgraph instead of learner it's a bit tricky to get learner and decide should we add
// normalization node or not, plus everywhere in code we leave that reposnsibility to TransformModel.
var normalizedView = input.ModelArray[0].TransformModel.Apply(host, input.TrainingData);
using (var ch = host.Start("CombineOvaModels"))
{
var schema = normalizedView.Schema;
var label = TrainUtils.MatchNameOrDefaultOrNull(ch, schema, nameof(input.LabelColumnName),
input.LabelColumnName,
DefaultColumnNames.Label);
var feature = TrainUtils.MatchNameOrDefaultOrNull(ch, schema, nameof(input.FeatureColumnName),
input.FeatureColumnName, DefaultColumnNames.Features);
var weight = TrainUtils.MatchNameOrDefaultOrNull(ch, schema, nameof(input.ExampleWeightColumnName),
input.ExampleWeightColumnName, DefaultColumnNames.Weight);
var data = new RoleMappedData(normalizedView, label, feature, null, weight);
return(new PredictorModelOutput
{
PredictorModel = new PredictorModelImpl(env, data, input.TrainingData,
OneVersusAllModelParameters.Create(host, input.UseProbabilities,
input.ModelArray.Select(p => p.Predictor as IPredictorProducing <float>).ToArray()))
});
}
}
private protected override MulticlassPredictionTransformer <OneVersusAllModelParameters> MakeTransformer(OneVersusAllModelParameters model, DataViewSchema trainSchema) => new MulticlassPredictionTransformer <OneVersusAllModelParameters>(Host, model, trainSchema, FeatureColumn.Name, LabelColumn.Name);
