selector

Type Members

1. sealed abstract class Distribution extends EnumEntry with Serializable

   

2. case class ModelEvaluation(modelUID: String, modelName: String, modelType: String, metricValues: EvaluationMetrics, modelParameters: Map[String, Any]) extends MetadataLike with Product with Serializable

   

   Evaluation summary of model

   Evaluation summary of model

   modelUID

   uid for winning model

   modelName

   unique name for model run

   modelType

   simple name of type of model

   metricValues

   evaluation metrics for model

   modelParameters

   parameter settings for model

3. trait ModelSelectorFactory extends AnyRef

   

   Creates the model selector class

4. case class ModelSelectorSummary(validationType: ValidationType, validationParameters: Map[String, Any], dataPrepParameters: Map[String, Any], dataPrepResults: Option[SplitterSummary], evaluationMetric: EvalMetric, problemType: ProblemType, bestModelUID: String, bestModelName: String, bestModelType: String, validationResults: Seq[ModelEvaluation], trainEvaluation: EvaluationMetrics, holdoutEvaluation: Option[EvaluationMetrics]) extends MetadataLike with Product with Serializable

   

   This is used to store all information about fitting and model selection generated by the model selector class

   This is used to store all information about fitting and model selection generated by the model selector class

   validationType

   type of validation performed to select hyper parameters

   validationParameters

   parameters on validation

   dataPrepParameters

   parameters on data preparation before hyper parameter tuning

   dataPrepResults

   changes made to the data in data preparation

   evaluationMetric

   metric used to select hyper parameters and model

   problemType

   type of modeling (eg binary classification, regressionm etc)

   bestModelUID

   best model UID

   validationResults

   model with parameters and metric for all evaluated

   trainEvaluation

   winning model performance on training data set

   holdoutEvaluation

   winning model performance on holdout data set

5. sealed trait ProblemType extends EnumEntry with Serializable

   

6. class RandomParamBuilder extends AnyRef

   

   Builder for a param sets used in random search-based model selection.

7. final class SelectedCombinerModel extends Model[SelectedCombinerModel] with OpTransformer3[RealNN, Prediction, Prediction, Prediction] with HasTestEval

   

8. trait SelectedCombinerParams extends Params

   

   Parameters for SelectorCombiner

9. final class SelectedModel extends Model[SelectedModel] with SparkWrapperParams[Model[_]] with OpTransformer2[RealNN, OPVector, Prediction] with HasTestEval

   

   Returned wrapped best model from model selector estimator

10. class SelectedModelCombiner extends Estimator[SelectedCombinerModel] with OpPipelineStage3[RealNN, Prediction, Prediction, Prediction] with SelectedCombinerParams with HasEval

    

    Class used to combine the predictions produced by two model selectors into a single prediction.

    Class used to combine the predictions produced by two model selectors into a single prediction. Does this by either taking the best models prediction or a combination of the two predictions that is either weighted by the accuracy measure or equal. Uses the summary information from the model selectors to determine the accuracy of the predictions and reruns evaluation (both train and test) when the predictions are combined.

11. sealed abstract class ValidationType extends EnumEntry with Serializable