magpie.models.regression

Interface AbstractRegressionModel

All Known Implementing Classes:

AbstractNonlinearRegression, AbstractParsedNonlinearRegression, BaseRegression, ClassificationRegression, CompositeRegression, CoulombEwaldMatrixRegression, CoulombSineMatrixRegression, GuessMeanRegression, LASSORegression, LinearCorrectedRegression, MetallurgicalHeuristicRegression, MixingRuleRegression, MultiObjectiveRegression, MultiPropertyRegression, NonlinearRegressionExpr, PolynomialRegression, PRDFRegression, RandomGuessRegression, ScikitLearnRegression, SimpleLinearModelExample, SingleGuessRegression, SplitRegression, StagedRegression, StructureKRRBasedRegression, WekaRegression

public interface AbstractRegressionModel

Interface for all regression models.

Method Summary

Modifier and Type	Method and Description
boolean	doRobustRegression() Whether robust regression is desired.
int	getNFittingParameters() Number of fitting parameters in a model.
double	getRobustRegressionQ() Get desired False Discovery Rate
void	robustTraining(Dataset TrainData) Robustly train a model.
void	setRobustRegressionQ(double Q) Set the desired False Discovery Rate for outlier detection.

Method Detail

robustTraining

void robustTraining(Dataset TrainData)

Robustly train a model. Is called from BaseModel.train(magpie.data.Dataset).

I recommend that you use methods outlined in a paper by Motulsky and Brown.

.

Parameters:

TrainData - Training data

setRobustRegressionQ

void setRobustRegressionQ(double Q)

Set the desired False Discovery Rate for outlier detection. A robust regression parameter of 0.01 is recommended. Zero turns it off. Don't give it less than zero.

See paper by Benjamini and Hochberg

Parameters:

Q - Desired FDR

getRobustRegressionQ

double getRobustRegressionQ()

Get desired False Discovery Rate

Returns:

Target FDR

doRobustRegression

boolean doRobustRegression()

Whether robust regression is desired. For now, this should be:
getRobustRegressionQ > 0

getNFittingParameters

int getNFittingParameters()

Number of fitting parameters in a model. Only pertinent to nonlinear regression. Otherwise, this should just return 0.

Returns:

Number of fitting parameters