io.improbable.keanu.model.regression

Class RegressionModelBuilder<OUTPUT extends Tensor>

java.lang.Object

io.improbable.keanu.model.regression.RegressionModelBuilder<OUTPUT>

public class RegressionModelBuilder<OUTPUT extends Tensor>
extends java.lang.Object

Builder class for doing linear regression without regularisation.

See Also:

RegressionModel

Constructor Summary

Constructors

Constructor and Description
RegressionModelBuilder(DoubleTensor inputTrainingData, OUTPUT outputTrainingData, java.util.function.Function<DoubleVertex,LinearRegressionGraph.OutputVertices<OUTPUT>> outputTransform)

Method Summary

Modifier and Type	Method and Description
RegressionModel<OUTPUT>	build()
RegressionModel<OUTPUT>	buildWithoutFitting()
RegressionModelBuilder	withPriorOnIntercept(double mean, double scaleParameter)
RegressionModelBuilder	withPriorOnIntercept(DoubleTensor mean, DoubleTensor scaleParameter)
RegressionModelBuilder	withPriorOnIntercept(DoubleVertex mean, DoubleVertex scaleParameter) Set the input parameters to the distribution describing the prior belief about the intercept of the regression model
RegressionModelBuilder	withPriorOnWeights(double means, double scaleParameters)
RegressionModelBuilder	withPriorOnWeights(DoubleTensor means, DoubleTensor scaleParameters)
RegressionModelBuilder	withPriorOnWeights(DoubleVertex means, DoubleVertex scaleParameters) Set the input parameters to the distribution describing the prior belief about the weights of the regression model
RegressionModelBuilder	withPriorOnWeightsAndIntercept(double mean, double scaleParameter) Set the input parameters to the distribution describing the prior belief about both the intercept and weights of the regression model
RegressionModelBuilder	withRegularization(RegressionRegularization regularization)
RegressionModelBuilder	withSampling(SamplingModelFitting sampling) Optional - use a sampling algorithm to fit the model instead of the default, which is gradient optimization.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

RegressionModelBuilder

public RegressionModelBuilder(DoubleTensor inputTrainingData,
                              OUTPUT outputTrainingData,
                              java.util.function.Function<DoubleVertex,LinearRegressionGraph.OutputVertices<OUTPUT>> outputTransform)

Method Detail

withRegularization

public RegressionModelBuilder withRegularization(RegressionRegularization regularization)

withPriorOnWeights

public RegressionModelBuilder withPriorOnWeights(DoubleVertex means,
                                                 DoubleVertex scaleParameters)

Set the input parameters to the distribution describing the prior belief about the weights of the regression model

Parameters:

means - An array of means of the distribution describing the prior belief about the regression weights

scaleParameters - An array of scale parameters of the distribution describing the prior belief about the regression weights. This will represent sigmas if no or ridge regularization is used and will represent betas if lasso regularization is used.

Returns:

this

withPriorOnWeights

public RegressionModelBuilder withPriorOnWeights(DoubleTensor means,
                                                 DoubleTensor scaleParameters)

withPriorOnWeights

public RegressionModelBuilder withPriorOnWeights(double means,
                                                 double scaleParameters)

withPriorOnIntercept

public RegressionModelBuilder withPriorOnIntercept(DoubleVertex mean,
                                                   DoubleVertex scaleParameter)

Set the input parameters to the distribution describing the prior belief about the intercept of the regression model

Parameters:

mean - The mean of the distribution describing the prior belief about the regression intercept

scaleParameter - The scale parameter of the distribution describing the prior belief about the regression intercept. This will represent sigmas if no or ridge regularization is used and will represent betas if lasso regularization is used.

Returns:

this

withPriorOnIntercept

public RegressionModelBuilder withPriorOnIntercept(DoubleTensor mean,
                                                   DoubleTensor scaleParameter)

withPriorOnIntercept

public RegressionModelBuilder withPriorOnIntercept(double mean,
                                                   double scaleParameter)

withPriorOnWeightsAndIntercept

public RegressionModelBuilder withPriorOnWeightsAndIntercept(double mean,
                                                             double scaleParameter)

Set the input parameters to the distribution describing the prior belief about both the intercept and weights of the regression model

Parameters:

mean - The mean of the distribution describing the prior belief about both the regression intercept and weights

scaleParameter - The scale parameter of the distribution describing the prior belief about both regression intercept and weights. This will represent sigmas if no or ridge regularization is used and will represent betas if lasso regularization is used.

Returns:

this

withSampling

public RegressionModelBuilder withSampling(SamplingModelFitting sampling)

Optional - use a sampling algorithm to fit the model instead of the default, which is gradient optimization.

Parameters:

sampling - Defines the number of samples to take and the algorithm to use, e.g. MetropolisHastings

Returns:

this

build

public RegressionModel<OUTPUT> build()

Returns:

A linear regression model from the data passed to the builder

buildWithoutFitting

public RegressionModel<OUTPUT> buildWithoutFitting()