Package jdistlib.inference

Class AcceleratedLogisticRegression

java.lang.Object

jdistlib.inference.AcceleratedLogisticRegression

All Implemented Interfaces:

AutoCloseable, BatchedDifferentiableLogDensity, ComputeBackedLogDensity, DifferentiableLogDensity, LogDensity

public final class AcceleratedLogisticRegression
extends Object
implements BatchedDifferentiableLogDensity, ComputeBackedLogDensity, AutoCloseable

Device-resident batched logistic-regression posterior with a spherical normal prior.

Constructor Summary

Constructors

Constructor	Description
AcceleratedLogisticRegression(ComputeBackend backend, double[][] design, double[] outcomes, double priorPrecision)

Method Summary

Modifier and Type	Method	Description
ComputeBackend	backend()
void	close()
ComputeBackend	computeBackend()
static AcceleratedLogisticRegression	create(Compute policy, double[][] design, double[] outcomes, double priorPrecision)	Selects a backend directly for vectorized or regular many-chain workflows.
int	dimensions()
static AcceleratedLogisticRegression	forNuts(SamplingOptions options, double[][] design, double[] outcomes, double priorPrecision)	Selects the backend from general sampling options for a NUTS workflow.
double	logDensityAndGradient(double[] state, double[] gradient)	Evaluates the log density and replaces gradient with its gradient.
void	logDensityAndGradientBatch(double[][] states, double[] logDensities, double[][] gradients)	Replaces every output row; implementations may execute the batch concurrently.
int	rows()

Methods inherited from class java.lang.Object

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

Methods inherited from interface jdistlib.inference.DifferentiableLogDensity

logDensity

Constructor Details

AcceleratedLogisticRegression

public AcceleratedLogisticRegression(ComputeBackend backend,
 double[][] design,
 double[] outcomes,
 double priorPrecision)

Method Details

forNuts

public static AcceleratedLogisticRegression forNuts(SamplingOptions options,
 double[][] design,
 double[] outcomes,
 double priorPrecision)

Selects the backend from general sampling options for a NUTS workflow.

create

public static AcceleratedLogisticRegression create(Compute policy,
 double[][] design,
 double[] outcomes,
 double priorPrecision)

Selects a backend directly for vectorized or regular many-chain workflows.

logDensityAndGradient

public double logDensityAndGradient(double[] state,
 double[] gradient)

Description copied from interface: DifferentiableLogDensity

Evaluates the log density and replaces gradient with its gradient.

Specified by:

logDensityAndGradient in interface DifferentiableLogDensity

logDensityAndGradientBatch

public void logDensityAndGradientBatch(double[][] states,
 double[] logDensities,
 double[][] gradients)

Description copied from interface: BatchedDifferentiableLogDensity

Replaces every output row; implementations may execute the batch concurrently.

Specified by:

logDensityAndGradientBatch in interface BatchedDifferentiableLogDensity

backend

public ComputeBackend backend()

computeBackend

public ComputeBackend computeBackend()

Specified by:

computeBackend in interface ComputeBackedLogDensity

rows

public int rows()

dimensions

public int dimensions()

close

public void close()

Specified by:

close in interface AutoCloseable