org.apache.spark.mllib.optimization

Class GradientDescent

Object

org.apache.spark.mllib.optimization.GradientDescent

All Implemented Interfaces:

java.io.Serializable, org.apache.spark.internal.Logging, Optimizer

public class GradientDescent
extends Object
implements Optimizer, org.apache.spark.internal.Logging

Class used to solve an optimization problem using Gradient Descent. param: gradient Gradient function to be used. param: updater Updater to be used to update weights after every iteration.

See Also:

Serialized Form

Nested Class Summary

Nested classes/interfaces inherited from interface org.apache.spark.internal.Logging

org.apache.spark.internal.Logging.SparkShellLoggingFilter

Method Summary

Modifier and Type	Method and Description
Vector	optimize(RDD<scala.Tuple2<Object,Vector>> data, Vector initialWeights) Runs gradient descent on the given training data.
scala.Tuple2<Vector,double[]>	optimizeWithLossReturned(RDD<scala.Tuple2<Object,Vector>> data, Vector initialWeights) Runs gradient descent on the given training data.
static void	org$apache$spark$internal$Logging$$log__$eq(org.slf4j.Logger x$1)
static org.slf4j.Logger	org$apache$spark$internal$Logging$$log_()
static scala.Tuple2<Vector,double[]>	runMiniBatchSGD(RDD<scala.Tuple2<Object,Vector>> data, Gradient gradient, Updater updater, double stepSize, int numIterations, double regParam, double miniBatchFraction, Vector initialWeights) Alias of runMiniBatchSGD with convergenceTol set to default value of 0.001.
static scala.Tuple2<Vector,double[]>	runMiniBatchSGD(RDD<scala.Tuple2<Object,Vector>> data, Gradient gradient, Updater updater, double stepSize, int numIterations, double regParam, double miniBatchFraction, Vector initialWeights, double convergenceTol) Run stochastic gradient descent (SGD) in parallel using mini batches.
GradientDescent	setConvergenceTol(double tolerance) Set the convergence tolerance.
GradientDescent	setGradient(Gradient gradient) Set the gradient function (of the loss function of one single data example) to be used for SGD.
GradientDescent	setMiniBatchFraction(double fraction) Set fraction of data to be used for each SGD iteration.
GradientDescent	setNumIterations(int iters) Set the number of iterations for SGD.
GradientDescent	setRegParam(double regParam) Set the regularization parameter.
GradientDescent	setStepSize(double step) Set the initial step size of SGD for the first step.
GradientDescent	setUpdater(Updater updater) Set the updater function to actually perform a gradient step in a given direction.

Methods inherited from class Object

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

Methods inherited from interface org.apache.spark.internal.Logging

$init$, initializeForcefully, initializeLogIfNecessary, initializeLogIfNecessary, initializeLogIfNecessary$default$2, initLock, isTraceEnabled, log, logDebug, logDebug, logError, logError, logInfo, logInfo, logName, logTrace, logTrace, logWarning, logWarning, org$apache$spark$internal$Logging$$log__$eq, org$apache$spark$internal$Logging$$log_, uninitialize

Method Detail

runMiniBatchSGD

public static scala.Tuple2<Vector,double[]> runMiniBatchSGD(RDD<scala.Tuple2<Object,Vector>> data,
                                                            Gradient gradient,
                                                            Updater updater,
                                                            double stepSize,
                                                            int numIterations,
                                                            double regParam,
                                                            double miniBatchFraction,
                                                            Vector initialWeights,
                                                            double convergenceTol)

Run stochastic gradient descent (SGD) in parallel using mini batches. In each iteration, we sample a subset (fraction miniBatchFraction) of the total data in order to compute a gradient estimate. Sampling, and averaging the subgradients over this subset is performed using one standard spark map-reduce in each iteration.

Parameters:

data - Input data for SGD. RDD of the set of data examples, each of the form (label, [feature values]).

gradient - Gradient object (used to compute the gradient of the loss function of one single data example)

updater - Updater function to actually perform a gradient step in a given direction.

stepSize - initial step size for the first step

numIterations - number of iterations that SGD should be run.

regParam - regularization parameter

miniBatchFraction - fraction of the input data set that should be used for one iteration of SGD. Default value 1.0.

convergenceTol - Minibatch iteration will end before numIterations if the relative difference between the current weight and the previous weight is less than this value. In measuring convergence, L2 norm is calculated. Default value 0.001. Must be between 0.0 and 1.0 inclusively.

initialWeights - (undocumented)

Returns:

A tuple containing two elements. The first element is a column matrix containing weights for every feature, and the second element is an array containing the stochastic loss computed for every iteration.

runMiniBatchSGD

public static scala.Tuple2<Vector,double[]> runMiniBatchSGD(RDD<scala.Tuple2<Object,Vector>> data,
                                                            Gradient gradient,
                                                            Updater updater,
                                                            double stepSize,
                                                            int numIterations,
                                                            double regParam,
                                                            double miniBatchFraction,
                                                            Vector initialWeights)

Alias of runMiniBatchSGD with convergenceTol set to default value of 0.001.

Parameters:

data - (undocumented)

gradient - (undocumented)

updater - (undocumented)

stepSize - (undocumented)

numIterations - (undocumented)

regParam - (undocumented)

miniBatchFraction - (undocumented)

initialWeights - (undocumented)

Returns:

(undocumented)

org$apache$spark$internal$Logging$$log_

public static org.slf4j.Logger org$apache$spark$internal$Logging$$log_()

org$apache$spark$internal$Logging$$log__$eq

public static void org$apache$spark$internal$Logging$$log__$eq(org.slf4j.Logger x$1)

setStepSize

public GradientDescent setStepSize(double step)

Set the initial step size of SGD for the first step. Default 1.0. In subsequent steps, the step size will decrease with stepSize/sqrt(t)

Parameters:

step - (undocumented)

Returns:

(undocumented)

setMiniBatchFraction

public GradientDescent setMiniBatchFraction(double fraction)

Set fraction of data to be used for each SGD iteration. Default 1.0 (corresponding to deterministic/classical gradient descent)

Parameters:

fraction - (undocumented)

Returns:

(undocumented)

setNumIterations

public GradientDescent setNumIterations(int iters)

Set the number of iterations for SGD. Default 100.

Parameters:

iters - (undocumented)

Returns:

(undocumented)

setRegParam

public GradientDescent setRegParam(double regParam)

Set the regularization parameter. Default 0.0.

Parameters:

regParam - (undocumented)

Returns:

(undocumented)

setConvergenceTol

public GradientDescent setConvergenceTol(double tolerance)

Set the convergence tolerance. Default 0.001 convergenceTol is a condition which decides iteration termination. The end of iteration is decided based on below logic.

- If the norm of the new solution vector is greater than 1, the diff of solution vectors is compared to relative tolerance which means normalizing by the norm of the new solution vector. - If the norm of the new solution vector is less than or equal to 1, the diff of solution vectors is compared to absolute tolerance which is not normalizing.

Must be between 0.0 and 1.0 inclusively.

Parameters:

tolerance - (undocumented)

Returns:

(undocumented)

setGradient

public GradientDescent setGradient(Gradient gradient)

Set the gradient function (of the loss function of one single data example) to be used for SGD.

Parameters:

gradient - (undocumented)

Returns:

(undocumented)

setUpdater

public GradientDescent setUpdater(Updater updater)

Set the updater function to actually perform a gradient step in a given direction. The updater is responsible to perform the update from the regularization term as well, and therefore determines what kind or regularization is used, if any.

Parameters:

updater - (undocumented)

Returns:

(undocumented)

optimize

public Vector optimize(RDD<scala.Tuple2<Object,Vector>> data,
                       Vector initialWeights)

Runs gradient descent on the given training data.

Specified by:

optimize in interface Optimizer

Parameters:

data - training data

initialWeights - initial weights

Returns:

solution vector

optimizeWithLossReturned

public scala.Tuple2<Vector,double[]> optimizeWithLossReturned(RDD<scala.Tuple2<Object,Vector>> data,
                                                              Vector initialWeights)

Runs gradient descent on the given training data.

Parameters:

data - training data

initialWeights - initial weights

Returns:

solution vector and loss value in an array