package jsat.math.optimization.stochastic;
import java.io.Serializable;
import jsat.linear.Vec;
/**
* This interface defines the method of updating some weight vector using a
* gradient and a learning rate. The method may then apply its own set of
* learning rates on top of the given learning rate in order to accelerate
* convergence in general or for specific conditions / methods.
*
* @author Edward Raff
*/
public interface GradientUpdater extends Serializable
{
/**
* Updates the weight vector {@code x} such that x = x-ηf(grad),
* where f(grad) is some function on the gradient that effectively returns a
* new vector. It is not necessary for the internal implementation to ever
* explicitly form any of these objects, so long as {@code x} is mutated to
* have the correct result.
* @param w the vector to mutate such that is has been updated by the
* gradient
* @param grad the gradient to update the weight vector {@code x} from
* @param eta the learning rate to apply
*/
public void update(Vec w, Vec grad, double eta);
/**
* Updates the weight vector {@code x} such that x = x-ηf(grad),
* where f(grad) is some function on the gradient that effectively returns a
* new vector. It is not necessary for the internal implementation to ever
* explicitly form any of these objects, so long as {@code x} is mutated to
* have the correct result.
*
* This version of the update method includes two extra parameters to make
* it easer to use when a scalar bias term is also used
*
* @param w the vector to mutate such that is has been updated by the
* gradient
* @param grad the gradient to update the weight vector {@code x} from
* @param eta the learning rate to apply
* @param bias the bias term of the vector
* @param biasGrad the gradient for the bias term
* @return the value to change the bias by, the update being
* {@code bias = bias - returnValue}
*/
public double update(Vec w, Vec grad, double eta, double bias, double biasGrad);
/**
* Sets up this updater to update a weight vector of dimension {@code d}
* by a gradient of the same dimension
* @param d the dimension of the weight vector that will be updated
*/
public void setup(int d);
public GradientUpdater clone();
}
Ads help maintain this website.