package jsat.math.optimization.stochastic;
import jsat.linear.Vec;
/**
* Performs unaltered Stochastic Gradient Decent updates computing
* x = x- η grad
*
* Because the SimpleSGD requires no internal state, it is not necessary to call
* {@link #setup(int) }.
*
* @author Edward Raff
*/
public class SimpleSGD implements GradientUpdater
{
private static final long serialVersionUID = 4022442467298319553L;
/**
* Creates a new SGD updater
*/
public SimpleSGD()
{
}
@Override
public void update(Vec x, Vec grad, double eta)
{
x.mutableSubtract(eta, grad);
}
@Override
public double update(Vec x, Vec grad, double eta, double bias, double biasGrad)
{
x.mutableSubtract(eta, grad);
return eta*biasGrad;
}
@Override
public SimpleSGD clone()
{
return new SimpleSGD();
}
@Override
public void setup(int d)
{
//no setup to be done
}
}
Ads help maintain this website.