jsat.math.decayrates

Interface DecayRate

All Superinterfaces:

Serializable

All Known Implementing Classes:

ExponetialDecay, InverseDecay, LinearDecay, NoDecay, PowerDecay

public interface DecayRate
extends Serializable

Many algorithms use a learning rate to adjust the step size by which the search space is covered. In practice, it is often useful to reduce this learning rate over time. In this way, large steps can be taken in the beginning when we are far from the solution, and smaller steps when we have gotten closer to the solution and do not want to step too far away.

Author:

Edward Raff

Method Summary

Modifier and Type	Method and Description
DecayRate	clone()
double	rate(double time, double initial) Decays the initial value over time.
double	rate(double time, double maxTime, double initial) Decays the initial value over time.

Method Detail

rate

double rate(double time,
            double maxTime,
            double initial)

Decays the initial value over time.

Parameters:

time - the current time through the algorithm in the range [0, maxTime]

maxTime - the maximum time step that will be seen

initial - the initial value

Returns:

the decayed value over time of the initial value

Throws:

ArithmeticException - if the time is negative

rate

double rate(double time,
            double initial)

Decays the initial value over time.

Parameters:

time - the current time step to return a value for

initial - the initial learning rate

Returns:

the decayed value

clone

DecayRate clone()