smile.math

Interface TimeFunction

All Superinterfaces:

java.io.Serializable

public interface TimeFunction
extends java.io.Serializable

A time-dependent function, e.g. the learning rate function, 1-dimensional neighborhood function, etc.

Method Summary

Modifier and Type	Method and Description
static TimeFunction	constant(double alpha) Returns the constant learning rate.
static TimeFunction	exp(double alpha, double C) Returns the power series learning rate lambda (alpha * exp(-t/C)).
static TimeFunction	inverse(double alpha, double C) Returns the inverse learning rate lambda (alpha * C / (C + t)).
static TimeFunction	linear(double alpha, double T) Returns the linear learning rate lambda (alpha * (1 - t/T)).
double	of(int t) Returns the learning rate at a given iteration.

Method Detail

of

double of(int t)

Returns the learning rate at a given iteration.

Parameters:

t - the order number of current iteration.

constant

static TimeFunction constant(double alpha)

Returns the constant learning rate.

Parameters:

alpha - the learning rate.

linear

static TimeFunction linear(double alpha,
                           double T)

Returns the linear learning rate lambda (alpha * (1 - t/T)).

Parameters:

alpha - the initial learning rate.

T - the number of iterations.

inverse

static TimeFunction inverse(double alpha,
                            double C)

Returns the inverse learning rate lambda (alpha * C / (C + t)). where C is typically a small percentage of the number of iterations T (e.g. T / 10).

Parameters:

alpha - the initial learning rate.

C - a small percentage of the number of iterations.

exp

static TimeFunction exp(double alpha,
                        double C)

Returns the power series learning rate lambda (alpha * exp(-t/C)). where C is typically smaller than the number of iterations T (e.g. T / 4).

Parameters:

alpha - the initial learning rate.

C - a small percentage of the number of iterations.