Table of Contents

Class PolynomialLRScheduler

Namespace
AiDotNet.LearningRateSchedulers
Assembly
AiDotNet.dll

Decays the learning rate using a polynomial function.

public class PolynomialLRScheduler : LearningRateSchedulerBase, ILearningRateScheduler
Inheritance
PolynomialLRScheduler
Implements
Inherited Members

Examples

// Polynomial decay with power=2 (quadratic)
var scheduler = new PolynomialLRScheduler(
    baseLearningRate: 0.1,
    totalSteps: 100,
    power: 2.0,
    endLearningRate: 0.001
);

Remarks

PolynomialLR decays the learning rate from the initial value to a minimum value using a polynomial function. The decay curve can be controlled by the power parameter - power=1 gives linear decay, power>1 gives faster initial decay, power<1 gives slower initial decay.

For Beginners: This scheduler provides flexible control over how fast the learning rate decreases. With power=1, it's a straight line decrease. With power=2, it decreases slowly at first then more rapidly. With power=0.5, it decreases rapidly at first then slows down. This flexibility lets you customize the decay curve to your specific training needs.

Formula: lr = (base_lr - end_lr) * (1 - step/total_steps)^power + end_lr

Constructors

PolynomialLRScheduler(double, int, double, double)

Initializes a new instance of the PolynomialLRScheduler class.

public PolynomialLRScheduler(double baseLearningRate, int totalSteps, double power = 1, double endLearningRate = 0)

Parameters

baseLearningRate double

The initial learning rate.

totalSteps int

Total number of steps over which to decay.

power double

The power of the polynomial. Default: 1.0 (linear)

endLearningRate double

The final learning rate. Default: 0

Exceptions

ArgumentException

Thrown when parameters are invalid.

Properties

EndLearningRate

Gets the end learning rate.

public double EndLearningRate { get; }

Property Value

double

Power

Gets the polynomial power.

public double Power { get; }

Property Value

double

TotalSteps

Gets the total number of steps.

public int TotalSteps { get; }

Property Value

int

Methods

ComputeLearningRate(int)

Computes the learning rate for a given step.

protected override double ComputeLearningRate(int step)

Parameters

step int

The step number.

Returns

double

The computed learning rate.

GetState()

Gets the scheduler state for serialization/checkpointing.

public override Dictionary<string, object> GetState()

Returns

Dictionary<string, object>

A dictionary containing the scheduler state.

Edit this page