SelfPacedScheduler<T>
Models & Types · AiDotNet.CurriculumLearning.Schedulers
Self-paced curriculum scheduler that adapts sample selection based on model performance.
Inheritance: CurriculumSchedulerBase<T> → SelfPacedScheduler<T>
Implements: ISelfPacedScheduler<T>
For Beginners
Unlike fixed schedulers, self-paced learning dynamically selects samples based on the model’s current ability. Samples with loss below a threshold are considered “easy enough” and included in training. The threshold increases over time to include progressively harder samples.
How It Works
How It Works:
Self-Paced Regularizers:
References:
Constructors
| Constructor | Summary |
|---|---|
SelfPacedScheduler(int totalEpochs, T initialLambda, T lambdaGrowthRate, T maxLambda, SelfPaceRegularizer regularizer) |
Initializes a new instance of the SelfPacedScheduler class. |
Properties
| Property | Summary |
|---|---|
T CurrentThreshold { get; } |
Gets the current pace threshold (lambda). |
T GrowthRate { get; set; } |
Gets or sets the growth rate for the pace parameter. |
T PaceParameter { get; set; } |
Gets or sets the current pace threshold (lambda). |
Vector<T> SampleWeights { get; } |
Gets sample weights from the last selection. |
string Name { get; } |
Gets the name of this scheduler. |
Methods
| Method | Summary |
|---|---|
Dictionary<string, object> GetStatistics() |
Gets scheduler-specific statistics. |
T GetDataFraction() |
Gets the current data fraction (estimated from lambda progression). |
ValueTuple<int[], Vector<T>> SelectSamplesWithWeights(Vector<T> losses) |
Selects samples based on current losses and pace threshold. |
Vector<T> ComputeSampleWeights(Vector<T> losses) |
Computes sample weights based on current losses and pace threshold. |
bool StepEpoch(CurriculumEpochMetrics<T> epochMetrics) |
Updates lambda threshold based on epoch metrics. |
int[] GetCurrentIndices(int[] sortedIndices, int totalSamples) |
Selects samples based on difficulty scores. |
void Reset() |
Resets the scheduler to initial state. |