LearningCurveOptions

Options & Configuration · AiDotNet.Evaluation.Options

Configuration options for learning curve analysis.

For Beginners

A learning curve plots model performance (y-axis) against training set size (x-axis). It helps answer questions like:

  • Will collecting more data help? (yes if training curve is still improving)
  • Is the model overfitting? (yes if train score >> validation score)
  • Is the model underfitting? (yes if both scores are low and flat)

How It Works

Learning curves show how model performance changes as training data size increases. This helps diagnose bias-variance tradeoffs and determine if more data would help.

Constructors

Constructor Summary
LearningCurveOptions()

Properties

Property Summary
CrossValidationStrategy? CVStrategy { get; set; } Cross-validation strategy for each point.
IReadOnlyList<string> MetricsToTrack { get; set; } Metrics to track on learning curve.
bool? ComputeBiasVarianceDecomposition { get; set; } Whether to compute bias-variance decomposition.
bool? ComputeConfidenceIntervals { get; set; } Whether to compute confidence intervals at each point.
bool? DiagnoseBiasVariance { get; set; } Whether to diagnose bias-variance condition.
bool? ExtrapolateCurve { get; set; } Whether to extrapolate learning curve beyond available data.
bool? ParallelExecution { get; set; } Whether to run evaluations in parallel.
bool? Shuffle { get; set; } Whether to shuffle before splitting.
bool? TrainSizesAreRatios { get; set; } Whether TrainSizes are ratios (0-1) or absolute counts.
double? ConfidenceLevel { get; set; } Confidence level for intervals.
double? ExtrapolationTarget { get; set; } Extrapolation target size (ratio or absolute).
double? HighBiasThreshold { get; set; } Threshold for high bias diagnosis (low scores).
double? HighVarianceThreshold { get; set; } Threshold for high variance diagnosis (train-test gap).
double? MinTrainSizeRatio { get; set; } Minimum training size ratio.
double[] TrainSizes { get; set; } Training set sizes to evaluate.
int? BiasVarianceBootstrapSamples { get; set; } Number of bootstrap samples for bias-variance decomposition.
int? CVFolds { get; set; } Number of CV folds.
int? MaxDegreeOfParallelism { get; set; } Maximum degree of parallelism.
int? NumberOfPoints { get; set; } Number of training sizes to auto-generate.
int? RandomSeed { get; set; } Random seed for reproducibility.