SubgroupAnalysisOptions

Options & Configuration · AiDotNet.Evaluation.Options

Configuration options for subgroup (slice-based) analysis.

For Beginners

Your model might have 90% accuracy overall, but:

  • 95% accuracy on common cases, 50% on rare cases
  • 92% on young users, 75% on elderly users
  • 98% on clean data, 60% on noisy data

Subgroup analysis reveals these hidden variations that overall metrics miss.

How It Works

Subgroup analysis computes metrics for different slices of your data, helping identify where the model performs well or poorly. This is essential for understanding model behavior.

Constructors

Constructor Summary
SubgroupAnalysisOptions()

Properties

Property Summary
ContinuousBinningStrategy? BinningStrategy { get; set; } Binning strategy for continuous features.
Dictionary<int, double[]> CustomBinEdges { get; set; } Custom bin edges for specific features.
IReadOnlyList<string> MetricsToCompute { get; set; } Metrics to compute per slice.
ReportFormat? ReportFormat { get; set; } Report format.
SliceSortOrder? SliceSortOrder { get; set; } Sort order for slices in report.
bool? ApplyMultipleTestingCorrection { get; set; } Whether to apply multiple testing correction.
bool? AutoDetectCategoricalFeatures { get; set; } Whether to auto-detect categorical features for slicing.
bool? BinContinuousFeatures { get; set; } Whether to bin continuous features for slicing.
bool? ComputeConfidenceIntervals { get; set; } Whether to compute confidence intervals per slice.
bool? ComputeSliceImportance { get; set; } Whether to compute slice importance (which slices matter most).
bool? ComputeSliceIntersections { get; set; } Whether to compute intersections of slices.
bool? GenerateSliceErrorAnalysis { get; set; } Whether to generate error analysis per slice.
bool? IdentifyUnderperformingSlices { get; set; } Whether to identify underperforming slices.
bool? IncludeDistributionStats { get; set; } Whether to include slice distribution statistics.
bool? IncludeRecommendations { get; set; } Whether to include recommendations per slice.
bool? ParallelExecution { get; set; } Whether to run analyses in parallel.
bool? TestSignificantDifference { get; set; } Whether to perform statistical test vs overall.
double? ConfidenceLevel { get; set; } Confidence level.
double? SignificanceLevel { get; set; } Significance level for difference tests.
double? UnderperformanceThreshold { get; set; } Threshold for underperformance (relative to overall).
int? ContinuousBins { get; set; } Number of bins for continuous features.
int? MaxCategoricalUniques { get; set; } Maximum unique values for auto-detected categorical features.
int? MaxIntersectionSize { get; set; } Maximum intersection size.
int? MaxSlicesInReport { get; set; } Maximum slices to include in report.
int? MinSliceSize { get; set; } Minimum samples per slice for reliable metrics.
int[] SliceFeatureIndices { get; set; } Feature indices to slice by.
string[] SliceFeatureNames { get; set; } Feature names for slicing (alternative to indices).