Expand description
FluxBench Statistical Engine
Provides robust statistical analysis for benchmark results including:
- Bootstrap resampling with BCa (Bias-Corrected and Accelerated) intervals
- Outlier detection via IQR method
- Percentile calculation preserving tail latency signals
- Distribution characterization
- A/B comparison with effect size and probability of regression
Structs§
- Bootstrap
Config - Bootstrap configuration
- Bootstrap
Result - Result of bootstrap analysis
- Comparison
Config - Configuration for comparison
- Comparison
Result - Result of comparing two distributions
- Confidence
Interval - Confidence interval bounds
- Cycles
Statistics - CPU cycles statistics (computed alongside time stats)
- Outlier
Analysis - Result of outlier analysis
- Percentiles
- Standard percentiles to compute
- Summary
Statistics - Comprehensive summary statistics
Enums§
- Bootstrap
Method - Which bootstrap method was used
- Comparison
Error - Errors from comparison operations
- Effect
Interpretation - Interpretation of effect size magnitude
- Outlier
Method - Method for outlier detection
Constants§
- BCA_
THRESHOLD - Threshold below which BCa method is used instead of percentile
- DEFAULT_
BOOTSTRAP_ ITERATIONS - Default number of bootstrap iterations
- DEFAULT_
CONFIDENCE_ LEVEL - Default confidence level (95%)
Functions§
- compare_
distributions - Compare two distributions using bootstrap resampling
- compute_
bootstrap - Compute bootstrap confidence interval for the mean
- compute_
cycles_ stats - Compute CPU cycles statistics from raw cycle counts
- compute_
percentile - Compute a single percentile from samples
- compute_
percentiles - Compute all standard percentiles
- compute_
summary - Compute summary statistics with proper separation of cleaned vs raw data
- detect_
outliers - Detect outliers in samples using specified method