Struct StatisticalAnalyzer 

pub struct StatisticalAnalyzer<T: Float + Debug + Send + Sync + 'static> { /* private fields */ }

Statistical analyzer for evaluation results

analysis_methods and multiple_comparison were configuration the analyzer stored and never consulted: it always produced descriptive statistics and never corrected a p-value, whatever they said. Both now drive behaviour — see StatisticalAnalyzer::analyze and StatisticalAnalyzer::adjusted_p_values — and the recorded StatisticalTests they operate on are readable through StatisticalAnalyzer::tests.

Implementations

impl<T: Float + Debug + Default + Sum + Send + Sync> StatisticalAnalyzer<T>

pub fn analysis_methods(&self) -> &[AnalysisMethod]

Analyses this analyzer performs, in order.

pub fn set_analysis_methods( &mut self, methods: Vec<AnalysisMethod>, ) -> Result<()>

Choose which analyses StatisticalAnalyzer::analyze performs.

Only the methods this analyzer actually implements are accepted. Enabling one it cannot perform is reported as OptimError::NotImplemented rather than accepted and silently skipped, which would let a caller believe a regression or cluster analysis had run.

pub fn analyze(&self, values: &[T]) -> Result<AnalysisReport<T>>

Run every enabled analysis over values.

pub fn tests(&self) -> &[StatisticalTest<T>]

Recorded statistical tests, in the order they were added.

pub fn multiple_comparison(&self) -> &MultipleComparisonCorrection

Multiple-comparison correction applied by StatisticalAnalyzer::adjusted_p_values.

pub fn set_multiple_comparison( &mut self, correction: MultipleComparisonCorrection, )

Choose the multiple-comparison correction.

pub fn adjusted_p_values(&self) -> Result<Vec<T>>

p-values of the recorded tests after the configured multiple-comparison correction, in the order the tests were added.

Every correction is closed-form and clamped to [0, 1]:

• None — the raw p-values.
• Bonferroni — m * p.
• Sidak — 1 - (1 - p)^m.
• HolmBonferroni — step-down (m - i) * p, made monotone.
• BenjaminiHochberg — step-up m * p / (i + 1), made monotone.
• BenjaminiYekutieli — Benjamini-Hochberg scaled by the harmonic number sum_{k=1..m} 1/k, which is the dependence-robust variant.

pub fn significant_tests(&self, alpha: T) -> Result<Vec<usize>>

Indices of the tests whose adjusted p-value is at or below alpha.

pub fn add_test(&mut self, test: StatisticalTest<T>)

Add a statistical test

pub fn get_threshold(&self, name: &str) -> Option<T>

Get significance threshold

pub fn set_threshold(&mut self, name: String, value: T)

Set significance threshold

pub fn compute_descriptive_stats( &self, values: &[T], ) -> Result<DescriptiveStats<T>>

Compute descriptive statistics.

An empty sample yields an all-zero summary with count == 0, which the caller can distinguish from a genuine all-zero sample. Fallible because the sample size has to reach the element type to be a divisor; there is no fallback divisor that would not silently report a different statistic.

Trait Implementations

impl<T: Debug + Float + Debug + Send + Sync + 'static> Debug for StatisticalAnalyzer<T>

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.