ldes_stats.h File Reference

The estimators of the native LDES engine: running integrals, the MSER-5 warmup filter, and the batch-means confidence intervals. More...

#include <algorithm>
#include <cmath>
#include <cstddef>
#include <cstdint>
#include <limits>
#include <string>
#include <vector>
#include "line/solvers/wrappers/ldes/ldes_options.h"
#include "line/util/matrix.h"

Include dependency graph for ldes_stats.h:

Go to the source code of this file.

Classes
struct	line::ldes::engine::Accum
	The running per-(station, class) integrals and tallies. More...
struct	line::ldes::engine::Truncation
	Where the warmup ended, and whether a truncation was applied at all. More...
struct	line::ldes::engine::Observations
	The event-spaced observation series MSER-5 and the CI both read. More...
struct	line::ldes::engine::StatTriple
	Grand mean, standard error and degrees of freedom of a batch-means estimate. More...
class	line::ldes::engine::Convergence
	Convergence-based stopping: batch the four metrics and stop once EVERY active (station, class) pair has reached the requested relative precision. More...

Namespaces
namespace	line
namespace	line::ldes
namespace	line::ldes::engine

Functions
int	line::ldes::engine::mser5_truncation (const std::vector< double > &obs, int batch_size)
	MSER-5 truncation point over a series of observations, in BATCHES.
double	line::ldes::engine::overlap_adjustment (double overlap)
	The variance inflation of OVERLAPPING batch means.
StatTriple	line::ldes::engine::bm_statistics (const std::vector< double > &obs, int batch_size)
	Non-overlapping batch means.
StatTriple	line::ldes::engine::obm_statistics (const std::vector< double > &obs, int batch_size, double overlap)
	Overlapping batch means.
double	line::ldes::engine::t_critical (double level, int df)
	The two-sided t critical value, from the REFERENCE'S TABLE.
StatTriple	line::ldes::engine::spectral_statistics (const std::vector< double > &obs, int batch_size, double low_freq_frac)
	Heidelberger-Welch spectral estimate of the variance of the sample mean.
StatTriple	line::ldes::engine::ci_statistics (const std::vector< double > &obs, int batch_size, const LdesOptions &o)
	Dispatch on cimethod.
void	line::ldes::engine::batch_means_ci (const Observations &obs, const Truncation &tr, const LdesOptions &o, LdesResult &res)
	Fill the half-width matrices of res from the post-warmup observation series.

Detailed Description

The estimators of the native LDES engine: running integrals, the MSER-5 warmup filter, and the batch-means confidence intervals.

THE INTEGRALS ARE LAZY. tot_qlen(i,r) is the time integral of the class-r queue length at station i, advanced only when that pair CHANGES. Advancing every pair on every event would be O(M*K) per event on a hot loop where the event touches one pair; the reference does the same thing with lastQueueUpdateTime, and the invariant is that update_qlen is called BEFORE any write to qlen, never after.

THE WARMUP FILTER IS MSER-5 over EVENT-SPACED observations, not time-spaced ones. That matters for what the estimators may then do with the series: an unweighted mean of per-interval averages overweights congested epochs, because a congested epoch contains more events and therefore more intervals per unit time. So the means below are integral DIFFERENCES over elapsed time, and only the CI – which needs a sequence of comparable observations, not a mean – reads the per-interval series directly.

Definition in file ldes_stats.h.