LogStats Class Reference

Record running statistics on a logarithmic scale. More...

#include <log-stats.h>

Public Member Functions
void	add (double x)
long	count () const
double	mean () const
double	stddev () const

Private Attributes
long	n = 0
double	m = 0
double	s = 0

Detailed Description

Record running statistics on a logarithmic scale.

This class is primarily meant to record statistics on the time steps dt and magnetization variations per time step maxdu. As these quantities may vary over multiple orders of magnitude, the statistics of the logarithms are likely to be more relevant that the statistics on the quantities themselves.

For the sake of numerical stability, this class uses Welford's online algorithm.

Member Function Documentation

add()

void LogStats::add ( double  x )

inline

Add a sample to the statistics.

count()

long LogStats::count ( ) const

inline

Return the count of samples added so far.

mean()

double LogStats::mean ( ) const

inline

Return the geometric mean, i.e. the exponential of the mean of the logarithms.

For a distribution that is wide on a logarithmic scale, this is usually a better indication of what a “typical” value is. For a log-normal distribution, this is the same as the median.

stddev()

double LogStats::stddev ( ) const

inline

Return the sample standard deviation of the logarithm.

For a narrow distribution, this can be interpreted as a “relative width”. For example, the value 0.05 means a width roughly equal to 5% of the mean. Larger values can be interpreted as expressing the width via multiplicative factors. For example, the value 2.3 (i.e. log(10)) can be interpreted as a distribution where most samples are within a factor 10 of the geometric mean.

Member Data Documentation

m

double LogStats::m = 0

private

sample mean

n

long LogStats::n = 0

private

sample count

s

double LogStats::s = 0

private

sum of squares of deviations

---

The documentation for this class was generated from the following file:

• log-stats.h