Boost C++ Libraries of the most highly regarded and expertly designed C++ library projects in the world. Herb Sutter and Andrei Alexandrescu, C++ Coding Standards


Logistic Distribution

#include <boost/math/distributions/logistic.hpp>
namespace boost{ namespace math{

template <class RealType = double,
          class Policy   = policies::policy<> >
class logistic_distribution;

template <class RealType, class Policy>
class logistic_distribution
   typedef RealType value_type;
   typedef Policy   policy_type;
   // Construct:
   logistic_distribution(RealType location = 0, RealType scale = 1);
   // Accessors:
   RealType location()const; // location.
   RealType scale()const; // scale.


typedef logistic_distribution<> logistic;

}} // namespaces

The logistic distribution is a continuous probability distribution. It has two parameters - location and scale. The cumulative distribution function of the logistic distribution appears in logistic regression and feedforward neural networks. Among other applications, United State Chess Federation and FIDE use it to calculate chess ratings.

The following graph shows how the distribution changes as the parameters change:

Member Functions
logistic_distribution(RealType u = 0, RealType s = 1);

Constructs a logistic distribution with location u and scale s.

Requires scale > 0, otherwise a domain_error is raised.

RealType location()const;

Returns the location of this distribution.

RealType scale()const;

Returns the scale of this distribution.

Non-member Accessors

All the usual non-member accessor functions that are generic to all distributions are supported: Cumulative Distribution Function, Probability Density Function, Quantile, Hazard Function, Cumulative Hazard Function, mean, median, mode, variance, standard deviation, skewness, kurtosis, kurtosis_excess, range and support.

The domain of the random variable is [-[max_value], +[min_value]]. However, the pdf and cdf support inputs of +∞ and -∞ as special cases if RealType permits.

At p=1 and p=0, the quantile function returns the result of +overflow_error and -overflow_error, while the complement quantile function returns the result of -overflow_error and +overflow_error respectively.


The logistic distribution is implemented in terms of the std::exp and the std::log functions, so its accuracy is related to the accurate implementations of those functions on a given platform. When calculating the quantile with a non-zero position parameter catastrophic cancellation errors can occur: in such cases, only a low absolute error can be guaranteed.



Implementation Notes


Using the relation: pdf = e-(x-u)/s / (s*(1+e-(x-u)/s)2)


Using the relation: p = 1/(1+e-(x-u)/s)

cdf complement

Using the relation: q = 1/(1+e(x-u)/s)


Using the relation: x = u - s*log(1/p-1)

quantile from the complement

Using the relation: x = u + s*log(p/1-p)




The same as the mean.



kurtosis excess



(π*s)2 / 3