honey::Bootstrap< SampleT, Dim, Real__ > Class Template Reference

Monte Carlo (random-based) method to estimate the interval of confidence in a function's result, when that function operates on samples from a complex or unknown distribution. More...

#include <Bootstrap.h>

Classes
struct	meanFunc
	Functor to estimate the sample mean. More...
struct	varianceFunc
	Functor to estimate the sample variance. This is the unbiased estimator (mean over all possible samples is the population variance) More...

Public Types
typedef Real__	Real
typedef Vec< dim, Real >	Vec
typedef vector< const SampleT * >	SampleList
typedef function< Vec(const SampleList &)>	Func

Public Member Functions
	Bootstrap (const Func &func, RandomGen &gen, const vector< SampleT > &samples, Real alpha=0.05, szt bootSampleCount=50)
	Constructor, set up constants for all calculation calls. More...
	~Bootstrap ()
void	calc (Real progressDelta=1)
	Perform bootstrap calculation. The calculation can be split up over multiple calls. More...
Real	progress () const
	Current progress of calculation, from 0 (start) to 1 (complete). More...
const Vec &	lower () const
	Get lower bound of confidence interval (calculation must be complete) More...
const Vec &	upper () const
	Get upper bound of confidence interval (calculation must be complete) More...

Static Public Attributes
static const sdt	dim = Dim

Detailed Description

template<class SampleT, sdt Dim = 1, class Real__ = Real>
class honey::Bootstrap< SampleT, Dim, Real__ >

Monte Carlo (random-based) method to estimate the interval of confidence in a function's result, when that function operates on samples from a complex or unknown distribution.

If a function operates on sample data from a distribution, it will produce confident answers with little variation if it has enough samples to work with. For example, a function to find the mean of the number of heads per 10 coin tosses would operate on a large set coin toss samples, and the more samples, the more the mean approaches a stable value – 5 (the mean of a binomial distribution with N=10, p=0.5) But what if the coin is rigged? The distribution becomes unknown, thus the approaching mean is unknown.

The bootstrap method uses the empirical data available (the toss samples) to estimate the stability of the function result (the mean). This stability is expressed as a 95% confidence interval, which says that the mean of any set of toss samples will lie within the interval 95% of the time. It's important to note that the confidence interval is an estimation – the more samples provided, the more accurate and tighter the interval.

Template Parameters

SampleT	Bootstrap sample data type.
Dim	Dimension of the functor result.

See also

Bootstrap()

Algorithm from "Better Bootstrap Confidence Intervals", Efron, 1987.

Member Typedef Documentation

template<class SampleT , sdt Dim = 1, class Real__ = Real>

typedef function<Vec (const SampleList&)> honey::Bootstrap< SampleT, Dim, Real__ >::Func

template<class SampleT , sdt Dim = 1, class Real__ = Real>

typedef Real__ honey::Bootstrap< SampleT, Dim, Real__ >::Real

template<class SampleT , sdt Dim = 1, class Real__ = Real>

typedef vector<const SampleT*> honey::Bootstrap< SampleT, Dim, Real__ >::SampleList

template<class SampleT , sdt Dim = 1, class Real__ = Real>

typedef Vec<dim,Real> honey::Bootstrap< SampleT, Dim, Real__ >::Vec

Constructor & Destructor Documentation

template<class SampleT , sdt Dim = 1, class Real__ = Real>

honey::Bootstrap< SampleT, Dim, Real__ >::Bootstrap ( const Func &  func, RandomGen &  gen, const vector< SampleT > &  samples, Real  alpha = 0.05, szt  bootSampleCount = 50  )

inline

Constructor, set up constants for all calculation calls.

The functor operates on a selection of the original samples, and returns a vector of dimension Dim. For example, a coin toss mean functor would count the number of heads per 10 bool samples, and return the average count (a Real) in the form of a 1D vector.

Parameters

func	Functor to process samples
gen	Random generator
samples	Sample data to bootstrap.
alpha	The non-confidence of the interval, usually 5%, which gives a 95% confidence interval.
bootSampleCount	The number of function samples to take for estimating the interval. The higher the better, but above 100 is excessive. The total functor call count is: bootSampleCount + samples.size()

template<class SampleT , sdt Dim = 1, class Real__ = Real>

honey::Bootstrap< SampleT, Dim, Real__ >::~Bootstrap ( )

inline

Member Function Documentation

template<class SampleT , sdt Dim, class Real >

void honey::Bootstrap< SampleT, Dim, Real >::calc

(

Real

progressDelta = 1

)

Perform bootstrap calculation. The calculation can be split up over multiple calls.

When the calculation is complete (progress == 1), the confidence interval for each dimension of the result vector can be retrieved with lower() and upper().

Parameters

progressDelta

Percentage of progress to complete in this call. Range [0,1]

template<class SampleT , sdt Dim = 1, class Real__ = Real>

const Vec& honey::Bootstrap< SampleT, Dim, Real__ >::lower ( ) const

inline

Get lower bound of confidence interval (calculation must be complete)

template<class SampleT , sdt Dim = 1, class Real__ = Real>

Real honey::Bootstrap< SampleT, Dim, Real__ >::progress ( ) const

inline

Current progress of calculation, from 0 (start) to 1 (complete).

template<class SampleT , sdt Dim = 1, class Real__ = Real>

const Vec& honey::Bootstrap< SampleT, Dim, Real__ >::upper ( ) const

inline

Get upper bound of confidence interval (calculation must be complete)

Member Data Documentation

template<class SampleT , sdt Dim = 1, class Real__ = Real>

const sdt honey::Bootstrap< SampleT, Dim, Real__ >::dim = Dim

static

---

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

• src/common/Honey/Math/Random/Bootstrap.h