# Boost-Commit :

Subject: [Boost-commit] svn:boost r49848 - sandbox/multivariate_distributions/doc
From: thijs_at_[hidden]
Date: 2008-11-20 09:23:11

Author: thijs
Date: 2008-11-20 09:23:10 EST (Thu, 20 Nov 2008)
New Revision: 49848
URL: http://svn.boost.org/trac/boost/changeset/49848

Log:
my first update!
Text files modified:
sandbox/multivariate_distributions/doc/multivariate_distributions.qbk | 115 +++++++++++++++++++++++++++++++++++++++
1 files changed, 114 insertions(+), 1 deletions(-)

Modified: sandbox/multivariate_distributions/doc/multivariate_distributions.qbk
==============================================================================
--- sandbox/multivariate_distributions/doc/multivariate_distributions.qbk (original)
+++ sandbox/multivariate_distributions/doc/multivariate_distributions.qbk 2008-11-20 09:23:10 EST (Thu, 20 Nov 2008)
@@ -11,6 +11,119 @@
]
]

+
+[section:introduction Introduction]
+
+in particular the multivariate Gaussian distribution.
+
+[endsect]
+
+
+[section:multivariate_distributions Multivariate Distribution]
+
+For the sake of testing the design and generic concepts of multivariate distributions,
+we will begin the inital design based two multivariate distributions.
+
+* Uniform
+* Gaussian
+
+[endsect]
+
+
+[section:multivariate_datatypes Datatype Concepts]
+
+(Non-)member function arguments and return types are (contrary to the univariate distribution)
+in need for multivariate extensions.
+
+To give an example, calculating the cdf of a multivariate Gaussian will involve the following steps:
+
+# construct a multivariate Gaussian using a *mean vector* and *covariance matrix*.
+# pass *the distribution object* and a *x vector*. The x vector represents the upperbound of the multivariate integration region.
+# return the probability which is a *scalar*.
+
+
+[h1 RealType]
+[h1 VectorType]
+[h1 MatrixType]
+[h1 DistributionType]
+
+[endsect]
+
+
+
+[section:anyvariate_nonmember Multivariate Non-member Accessors that have univariate equivalents]
+
+In this section we will discuss the compatibilities of functions accesors between the uni
+and miltivariate case.
+
+[h1 cdf]
+The cummulative probability distribution.
+
+* Input: VectorType, the upperbound of the multivariate integral.
+* Return: RealType
+
+[h1 cdf complement]
+[h1 chf]
+[h1 hazard]
+[h1 kurtosis]
+[h1 kurtosis_excess]
+
+[h1 mean]
+* Input: none
+* Return: VectorType
+
+[h1 median]
+* Input: none
+* Return: VectorType
+
+[h1 mode]
+
+[h1 pdf]
+
+* Input: VectorType, the density location.
+* Return: RealType
+
+[h1 range]
+
+[h1 quantile]
+
+[h1 quantile from the complement]
+
+[h1 skewness]
+
+[h1 standard_deviation]
+
+[h1 support]
+
+[h1 variance]
+
+[endsect]
+
+
+[section:multivariate_nonmember Multivariate Non-member Accessors that don\'t have an univariate equivalents]
+
+[h1 marginal]
+The marginal distribution returns the marginal distribution. A specific application wil be to create a univariate density out of a multivariate one!
+This is great, to combine with other functions, e.g. to get the variance along the first dimension of a multivariate Gaussian, you could use
+
+``
+ variance( marginal( multivariate_gaussian({0,0,0}, {1,2,3}), 1));
+``
+
+Input: ? Some way to define a set of variables to marginalize on. Could be a list of dimension indices, or a boolean mask.
+Return: DistributionType
+
+[h1 conditional]
+The conditional distribution.
+
+* Input: ? Some way to define a set of variables to condition on. Could be a list of dimension indices, or a boolean mask.
+* Input: VectorType the values of the conditioned variables
+* Return: DistributionType
+
+[endsect]
+
+