close Warning: Can't synchronize with repository "(default)" (/var/svn/tolp does not appear to be a Subversion repository.). Look in the Trac log for more information.

Opened 16 years ago

Last modified 16 years ago

#757 closed enhancement

Allowing arbitrary covariance matrix on BSR Ascii and Import API's — at Version 2

Reported by: Víctor de Buen Remiro Owned by: Víctor de Buen Remiro
Priority: highest Milestone: BSR API
Component: Math Version: 2.0.1
Severity: blocker Keywords:
Cc:

Description (last modified by Víctor de Buen Remiro)

Kernel of BSR can handle with arbitrary covariance matrices but Ascii nor Import API's does it.

Internally, they are needed three matrices:

  1. \Sigma : Simmetric positive define covariance matrix
  2. L : Choleski decomposition of covariance, \Sigma=L\cdot L^{T}
  3. L^{-T} : Choleski decomposition of inverse of covariance \Sigma^{-1}=L^{-T}\cdot L^{-1}

In real problems it's posible that we have precalculated some of these matrices. So, in order to be efficient, specially for large cases, it could be a good feature to admit at least one of these representations:

  1. Cov: When we have only the covariance. For example in prior nodes.
  2. CovInv : When we have the inverse of the covariance. For example, when it results from a previous linear regression.
  3. CovChol : When we have precalculate the Choleski decomposition
  4. CovInvChol : When we have precalculate the Choleski decomposition of inverse of covariance matrix.

Change History (2)

comment:1 Changed 16 years ago by Víctor de Buen Remiro

Status: new → accepted

comment:2 Changed 16 years ago by Víctor de Buen Remiro

Description: modified (diff)
Note: See TracTickets for help on using tickets.