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

Closed 16 years ago

Last modified 16 years ago

#691 closed defect (fixed)

BSR with ARIMA block estimation becomes slower in v1.1.7 b.9 2009-05-07 15:34:13 i686-win

Reported by: Víctor de Buen Remiro Owned by: Víctor de Buen Remiro
Priority: highest Milestone: Mantainance
Component: Math Version: 1.1.7
Severity: blocker Keywords: BSR, ARIMA
Cc:

Description

BSR with ARIMA block estimation becomes slower in v1.1.7 b.9 2009-05-07 15:34:13 i686-win, since ticket #676 was fixed.

Change History (3)

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

Resolution: → fixed
Status: new → closed

(In [991]) Fixes #691

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

(In [992]) Fixes #691

comment:3 Changed 16 years ago by Víctor de Buen Remiro

Certain cell by cell VMatrix functions like Log were skiping non explicit cells of sparse matrix but Log(0) is not zero. Now, functions non matching f(0)==0 must to create a dense copy of tha matrix and apply to it.

When we use this expression VMatSum(Log(SubBand(L,0) to calculate the logarithm of determinant of a triangular sparse matrix in order to evaluate ARMA blocks, SubBand returns a matrix with the same dimensions than L and Log needs to build a dense copy of L that could be very large. So simulation becomes very slower and afterwords logarithm of determinant is calculated as -Inf

Problem is fixed simply by using SubDiag, that returns a row vector, instead of SubBand, that returns a full matrix.

Note: See TracTickets for help on using tickets.