Project

General

Profile

Design #602

OrdMat: should it be a reference to a MatrixView in all PPOrderings?

Added by Anna Maria Bigatti over 9 years ago. Updated about 8 years ago.

Status:
Closed
Priority:
Normal
Assignee:
John Abbott
Category:
New Function
Target version:
CoCoALib-0.99540 Feb 2016
Start date:
31 Jul 2014
Due date:
% Done:

100%

Estimated time:
5.50 h
Spent time:
5.20 h

Description

The function OrdMat needs to create a new copy of the ordering matrix because not all PPOrdering objects actually contain a matrix (e.g. lex does not)
We could store a MatrixView in each PPOrdering.

Related issues

Related to CoCoALib - Feature #310: ordering and grading (weights) matrixClosed2013-02-13

Related to CoCoALib - Feature #803: PPOrdering: use it to compute WDeg?In Progress2015-11-11

History

#1 Updated by John Abbott over 8 years ago

  • Description updated (diff)

How important is it that OrdMat is fast? Is there a good example where we would want to compute repeatedly the OrdMat for an ordering? [JAA has doubts]

Remember KISS!

#2 Updated by John Abbott over 8 years ago

JAA now thinks that there is an example (see issue #803) where it could be useful to have the order matrix(view) quickly available from a PPOrdering.

Having the order matrix readily accessible may also help simplify (a little) the impl in PPMonoidEv.C of the various concrete subclasses of CmpBase.

#3 Updated by John Abbott over 8 years ago

  • Status changed from New to In Progress
  • Assignee set to John Abbott
  • Target version changed from CoCoALib-1.0 to CoCoALib-0.99540 Feb 2016
  • % Done changed from 0 to 10

JAA is now convinced that having OrdMat return a "reference" (namely a ConstMatrixView) is correct: it was relatively easy to implement and should incur no measurable run-time overhead.

I have a running impl, but it is not yet checked in because there are some other related issues still to be resolved.

One important detail is the ring over which the matrices will be. In practice we require that the entries of the matrix be integer, so it seems to make most sense for the matrices to be over ZZ. However, previously the matrices were over QQ, though JAA thinks there was no compelling technical reason for this.

JAA notes that some C5 tests involved checks similar to OrdMat(P) = mat([[1,2],[3,4]]) and in C5 a matrix created from INT/RAT values is automatically over QQ even if all the values are actually integer -- this was to make interactive use simpler e.g. inverse(mat([[1,2],[3,4]])) fails if the matrix is over ZZ.

Making OrdMat return a matrix over ZZ means the tests have to be modified slightly: e.g. OrdMat(P) = mat(ZZ,[[1,2],[3,4]]). JAA thinks the extra onus of having to specify ZZ explicitly in the tests is quite acceptable.

Anna, what do you think?

#4 Updated by John Abbott about 8 years ago

  • Status changed from In Progress to Closed
  • % Done changed from 10 to 100
  • Estimated time changed from 6.00 h to 5.50 h

Also available in: Atom PDF