Project

General

Profile

Design #806

AssignZero for matrix

Added by John Abbott over 8 years ago. Updated about 8 years ago.

Status:
Closed
Priority:
Low
Assignee:
John Abbott
Category:
Safety
Target version:
CoCoALib-0.99540 Feb 2016
Start date:
14 Nov 2015
Due date:
% Done:

100%

Estimated time:
1.10 h
Spent time:
1.10 h

Description

There is a procedure AssignZero for a matrix; it calls the mem fn myAssignZero.

What should these fns do?

History

#1 Updated by John Abbott over 8 years ago

  • Status changed from New to In Progress
  • Priority changed from Normal to Low
  • % Done changed from 0 to 10

Before spending too much time deciding the exact semantics, we should also decide if the procedure is even useful -- when could it be useful?

For a normal matrix I think the meaning is fairly clear: the matrix size stays the same, and all entries are set to zero (of the ring of the matrix).

It is less clear to me what should happen to a "special" matrix, for instance a DiagMatrix. I see two obvious choices:
  • (A) set all the diag entries to zero (the rest are already zero)
  • (B) give error

The problem with approach (A) is that for a DiagMatrix called DM the call AssignZero(DM) is not equivalent to

for (int i=0; i < NumRows(DM); ++i)
  for (int j=0; j < NumCols(DM); ++j)
    DM.mySetEntry(i,j,0);

because the explicit loop will give an error (provided DM has dimension at least 2x2).

If we adopt approach (A), what should happen when attempting to AssignZero an object of type ZeroMatrix? Presumably nothing will happen, but perhaps an error would be more appropriate?

#2 Updated by John Abbott over 8 years ago

In some ways it would be nice to have a default impl of AssignZero; presumably as the mem fn MatrixView::myAssignZero. The most obvious default impl is the double loop in comment 1, and this would give error both for a DiagMat and a ZeroMat if they do not have their own "private" impls of the mem fn myAssignZero.

An alternative could be the following "intelligent" loop:

for (int i=0; i < NumRows(DM); ++i)
  for (int j=0; j < NumCols(DM); ++j)
  {
    if (IsZero(DM(i,j))) continue; // regardless of whether writable!!
    if (!DM.IsWritable(i,j))  CoCoA_ERROR(...);
    DM.mySetEntry(i,j,0);
  }

If we accept the "intelligent" loop definition, what should happen if someone tries Id.mySetEntry(0,0,1); where Id is an identity matrix? In other words, if we have a const matrix, is it then OK to set its entries to the values they already have? The "intelligent" loop sets a worrying precedent since it does this for entries containing 0.

Comments? Opinions?

NOTE (2015-11-26) corrected sample impl above!

#3 Updated by Anna Maria Bigatti over 8 years ago

John Abbott wrote:

In some ways it would be nice to have a default impl of AssignZero; presumably as the mem fn MatrixView::myAssignZero. The most obvious default impl is the double loop in comment 1, and this would give error both for a DiagMat and a ZeroMat if they do not have their own "private" impls of the mem fn myAssignZero.

The "intelligent loop" could become quite tricky and I'm not sure we should make a default implementation...
I think we could have the "trivial loop" as a "generic implementation" which could be called by the concrete classes where it makes sense.

#4 Updated by John Abbott over 8 years ago

  • Status changed from In Progress to Resolved
  • Assignee set to John Abbott
  • % Done changed from 10 to 80

Since the fn AssignZero (for matrices) was no longer used anywhere, and especially since its semantics were rather unclear (with no obvious clarification), I have removed it.

Removed the doc; checked in.

#5 Updated by John Abbott about 8 years ago

  • Status changed from Resolved to Closed
  • % Done changed from 80 to 100
  • Estimated time set to 1.10 h

Also available in: Atom PDF