Re: rcond, condest, and a block 1-norm estimator

[John W. Eaton]

On 22-Nov-2007, David Bateman wrote:

| I'm moving this to the maintainers list as this is now a patch against
| Octave. What I've done in this patch relative to your code is
| 
| 1) Relicense under the GPL. I kept the original license as a comment,
| but perhaps that should be removed as well.
| 
| 2) Change the style to be closer to the rest of Octave.
|   - Limit the uppercase variables
|   - Help text immediately following the copyright
|   - write "if (a > 1)" and not "if a>1,"
| 
| 3) Eliminate the rcond function as 1/condest(...) gives the same..
| 
| 4) Change the name of block_onenorm_est to onenormest as something
| shorter. I hesitated to make this an internal function, but thought that
| a 1-norm estimator is useful in its own right. Also I didn't want to use
| the name normest1 as the interface is not the same as the matlab
| normest1 function.
| 
| 5) Replace the block of code
| 
|       colord = colamd (A);
|       Pc = speye (n);
|       Pc(:, colord) = Pc;
|       [L, U, P] = lu (A(:, Pc));
| 
| in condest.m with
| 
|       [L, U, P, Pc] = splu (A);
| 
| letting UMFPACK itself decide on the sparsity preserving permutation.
| 
| 6) Add condest to the sparse documentation. Maybe onenormest should be
| added as well.
| 
| Are you happy with the proposed changes? Can you suggest any further
| changes?
| 
| D.
| 
| 2007-11-22  Jason Riedy  <address@hidden>
| 
|       * linear-algebra/condest.m, linear-algebra/__onenormest__.m: New
|       functions.
|       * linear-algebra/Makefile.in (SOURCES): Add them to the sources.
| 
| 2007-11-22  David Bateman  <address@hidden>
| 
|       * interpreter/sparse.txi: Document condest.
| 
| 2007-11-22  David Bateman  <address@hidden>
| 
|       * PROJECTS: condest now implemented.

Please apply this patch.

Thanks,

jwe