Re: bug in ols.m

[Jordi Gutiérrez Hermoso]

On 29 March 2011 06:18, Michael Creel <address@hidden> wrote:
> Last week I reported (on the tracker) a bug in ols.m that affects
> 3.4.0 and the devel sources, and which leads to incorrect results. It
> is a simple error in the degrees of freedom computation.  3.2.4 does
> not have the problem. I just wanted to call attention to it.

Thanks. I've pushed your patch:

     http://hg.savannah.gnu.org/hgweb/octave/rev/17997376291b

While I was there, I noticed that there were a few other improvements
in ols.m that could be done. It could use more tests that use the
other return values, which would have caught this bug. The docstring
needs some improvement, since the discrepancy with for example how S
is rendered leads it to sometimes being lowercase and sometimes being
uppercase in the plain text docstring (the one you read when you do
"help ols"); plus, it's not immediately clear to me which identity
matrix they mean in the Kronecker product. I had to think about these
for a minute. Also, momma always said to never do inv(A)*B, always
A\B, but perhaps it doesn't matter so much if A is symmetric positive
definite.

Would you consider writing patches for these?

Thanks,
- Jordi G. H.