On Sat, Mar 16, 2013 at 4:28 PM, Daniel J Sebald <address@hidden
<mailto:address@hidden>> wrote:
On 03/16/2013 05:18 PM, Ed Meyer wrote:
I fixed the tests last fall according to what I consider the
right way to
do tests like this. As I have said several times before tests
should not
in general use absolute tolerances, rather they should take into
account
the size of the data. So for example instead of
assert (fftfilt (b, r * x),r * r * [1, 1, 0, 0, 0, 0, 0, 0, 0,
0],eps)
I used
%! y = r*r*[1 1 0 0 0 0 0 0 0 0];
%! tol = 2*norm(y)*eps;
%! assert (fftfilt (b, r*x), y, tol);
There are a number of tests like this (eig.cc for example) with
absolute
tolerances which should be using norms. But don't take my word
for it,
see, e.g. Golub & Van Loan, "Matrix Computations".
Yes, I'm a bit sketchy, but I remember looking at the work you had
done. I thought you checked it in. But this file log would suggest
it wasn't:
http://hg.savannah.gnu.org/__hgweb/octave/log/7d0848353eaf/__scripts/signal/fftfilt.m
<http://hg.savannah.gnu.org/hgweb/octave/log/7d0848353eaf/scripts/signal/fftfilt.m>
Dan
As I recall you made some changes that wiped out what I sent you and we
never got them back in before you checked it in. I could put a patch up but
for some reason other patches I've put up don't get applied so there is
something
I don't understand about the process.