Re: test fail in ode/decic.m

[Rik]

On 11/30/2018 09:00 AM, address@hidden wrote:

Subject:

Re: test fail in ode/decic.m

From:

Kai Torben Ohlhus <address@hidden>

Date:

11/29/2018 12:59 PM

To:

address@hidden

CC:

address@hidden

List-Post:

<mailto:address@hidden>

Precedence:

list

MIME-Version:

1.0

References:

<address@hidden>

In-Reply-To:

<address@hidden>

Message-ID:

<address@hidden>

Content-Type:

multipart/alternative; boundary="000000000000d5ed01057bd3f4e1"

Message:

2

On Thu, Nov 29, 2018 at 8:43 PM Dmitri A. Sergatskov <address@hidden> wrote:

Buildbots are failing at ode/decic.m test since about 4 days ago.

Dmitri.

For me as well.  Since [1] Octave's fminunc is now MATLAB compatible, but that test [2] needs to be relaxed to 1e-8 or 1e-7

!!!!! test failed

ASSERT errors for:  assert ([ynew(1:end), ypnew(1:end)],[ref1(1:end), ref2(1:end)],1e-10)

  Location  |  Observed  |  Expected  |  Reason

   (3,1)     -4.2594e-09       0         Abs err 4.2594e-09 exceeds tol 1e-10 by 4e-09

   (1,2)        -0.04        -0.04       Abs err 4.5867e-09 exceeds tol 1e-10 by 4e-09

   (2,2)         0.04         0.04       Abs err 1.3676e-08 exceeds tol 1e-10 by 1e-08

Kai

[1] https://hg.savannah.gnu.org/hgweb/octave/rev/1ae11ca7dceb

[2] https://hg.savannah.gnu.org/hgweb/octave/file/242aa7b2c783/scripts/ode/decic.m#l202

Yes.  Apparently decic.m internally calls fminunc.

The code is

  opt = optimset ("tolfun", TolFun, "tolx", TolX, "display", "iter-detailed");
  x = ...
    fminunc (@(x) objective (x, t0, y0, fixed_y0, yp0, fixed_yp0, nl, nu, fun),
             x0, opt);

I reverted to the old behavior by adding "FinDiffType", "central" to the list of options.  Also, there seemed to be an issue with the opts.  The property "display" is not implemented for fminunc.m, and even if it was the value "iter-detailed" is invalid.  I just removed that.

See https://hg.savannah.gnu.org/hgweb/octave/rev/7bd67e786e5f.

--Rik