[Libreoffice-bugs] [Bug 135679] New: Unwanted recalculation of RAND() data when using SOLVER, SOLVER does not use given start-values.

[bugzilla-daemon at bugs.documentfoundation.org]

https://bugs.documentfoundation.org/show_bug.cgi?id=135679

            Bug ID: 135679
           Summary: Unwanted recalculation of RAND() data when using
                    SOLVER, SOLVER does not use given start-values.
           Product: LibreOffice
           Version: 6.4.5.2 release
          Hardware: All
                OS: All
            Status: UNCONFIRMED
          Severity: normal
          Priority: medium
         Component: Calc
          Assignee: libreoffice-bugs at lists.freedesktop.org
          Reporter: gmildner at t-online.de

Description:
I generated ca. 240 (hopefully) gauss-distributed random numbers (as a column)
using NORMINV(RAND();NDmean;NDsigma), then made histogram data using
FREQUENCY(C6:C249;F6:F16).
The histogram is displayed by the Diagram feature.
Then a row with theoretical gauss-distributed data was generated using
NORMDIST(F6;mean;sigma;0)*nfac, plus another with the differencies of the 2
datasets.
Up to now everything worked well.
A nonlinear regression with the SOLVER worked, but extremely slow and it
stopped at 2000 iterations. "Continue"(Fortfahren) minimized the sum-of-squares
further.
(I have another file with a gauss-distribution fit and comparaby number of
classes. This one computes fast and comes to a solution at <2000 iterations.
The difference is: there is no RAND() function included!)
The 2nd and more grave problem with SOLVER is: when Stop (Anhalten) or OK has
been clicked to take-over the fitted params, a new set of Random-data is
generated which makes 
the fit useless.

Another unwanted behaviour is: SOLVER apparently does not use the start-values
given in the changeable cells.

You can reproduce this behaviour easily by the attached file.

Thanx & best wishes,
Gerd

Steps to Reproduce:
1.see text
2.see attatched ods-file
3.

Actual Results:
SOLVER-related Buttons trigger RAND()
Nonlin. Regression unuable when RAND() is present.

Expected Results:
obviously the negative of the above.


Reproducible: Always


User Profile Reset: No



Additional Info:
OS: Win10, LO: (64Bit)

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/libreoffice-bugs/attachments/20200812/9b1a1361/attachment-0001.htm>