Re: fminuc problem

[Kai Torben Ohlhus]

On Sat, May 25, 2019 at 3:23 PM madhav <address@hidden> wrote:
>
> Hi,
>
> I tried to find the optimum theta using fminunc function with rand initial 
> theta for simple logistic regression model(not NN). But i got the following 
> warning as,
>
>                              warning: division by zero
>                              warning: called from
>                                fminunc at line 288 column 15
>
> But it provided an output theta which when i verified taking the training set 
> as testing data and it gave utter incorrect results and hence the output 
> theta is not the optimum theta. But when i initialized the initial theta as 
> zero vector, i got the correct optimum theta. Hence this function works good 
> only for zero vector initialization. So there must be some bug in the code of 
> the function fminunc. Is there any alternative function for fminunc?
>

Can you provide a minimal complete working example, including the
version number of Octave and your operating system to reproduce the
issue?  Nothing in the documentation of fminunc [1] is called "theta".
Do you mean the starting guess "x0"?  Please be more precise with your
terminology (best by providing code).  Otherwise I cannot understand
and help you.

General request:  Please provide a short and meaningful email subject
and answer at the bottom of the previous mail to make our conversation
valuable for this mailing list.

Best,
Kai

[1] https://octave.org/doc/v5.1.0/XREFfminunc.html