[Rd] stats package, optim and other optimization methods: more parameters?

Mike Preuss mike.preuss at cs.uni-dortmund.de
Wed Nov 14 23:19:57 CET 2007


Dear Brian,

thanks for the hint. But my question is more
concerned with the technical procedure. Of
course, I can change the C code locally. And
then use a locally changed package. But then
nobody else beside myself can use it, right?
If changes were made in the official code
(and it's not really changing the code, only
adding some parameters to the interface),
my experiments could be repeated by others.

And if the official code is then changed
again, I'd have to update the local code
every time. Or is that unlikely? Or am I
missing some important point here?

Best, Mike



Prof Brian Ripley wrote:
> On Wed, 14 Nov 2007, Mike Preuss wrote:
> 
>> Dear R developers,
>>
>> concerning the stats package,
>> is there a way to access more of the 'internal'
>> parameters of the optim() methods (e.g. BFGS),
>> without me rewriting optim code and using it as
>> an internal, patched stats package? It may not
>> sound useful to you, but for me it is important
>> to be able to change the parameters as I want
>> to apply parameter tuning methods. Currently,
>> I'm doing so only on the very limited parameter
>> set available via the control parameter, but
>> e.g. for the CG and BFGS methods, this is practically
>> empty. But there are internally set parameters
>> (e.g. for the line searches), just they are not
>> accessible from the outside.
> 
> Why don't you read the C code to find out?
> 
> You probably also need to read the reference to understand the choices
> made, and why the author didn't feel it useful to make them accessible.
>



More information about the R-devel mailing list