nlminb.control

["Bradley, Ronan" <Ronan.Bradley@ILIM.COM>]

Hi,

I'm trying to optimise a function as follows using nlminb()

nlminb( start = x0,  f ,   control = nlminb.control( diff.g = 0.01) )

where f is a function of x, x taking the intial values x0, and x is a vector
of seven free parameters each ranging between 0 and 1.

f is a fairly large numerical evaluation with constant value over short distances and some degree of noise
over intermediate distances.  As such there is no _expression_
for the gradient or hessian so the gradient must be got by finite-difference. I expect that
the step size parameter for the finite difference gradient approximation would need to be of
the order 0.01 to overcome the noise/grainyness of the solution.

The problem is that supplying the arguments,

control = nlminb.control( diff.g = 0.01)

seems to have no effect of the step size used in the finite difference calculation.  The default
step size of  1.5e^-8  is being used, and is too small to measure a gradient on this funcion.
I'm using S-Plus 6.2.  

Does anyone know how to get this working?

regards

Ronan

Ronan Bradley
Irish Life Investment Managers
ronan.bradley@ilim.com

Irish Life Investment Managers winner of the Investment Management KPMG Financial Services Excellence Award 2007

 

**********************************************************************************

Irish Life Investment Managers Limited is authorised by the Financial Regulator under Section 10 of the Investment Intermediaries Act, 1995. Irish Life Investment Managers Limited Registered Office: Beresford Court, Beresford Place, Dublin 1. Registered in Ireland Number 116000

 

While Irish Life Investment Managers uses reasonable efforts to ensure that the information contained in this email is current, accurate and complete at the date of publication, no representations or warranties are made (express or implied) as to the reliability, accuracy or completeness of such information. Irish Life Investment Managers therefore cannot be held liable for any loss arising directly or indirectly from the use of, or any action taken in reliance on, any information contained in this email.

 

This material is for information only and does not constitute an offer or recommendation to buy or sell any investment, or subscribe to any investment management or advisory service. It is intended for the use of institutional and other professional investors. Past performance is not indicative of future results. The value of funds we manage may fall as well as rise.

 

This email and any files transmitted with it are confidential and intended solely for the use of the individual or entity to whom they are addressed. If you have received this email in error please notify the system manager. This footnote also confirms that this email message has been swept for the presence of computer viruses.

**********************************************************************************