Re: time-series smoothing

From: A.G.McDowell (mcdowella_at_mcdowella.demon.co.uk)
Date: 03/17/05

  • Next message: Everett M. Greene: "Re: Flexible Floating-Point Standard Proposal" 
    Date: Thu, 17 Mar 2005 19:09:58 +0000

    In article <4239ACF1.55B96907@ever.com>, Jake <wh@ever.com> writes
    >I have a problem of an apparent cross-disciplinary nature so I hope that
    >justifies the cross-posting.
    >
    (trimmed)
    >
    >2] Would it make sense to use Nelder-Mead (downhill simplex) method for
    >minimizing, even though this problem seems to be a constrained
    >optimization? There are three parameters, all constrained to positive
    >values and one constrained to integer values, so differentiation doesn't
    >seem to be an option. Is there a better derivative-free optimizer for
    >this problem? Is there a differentiation method that would work better,
    >even for a function that is only partially differentiable?
    >
    I note that the Nelder-Mean algorithm is mentioned in Numerical Recipes
    to need restarting in practice and is now known to sometimes fail to
    converge (google on Nelder simplex counter-example yields e.g.
    http://portal.acm.org/citation.cfm?id=589108: the body of the text
    requires a subscription, but the abstract is accessible to all). Torczon
    has provided a variety of convergence proofs for direct search
    algorithms, one of which looks like a variant of the Nelder-Mead
    algorithm: googling on Torczon Simplex yields http://www.cs.wm.edu/~va/r
    esearch/, which includes a paper "On the convergence of the
    multidirectional search algorithm".

    Are these methods destined to replace Nelder-Mead? Nice Theoretical
    curiosities? Something in between? 

    -- 
A.G.McDowell
  • Next message: Everett M. Greene: "Re: Flexible Floating-Point Standard Proposal"

    Relevant Pages

    • Re: time-series smoothing
      ... Is there a differentiation method that would work better, ... I note that the Nelder-Mean algorithm is mentioned in Numerical Recipes ... Are these methods destined to replace Nelder-Mead? ...
      (sci.stat.edu)
    • Re: time-series smoothing
      ... Is there a differentiation method that would work better, ... > I note that the Nelder-Mean algorithm is mentioned in Numerical Recipes ... one of which looks like a variant of the Nelder-Mead ... > multidirectional search algorithm". ...
      (sci.math.num-analysis)
    • Re: time-series smoothing
      ... Is there a differentiation method that would work better, ... > I note that the Nelder-Mean algorithm is mentioned in Numerical Recipes ... one of which looks like a variant of the Nelder-Mead ... > multidirectional search algorithm". ...
      (sci.stat.edu)
    • Re: time-series smoothing
      ... >I note that the Nelder-Mean algorithm is mentioned in Numerical Recipes ... >has provided a variety of convergence proofs for direct search ... one of which looks like a variant of the Nelder-Mead ...
      (sci.stat.edu)
    • Re: time-series smoothing
      ... >I note that the Nelder-Mean algorithm is mentioned in Numerical Recipes ... >has provided a variety of convergence proofs for direct search ... one of which looks like a variant of the Nelder-Mead ...
      (sci.math.num-analysis)