Nitpicker! [Two-Stage / GS Designs]

posted by ElMaestro  – Belgium?, 2017-08-22 17:59  – Posting: # 17735
Views: 22,886

Hi again,

» (...) I don’t get the point of initializing a large vector of values and later picking out the suitable one.

The secant method is terrible if the initial guess is wrong. Terrible as in, converges to plus/minus infinity and will give errors, so it isn't a matter of many iterations.
What the code above does: The n'th CV is probably going to be not so far from the (n-1)'th CV, and therefore it is a good guess. Therefore, if we only guess one CV reasonably (the 6th) then we get the corresponding 6'th critical CV, and from that point and onwards we can use the previously generated critical CV as the guess, i.e. when we want the 7'th value we use the 6th value as a guess, and so on. That's how these lines seem to work very fast.

You can also play around with regula falsi or whatever the heck Wikipedia calls it:

FindCritCV.rf=function(Pwr.T, GMR, Nps, Is.St2=1, toler=0.000001)
{
  CV1=0.0123
  CV2=8.7654
  Iter=0

  while (abs(CV1-CV2)>toler)
  {
   p1=Power.calc(Nps, GMR, CV1, Is.St2)-Pwr.T
   p2=Power.calc(Nps, GMR, CV2, Is.St2)-Pwr.T
   Iter=Iter+1

   CV3=CV2-p2*((CV2-CV1)/(p2-p1))


   ##cat("Noi=", Noi, "\n")
   tmp=Power.calc(Nps, GMR, CV3, Is.St2) -Pwr.T
   if (tmp<0) CV2=CV3
      else    CV1=CV3
   cat("CV1=", CV1, "CV2=", CV2, "dCV=", abs(CV2-CV1), "\n")
  }
  ##cat("Iterations RF:", Iter,"\n")
  return(0.5*(CV2+CV1) )
}



It is a little faster than bisection and does not require a good guess apart from the above. It is stable for power here since the power does not misbehave within any CV-interval of interest; there are no minima or maxima or things that can screw it badly up :-)

I could be wrong, but...
Best regards,
ElMaestro

Complete thread:

Activity
 Admin contact
20,144 posts in 4,248 threads, 1,385 registered users;
online 9 (0 registered, 9 guests [including 3 identified bots]).
Forum time (Europe/Vienna): 22:57 CET

Maturity is the capacity to endure uncertainty.    John Finley

The Bioequivalence and Bioavailability Forum is hosted by
BEBAC Ing. Helmut Schütz
HTML5