[BioC] RE : RE : Optimisation

[Simon Noël]

   I tought that big formula will scare some people...  But I will try...  So I
   have that function :

   variance <- function(w, result)
   {
    w22 = w[1]
    w21 = w[2]
    w12 = w[3]

    w11 = 1 - w12 - w21 - w22

    (  ((1/2  *  w22)^2 * result[1]) + ((w21)^2 * result[2]) + ((w12)^2 *
   result[3]) + ((2 * w11)^2 * result[4]) + (w22 * w21 * result[5]) + (w22
   * w12  *  result[6])  + (2 * w22 * w11 * result[7]) + (2 * w21 * w12 *
   result[8]) + (4 * w21 * w11 * result[9]) + (4 * w12 * w11 * result[10]) )
   }

   I want to find the combinaison of value for my W that will give the best
   minimisation of my equation.  Each w must be between 1 and 0.  Sum of all
   om them must = 1.

   Until now, everything seem to be ok.  I use constrOptim like that :

    Amat <- matrix(c(1,-1,0,0,0,0,0,0,1,-1,0,0,0,0,0,0,1,-1),6,3)
    Bvec <- c(0,-1,0,-1,0,-1)

    varLamda   =  constrOptim(theta=w,  f=variance,  grad=NULL,  ui=Amat,
   ci=Bvec, result = result)

   result is defined as

   variance1 <- function(i22, i21, i12, i11)
   {
    result = vector(length = 10)

    result[1] = VLogRc(i22)
    result[2] = VLogRc(i21)
    result[3] = VLogRc(i12)
    result[4] = VLogRc(i11)
    result[5] = covlogi(i22, i21)
    result[6] = covlogi(i22, i12)
    result[7] = covlogi(i22, i11)
    result[8] = covlogi(i21, i12)
    result[9] = covlogi(i21, i11)
    result[10] = covlogi(i12, i11)

    result
   }

   and
   covlogi <- function(i1, i2)
   {
    i = vector(length = length(i1))
    for(j in 1:length(i1))
    {
     if(names(i1)[j] == names(i2)[j])
     {
      i[j] = 1/i1[j]
     }else
     {
      i[j] = 0
     }
    }
    sum(i)
   }


   VLogRc <- function(i)
   {
    ( (1/i[1]) + (1/i[2]) + (1/i[3]) + (1/i[4]) )
   }

   my i are
    #i11 = (q11 * q00) / (q10 * q01)
    #i12 = (q12 * q00) / (q10 * q02)
    #i21 = (q21 * q00) / (q20 * q01)
    #i22 = (q22 * q00) / (q20 * q02)
    i11[1] = Q[5]
    i11[2] = Q[1]
    i11[3] = Q[4]
    i11[4] = Q[2]
    i12[1] = Q[6]
    i12[2] = Q[1]
    i12[3] = Q[4]
    i12[4] = Q[3]
    i21[1] = Q[8]
    i21[2] = Q[1]
    i21[3] = Q[7]
    i21[4] = Q[2]
    i22[1] = Q[9]
    i22[2] = Q[1]
    i22[3] = Q[7]
    i22[4] = Q[3]

   and each of my Q are the number of subkect with a particular combinaison of
   SNPs.

   It may happen that  one  of  the  Q  =  0.  Like  if q22  = 0, we will
   have with VLogRc =  ( (1/i[1]) + (1/i[2]) + (1/i[3]) + (1/i[4]) ) where i[1]
   = 0 so the result is INF and that's really bad.

   One way  to avoid that is  to write  a  new formula without w22..  And
   an other when a problem come for w21 and an other for w11 and one other
   for w12 and an other for w22 and w21, etc.  There also a lot of if/else
   for each  possible  q  combinaison with  0  value.  So it's not a goot
   solution...  Something  simple is  if  I  have an INF, and w = 0, ther
   the problem will be solved...

   If we take the exemple of q22 =0, w22 must = 0 in orther that the program
   work...  But how to corce that with constrOptim?  Is there a better fonction
   to solve my problem?

   Simon Noël
   CdeC
     _________________________________________________________________

   De : Tim Triche, Jr. [tim.triche at gmail.com]
   Date d'envoi : 22 février 2012 16:11
   À : Simon Noël
   Cc : bioconductor at r-project.org
   Objet : Re: [BioC] RE : Optimisation
   you probably want to post this on r-help or perhaps look at the optimx
   package.
   you might also want to include a piece of code showing what precisely it is
   that you are trying to do.
   this is more likely to get a response than the fairly vague message below.
   hope it helps,
   --t
   2012/2/22 Simon Noël <[1]simon.noel.2 at ulaval.ca>

       Hello,
       Anyone can help?
       Simon Noël
       CdeC
         _________________________________________________________________
       De : Simon Noël
       Date d'envoi : 9 février 2012 13:37
       À : [2]bioconductor at r-project.org
       Objet : Optimisation

     Hi every one,
     I have a big formula with some parameter that I want to minimise with some
     condition like sum  of all of them must be one, in some condition, one
     of them must equal 0, etc.
     I am trying to solve that with constrOptim but I got some broblem...  Like
     how  to have a parametter equal to 0  Is there any package who is more
     user friendly that you can suggest to me?
     Simon Noël
     CdeC
   _______________________________________________
   Bioconductor mailing list
   [3]Bioconductor at r-project.org
   [4]https://stat.ethz.ch/mailman/listinfo/bioconductor
   Search the archives:
   [5]http://news.gmane.org/gmane.science.biology.informatics.conductor

   --
   A model is a lie that helps you see the truth.
   [6]Howard Skipper

References

   1. mailto:simon.noel.2 at ulaval.ca
   2. mailto:bioconductor at r-project.org
   3. mailto:Bioconductor at r-project.org
   4. https://stat.ethz.ch/mailman/listinfo/bioconductor
   5. http://news.gmane.org/gmane.science.biology.informatics.conductor
   6. http://cancerres.aacrjournals.org/content/31/9/1173.full.pdf