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 No??l CdeC _________________________________________________________________ De : Tim Triche, Jr. [tim.triche at gmail.com] Date d'envoi : 22 f??vrier 2012 16:11 ?? : Simon No??l 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 No??l <[1]simon.noel.2 at ulaval.ca> Hello, Anyone can help? Simon No????l CdeC _________________________________________________________________ De : Simon No????l Date d'envoi : 9 f????vrier 2012 13:37 ????? : [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 No????l 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