From: Dimitri Liakhovitski <dimitri.liakhovitski_at_gmail.com>

Date: Mon, 07 Mar 2011 17:10:55 -0500

(pred)

DV<-c(0.55351297,0.27616943,0.58134926,0.33887159,0.03092546,0.14928061,

*(DV)
*

Date: Mon, 07 Mar 2011 17:10:55 -0500

Hello!

I have 2 variables - predictor "pred" and response variable "DV":

5728.821, 2054514.244, 301954.819, 773955.355, 735497.506, 347355.976, 1678175.153, 133082.395, 591326.289, 30866.182, 27235.846, 118372.342, 71590.969, 84813.299, 366146.153, 1391725.205, 763199.746, 1216661.202,263878.157, 930832.769, 261270.130, 589303.561, 455137.946, 954655.201, 873434.054)

(pred)

DV<-c(0.55351297,0.27616943,0.58134926,0.33887159,0.03092546,0.14928061,

0.11836759,0.01719463,0.03258188,1.81205587,2.86657699,2.49491195, 3.09727230,1.95648776,2.28106268,1.78978179,1.74003678,1.22520393, 0.54245878,0.41483039,1.08731493,2.19581289,1.60516129,1.30723431, 1.41822649,1.31530539,2.02406576,1.22211412,0.52055790,0.12975522, 0.01416903,0.61043485,0.44141748,0.64327070,0.53607039,0.32603820, 1.77261016,0.42035756,0.37853917,0.12342486,0.06607710,0.02383682, 0.08421590,0.09255332,0.23644909,1.67921092,1.26864432,1.38654574, 1.29833020,1.76873555,0.93363677,1.01857658,0.81359775,2.14758239,2.41583852)

Both "pred" and "DV" above are time series (observed across 55 months). The relationship between them is pre-specified. In this relationship, the (predicted) "DV" at time t is a specific function of itself at time t-1, of "pred" at time t, and of 2 scalars - a and b. I have to find optimal a and b that would ensure the best fit between the observed DV and the predicted DV. Below is the function I have to optimize:

DV_pred[i] <- 1 - ( (1 - DV_pred[i-1] * a) / (exp(pred[i] * b)) )
}

DV_pred[1]<-DV[1]

correl <- cor(DV,DV_pred)

return(correl)

}

a has to be between 0.001 and 0.75

b has to be positive.

I apologize if it's a simple question for statisticians but I am not a
mathematician/statistician by training. I didn't think optim would
work here. The only thing I could think of is genetic optimization,
for example in rgenoud below. However, I don't think I could use it
for 2 reasons: (1) Solutions do not seem stable and depend on starting
values, set.seed, and domains chosen; (2) It takes too long - I have
to do the task I described above about 1,200 times and it would take
me forever.

Could someone maybe provide a pointer: what would be a faster/more
efficient optimization approach for such a function as mine? Thanks a
lot!

library(rgenoud)

genoud.opt<-genoud(my.function, nvars=2, max=TRUE, pop.size=10000,
hard.generation.limit=TRUE,

max.generations=20, wait.generations=5, starting.values=c(0.5,1), Domains=matrix(c(0.001,0.75,0.00000001,2),ncol=2,byrow=T),boundary.enforcement=2)

-- Dimitri Liakhovitski Ninah Consulting www.ninah.com ______________________________________________ R-help_at_r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.Received on Mon 07 Mar 2011 - 22:36:11 GMT

Archive maintained by Robert King, hosted by
the discipline of
statistics at the
University of Newcastle,
Australia.

Archive generated by hypermail 2.2.0, at Tue 08 Mar 2011 - 03:30:19 GMT.

*
Mailing list information is available at https://stat.ethz.ch/mailman/listinfo/r-help.
Please read the posting
guide before posting to the list.
*