Re: [R] different forms of nls recommendations

2010-03-21 Thread Dieter Menne 


emorway wrote:
 
 
 So I wanted to try a different equation of the general form a/(b+c*x^d)
 
 US.nls.2-nls(US.final.values$ECe~(a/(b+c*US.final.values$WTD^d)),data=US.final.values,start=list(a=100.81,b=73.7299,c=0.0565,d=-6.043),trace=TRUE,algorithm=port)
 
 but that ended with Convergence failure: false convergence (8).  I tried
 relaxing the convergence 
 
 

You want 4 parameters from a set of data that has a very large variance. Try
to fix d and c to a reasonable constant; it probably will converge.

From looking at your data and the variances, I would suggest to
log-transform first. If you feel bad about it, you can always fit the
original data later after you have found a good model.

Dieter


-- 
View this message in context: 
http://n4.nabble.com/different-forms-of-nls-recommendations-tp1676330p1676547.html
Sent from the R help mailing list archive at Nabble.com.

__
R-help@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.

Re: [R] different forms of nls recommendations

2010-03-21 Thread Bernardo Rangel Tura 
On Sat, 2010-03-20 at 14:55 -0800, emorway wrote:
 Hello, 
 
 Using this data:
 http://n4.nabble.com/file/n1676330/US_Final_Values.txt US_Final_Values.txt 
 
 and the following code i got the image at the end of this message:
 
 US.final.values-read.table(c:/tmp/US_Final_Values.txt,header=T,sep= )
 US.nls.1-nls(US.final.values$ECe~a*US.final.values$WTD^b+c,data=US.final.values,start=list(a=2.75,b=-0.95,c=0.731),trace=TRUE)
 f.US1-function(x){coef(US.nls.1)[a]*x^coef(US.nls.1)[b]+coef(US.nls.1)[c]}
 xvals.US1-seq(min(US.final.values$WTD),max(US.final.values$WTD),length.out=75)
 yvals.US1-f.US1(xvals.US1)
 Rsq.nls.1-sum((predict(US.nls.1)-mean(US.final.values$ECe))^2/sum((US.final.values$ECe-mean(US.final.values$ECe))^2))
 plot(US.final.values$WTD,US.final.values$ECe,col=red,pch=19,cex=.75)
 lines(xvals.US1,yvals.US1,col=blue)
 
 but the r^2 wasn't so hot.  
 Rsq.nls.1
 [1] 0.2377306
 
 So I wanted to try a different equation of the general form a/(b+c*x^d)
 
 US.nls.2-nls(US.final.values$ECe~(a/(b+c*US.final.values$WTD^d)),data=US.final.values,start=list(a=100.81,b=73.7299,c=0.0565,d=-6.043),trace=TRUE,algorithm=port)
 
 but that ended with Convergence failure: false convergence (8).  I tried


Hi emorway,

Do you have 657 obs and 4 parameters to fit.
In my opinion you have few obs...
I think do you  fit in steps:

US.nls.2-nls(ECe~(a/(b + c *
WTD^d)),data=US.final.values,start=list(a=100.81,b=73.7299,c=0.0565,d=-6.043),trace=TRUE,algorithm=port)
temp_nls1 - nls(ECe~(100/(73 + .05 *
WTD^d)),data=US.final.values,start=list(d=-6.043),trace=TRUE,algorithm=port)
temp_nls2 - nls(ECe~(100/(73 + .05 *
WTD^d)),data=US.final.values,start=list(d=-1.01613),trace=TRUE,algorithm=port)
temp_nls3 - nls(ECe~(100/(73 + c *
WTD^(-1.01613))),data=US.final.values,start=list(c=0.05),trace=TRUE,algorithm=port)
temp_nls4 - nls(ECe~(100/(73 + c *
WTD^(-1.01613))),data=US.final.values,start=list(c=-14.7127),trace=TRUE,algorithm=port)
temp_nls5 - nls(ECe~(100/(b-14.7127 *
WTD^(-1.01613))),data=US.final.values,start=list(b=73),trace=TRUE,algorithm=port)
temp_nls6 - nls(ECe~(100/(b-14.7127 *
WTD^(-1.01613))),data=US.final.values,start=list(b=70.4936),trace=TRUE,algorithm=port)
temp_nls7 - nls(ECe~(a/(70.4936-14.7127 *
WTD^(-1.01613))),data=US.final.values,start=list(a=100),trace=TRUE,algorithm=port)
  0: 2243.9898:  100.000
  1: 2122.8218:  106.219
  2: 1359.8819:  187.530
  3: 1359.8819:  187.530



-- 
Bernardo Rangel Tura, M.D,MPH,Ph.D
National Institute of Cardiology
Brazil

__
R-help@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.

[R] different forms of nls recommendations

2010-03-20 Thread emorway 

Hello, 

Using this data:
http://n4.nabble.com/file/n1676330/US_Final_Values.txt US_Final_Values.txt 

and the following code i got the image at the end of this message:

US.final.values-read.table(c:/tmp/US_Final_Values.txt,header=T,sep= )
US.nls.1-nls(US.final.values$ECe~a*US.final.values$WTD^b+c,data=US.final.values,start=list(a=2.75,b=-0.95,c=0.731),trace=TRUE)
f.US1-function(x){coef(US.nls.1)[a]*x^coef(US.nls.1)[b]+coef(US.nls.1)[c]}
xvals.US1-seq(min(US.final.values$WTD),max(US.final.values$WTD),length.out=75)
yvals.US1-f.US1(xvals.US1)
Rsq.nls.1-sum((predict(US.nls.1)-mean(US.final.values$ECe))^2/sum((US.final.values$ECe-mean(US.final.values$ECe))^2))
plot(US.final.values$WTD,US.final.values$ECe,col=red,pch=19,cex=.75)
lines(xvals.US1,yvals.US1,col=blue)

but the r^2 wasn't so hot.  
Rsq.nls.1
[1] 0.2377306

So I wanted to try a different equation of the general form a/(b+c*x^d)

US.nls.2-nls(US.final.values$ECe~(a/(b+c*US.final.values$WTD^d)),data=US.final.values,start=list(a=100.81,b=73.7299,c=0.0565,d=-6.043),trace=TRUE,algorithm=port)

but that ended with Convergence failure: false convergence (8).  I tried
relaxing the convergence criteria to no avail.  Assuming the form of the
equation I'm trying to use is the problem, I've been unable to track down a
source that shows the shapes of various non-linear equations that I might be
able to try as alternatives.  Any suggestions?

http://n4.nabble.com/file/n1676330/nls_image.jpg 

-- 
View this message in context: 
http://n4.nabble.com/different-forms-of-nls-recommendations-tp1676330p1676330.html
Sent from the R help mailing list archive at Nabble.com.

__
R-help@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.