[R] nls() and lines()
Peter Ehlers
ehlers at ucalgary.ca
Mon Jul 18 15:52:09 CEST 2011
On 2011-07-18 06:38, Steven Ranney wrote:
> Provided, of course, that I alter the lines for different data sets
> and data frames, the code to plot a line derived from nls() onto a
> plot works with no problems.
>
> Here's an example:
>
> Year NOP
> 2002 6
> 2003 8
> 2004 11
> 2005 19
> 2006 26
> 2007 25
>
> mod1<- nls(NOP~alpha*exp(beta*Year), data=aic,
> start=list(alpha=1e-278, beta=0.3205), trace=T,
> nls.control(maxiter=30000, minFactor=0.000005))
> plot(NOP~Year, data=aic, pch=19, ylab="Number of papers")
> mod1a=seq(2002, 2007, by=.0001)
> lines(mod1a, predict(mod1, list(Year = mod1a)))
Yes, but here's what you posted as your 'last line of code':
lines(modb, predict(nls.2009, lines(as.numeric(x)=modb)))
Perhaps just a typo: lines -> list???
In any case, the newdata in predict should have a variable 'Year'.
Peter Ehlers
>
> I've been using this code for several years not to get models from
> nls() onto plot and I've never had an issue with it until the dataset
> I referenced in my initial email.
>
> Thanks for your assistance.
>
> SR
>
> Steven H. Ranney
>
> http://www.steven-ranney.com
> http://stevenranney.blogspot.com
>
>
> On Mon, Jul 18, 2011 at 2:55 AM, Peter Ehlers<ehlers at ucalgary.ca> wrote:
>> On 2011-07-17 17:37, Steven Ranney wrote:
>>>
>>> All -
>>>
>>> I'm having an issue with trying to plot a model derived from nls()
>>> onto a simple plot. I have included a sample data set and the code
>>> that I've been using.
>>>
>>> year month day date location mileage cost gallon cpg
>>> mpg x
>>> 2009 1 4 1/4/2009 BZN 124585 19.39 14.37 1.349339
>>> 10.71677 2009-01-04
>>> 2009 1 15 1/15/2009 BZN 124888 23.2 16.12 1.439206
>>> 18.79653 2009-01-15
>>> 2009 1 27 1/27/2009 BZN 125133 21.51 14.35 1.498955
>>> 17.07317 2009-01-27
>>> 2009 2 16 2/16/2009 BZN 125429 27.96 15.54 1.799228
>>> 19.04762 2009-02-16
>>> 2009 2 27 2/27/2009 BZN 125702 26.82 14.27 1.879467
>>> 19.13104 2009-02-27
>>> 2009 3 19 3/19/2009 BZN 125941 24.38 12.91 1.888459
>>> 18.51278 2009-03-19
>>> 2009 4 9 4/9/2009 BZN 126260 32.59 16.30 1.999387
>>> 19.57055 2009-04-09
>>> 2009 4 28 4/28/2009 BZN 126587 34.58 16.79 2.059559
>>> 19.47588 2009-04-28
>>> 2009 5 17 5/17/2009 BZN 126925 35.78 16.57 2.159324
>>> 20.39831 2009-05-17
>>> 2009 5 27 5/27/2009 BZN 127240 35.57 15.01 2.369753
>>> 20.98601 2009-05-27
>>> 2009 6 7 6/7/2009 BZN 127590 40.99 16.60 2.469277
>>> 21.08434 2009-06-07
>>> 2009 6 21 6/21/2009 BZN 127910 41.52 15.64 2.654731
>>> 20.46036 2009-06-21
>>> 2009 7 21 7/21/2009 BZN 128264 43.37 16.67 2.601680
>>> 21.23575 2009-07-21
>>> 2009 8 11 8/11/2009 BZN 128618 42.68 16.42 2.599269
>>> 21.55907 2009-08-11
>>> 2009 8 27 8/27/2009 BZN 128947 43.12 16.60 2.597590
>>> 19.81928 2009-08-27
>>> 2009 9 21 9/21/2009 BZN 129255 40.44 15.56 2.598972
>>> 19.79434 2009-09-21
>>> 2009 10 1 10/1/2009 BZN 129541 38.55 14.83 2.599461
>>> 19.28523 2009-10-01
>>> 2009 10 11 10/11/2009 BZN 129806 36.65 14.10 2.599291
>>> 18.79433 2009-10-11
>>> 2009 10 22 10/22/2009 BZN 130027 30.18 11.61 2.599483
>>> 19.03531 2009-10-22
>>> 2009 11 9 11/9/2009 BZN 130358 43.19 16.62 2.598676
>>> 19.91576 2009-11-09
>>> 2009 11 22 11/22/2009 BZN 130631 39.23 15.09 2.599735
>>> 18.09145 2009-11-22
>>> 2009 12 5 12/5/2009 BZN 130950 44.43 17.09 2.599766
>>> 18.66589 2009-12-05
>>> 2009 12 30 12/30/2009 BZN 131239 42.14 16.70 2.523353
>>> 17.30539 2009-12-30
>>>
>>> After converting my dates into R-usable dates:
>>>
>>> #convert my dates to R-usable dates
>>> x<- strptime(date, format="%m/%d/%Y")
>>> x
>>> mileage<- cbind(mileage, x)
>>>
>>> I plot the data and model mpg as a function of date. In the nls()
>>> statement, I convert x back to a numeric value so that I can conduct
>>> the regression:
>>>
>>> plot(mpg~x, data=mileage[year==2009,], ylab="Miles per gallon",
>>> xlab="2009", yaxs="i", ylim=c(10,30))
>>> nls.2009<-
>>> nls(mpg~(alpha*(as.numeric(x)^2))+(bravo*as.numeric(x))+(charlie),
>>> data=mileage[year==2009,], start=list(alpha=-2e-14, bravo=5e-5,
>>> charlie=-31407),
>>> trace=T, na.action=na.omit,
>>> nls.control(minFactor=0.000000000000000000001))
>>> plot(mpg~x, data=mileage[year==2009,])
>>> modb=seq(min(as.numeric(x)), max(as.numeric(x)), by=10000)
>>> lines(modb, predict(nls.2009, lines(as.numeric(x)=modb)))
>>>
>>> Unfortunately, when I run the final line of this code, I get the
>>> following:
>>>
>>> Error: unexpected '=' in " lines(modb, predict(nls.2009,
>>> lines(as.numeric(x)="
>>>
>>> In other similar analyses, I've been able to plot an nls() model using
>>> this exact code--altered of course according to information--but here
>>> I'm at a loss. I'm certain it has something to do with the
>>> lines(...as.numeric(x)) value I'm trying to plot, but I can't figure
>>> out what I'm doing wrong.
>>
>> That last line of code doesn't look right to me. The arguments
>> that you need to supply to predict() are 'object' and 'newdata',
>> where 'newdata' must have the appropriate form. Unless you have
>> your own function lines(), I don't think that lines(as.numeric(x)=modb)
>> would qualify as newdata.
>>
>> It's usually a bad idea to shove too much stuff into a single command
>> and a good idea to use str() often.
>>
>> This 'exact' code worked in the past?
>>
>> Peter Ehlers
>>
>>>
>>> The model is fine, but it's the plotting of the model that escapes me.
>>>
>>> I'm running R version 2.12.1 on a Windows 7 machine.
>>>
>>> Thanks for your help -
>>>
>>> Steven H. Ranney
>>>
>>> http://stevenranney.blogspost.com
>>> http://www.steven-ranney.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.
>>
>>
More information about the R-help
mailing list