Do YOU know an equation for splines (ns)?

[Ranae]

Hi,

I am looking at the change in N concentration in plant roots over 4 time points and I have fit a spline to the data using ns and lme:

fit10 <- lme( N~ns(day, 3), data = rcn10G)

I may want to adjust the model a little bit, but for now, let's assume it's good.  I get output for the fixed effects:

Fixed: N ~ ns(day, 3)
(Intercept) ns(day, 3)1 ns(day, 3)2 ns(day, 3)3
 1.15676524  0.14509171  0.04459627  0.09334428
 
and coefficients for each experimental unit in my experiment:

   (Intercept) ns(day, 3)1 ns(day, 3)2 ns(day, 3)3
24    1.050360 -0.42666159 -0.56290877 -0.10714407
13    1.104464 -0.30825350 -0.53311653 -0.05558150
31    1.147878 -0.14548512 -0.78673906 -0.07231781
46    1.177781 -0.22278380 -0.80278177 -0.02321460
15    1.144215 -0.04484519 -0.06084798  0.07633663
32    1.213007  0.00741061  0.03896933  0.15325849
23    1.274615  0.16477514  0.00872224  0.23128320
41    1.215626  0.57050767  0.11415467  0.10608867
43    1.134203  0.48070741  0.72112899  0.18108193
12    1.091422  0.39563632  1.01521528  0.22597459
21    1.100631  0.44589314  0.98526322  0.23535739
35    1.226980  0.82419937  0.39809568  0.16900841

NOW, I want to write a spline function where I can incorporate these coefficients to get the predicted N concentration value for each day.  However, I am having trouble finding the right spline equation, since there are many forms on the internets.  

I know it won't be a simple one, but can some one direct me to the equation that would be best to use for ns?

Thanks a lot,

Ranae

[Bert Gunter]

Does
?predict.ns
not do what you want without having to explicitly manipulate the spline basis?

-- Bert

On Tue, Jun 5, 2012 at 1:56 PM, Ranae <[hidden email]> wrote:

> Hi,
>
> I am looking at the change in N concentration in plant roots over 4 time
> points and I have fit a spline to the data using ns and lme:
>
> fit10 <- lme( N~ns(day, 3), data = rcn10G)
>
> I may want to adjust the model a little bit, but for now, let's assume it's
> good.  I get output for the fixed effects:
>
> Fixed: N ~ ns(day, 3)
> (Intercept) ns(day, 3)1 ns(day, 3)2 ns(day, 3)3
>  1.15676524  0.14509171  0.04459627  0.09334428
>
> and coefficients for each experimental unit in my experiment:
>
>   (Intercept) ns(day, 3)1 ns(day, 3)2 ns(day, 3)3
> 24    1.050360 -0.42666159 -0.56290877 -0.10714407
> 13    1.104464 -0.30825350 -0.53311653 -0.05558150
> 31    1.147878 -0.14548512 -0.78673906 -0.07231781
> 46    1.177781 -0.22278380 -0.80278177 -0.02321460
> 15    1.144215 -0.04484519 -0.06084798  0.07633663
> 32    1.213007  0.00741061  0.03896933  0.15325849
> 23    1.274615  0.16477514  0.00872224  0.23128320
> 41    1.215626  0.57050767  0.11415467  0.10608867
> 43    1.134203  0.48070741  0.72112899  0.18108193
> 12    1.091422  0.39563632  1.01521528  0.22597459
> 21    1.100631  0.44589314  0.98526322  0.23535739
> 35    1.226980  0.82419937  0.39809568  0.16900841
>
> NOW, I want to write a spline function where I can incorporate these
> coefficients to get the predicted N concentration value for each day.
> However, I am having trouble finding the right spline equation, since there
> are many forms on the internets.
>
> I know it won't be a simple one, but can some one direct me to the equation
> that would be best to use for ns?
>
> Thanks a lot,
>
> Ranae
>
>
> --
> View this message in context: http://r.789695.n4.nabble.com/Do-YOU-know-an-equation-for-splines-ns-tp4632440.html
> Sent from the R help mailing list archive at Nabble.com.
>
> ______________________________________________
> [hidden email] 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.

--

Bert Gunter
Genentech Nonclinical Biostatistics

Internal Contact Info:
Phone: [phone]
Website:
http://pharmadevelopment.roche.com/index/pdb/pdb-functional-groups/pdb-biostatistics/pdb-ncb-home.htm

______________________________________________
[hidden email] 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.

[Ranae]

I have not been able to get "predict" (or most functions) to run well with grouped data in nlme.  I may not have it coded right, but this is what it looks like:

spline.txt

library(nlme)
library(splines)

rootCN<-read.table("spline.txt", header=TRUE)
rootCN$plotF<-as.factor(rootCN$plot)

rcn10G<-groupedData(N ~ day | plotF, data=rcn10)

fit10 <- lme( N~ns(day, 3), data = rcn10G)

plot(augPred(fit10))

num<- seq(88,300, len=200)
lines(num, predict(fit10, data.frame(day=num)))

-Ranae


Does
 ?predict.ns
 not do what you want without having to explicitly manipulate the spline basis?

-- Bert

On Tue, Jun 5, 2012 at 1:56 PM, Ranae <[hidden email]> wrote:

> Hi,
>
> I am looking at the change in N concentration in plant roots over 4 time
> points and I have fit a spline to the data using ns and lme:
>
> fit10 <- lme( N~ns(day, 3), data = rcn10G)
>
> I may want to adjust the model a little bit, but for now, let's assume it's
> good.  I get output for the fixed effects:
>
> Fixed: N ~ ns(day, 3)
> (Intercept) ns(day, 3)1 ns(day, 3)2 ns(day, 3)3
>  1.15676524  0.14509171  0.04459627  0.09334428
>
> and coefficients for each experimental unit in my experiment:
>
>   (Intercept) ns(day, 3)1 ns(day, 3)2 ns(day, 3)3
> 24    1.050360 -0.42666159 -0.56290877 -0.10714407
> 13    1.104464 -0.30825350 -0.53311653 -0.05558150
> 31    1.147878 -0.14548512 -0.78673906 -0.07231781
> 46    1.177781 -0.22278380 -0.80278177 -0.02321460
> 15    1.144215 -0.04484519 -0.06084798  0.07633663
> 32    1.213007  0.00741061  0.03896933  0.15325849
> 23    1.274615  0.16477514  0.00872224  0.23128320
> 41    1.215626  0.57050767  0.11415467  0.10608867
> 43    1.134203  0.48070741  0.72112899  0.18108193
> 12    1.091422  0.39563632  1.01521528  0.22597459
> 21    1.100631  0.44589314  0.98526322  0.23535739
> 35    1.226980  0.82419937  0.39809568  0.16900841
>
> NOW, I want to write a spline function where I can incorporate these
> coefficients to get the predicted N concentration value for each day.
> However, I am having trouble finding the right spline equation, since there
> are many forms on the internets.
>
> I know it won't be a simple one, but can some one direct me to the equation
> that would be best to use for ns?
>
> Thanks a lot,
>
> Ranae

[Bert Gunter]

Ah ...
Iirc believe the problem is that you need to explicitly generate the
spline basis and then the predicted values via predict.ns and feed
that to predict.lme; i.e.

splineBas <- with(rcn10,ns(day,3))

newvals <- data.frame( predict(splineBas, num))

## then once you've fitted your model:
lines(num, predict(fit10, newvals))


I have NOT checked this though, so please post back to me and the list
whether this works.

-- Bert




On Wed, Jun 6, 2012 at 10:38 AM, Ranae <[hidden email]> wrote:

> I have not been able to get "predict" (or most functions) to run well with
> grouped data in nlme.  I may not have it coded right, but this is what it
> looks like:
>
> http://r.789695.n4.nabble.com/file/n4632567/spline.txt spline.txt
>
> library(nlme)
> library(splines)
>
> rootCN<-read.table("spline.txt", header=TRUE)
> rootCN$plotF<-as.factor(rootCN$plot)
>
> rcn10G<-groupedData(N ~ day | plotF, data=rcn10)
>
> fit10 <- lme( N~ns(day, 3), data = rcn10G)
>
> plot(augPred(fit10))
>
> num<- seq(88,300, len=200)
> lines(num, predict(fit10, data.frame(day=num)))
>
> -Ranae
>
>
> Does
>  ?predict.ns
>  not do what you want without having to explicitly manipulate the spline
> basis?
>
> -- Bert
>
> On Tue, Jun 5, 2012 at 1:56 PM, Ranae <[hidden email]> wrote:
>
>> Hi,
>>
>> I am looking at the change in N concentration in plant roots over 4 time
>> points and I have fit a spline to the data using ns and lme:
>>
>> fit10 <- lme( N~ns(day, 3), data = rcn10G)
>>
>> I may want to adjust the model a little bit, but for now, let's assume
>> it's
>> good.  I get output for the fixed effects:
>>
>> Fixed: N ~ ns(day, 3)
>> (Intercept) ns(day, 3)1 ns(day, 3)2 ns(day, 3)3
>>  1.15676524  0.14509171  0.04459627  0.09334428
>>
>> and coefficients for each experimental unit in my experiment:
>>
>>   (Intercept) ns(day, 3)1 ns(day, 3)2 ns(day, 3)3
>> 24    1.050360 -0.42666159 -0.56290877 -0.10714407
>> 13    1.104464 -0.30825350 -0.53311653 -0.05558150
>> 31    1.147878 -0.14548512 -0.78673906 -0.07231781
>> 46    1.177781 -0.22278380 -0.80278177 -0.02321460
>> 15    1.144215 -0.04484519 -0.06084798  0.07633663
>> 32    1.213007  0.00741061  0.03896933  0.15325849
>> 23    1.274615  0.16477514  0.00872224  0.23128320
>> 41    1.215626  0.57050767  0.11415467  0.10608867
>> 43    1.134203  0.48070741  0.72112899  0.18108193
>> 12    1.091422  0.39563632  1.01521528  0.22597459
>> 21    1.100631  0.44589314  0.98526322  0.23535739
>> 35    1.226980  0.82419937  0.39809568  0.16900841
>>
>> NOW, I want to write a spline function where I can incorporate these
>> coefficients to get the predicted N concentration value for each day.
>> However, I am having trouble finding the right spline equation, since
>> there
>> are many forms on the internets.
>>
>> I know it won't be a simple one, but can some one direct me to the
>> equation
>> that would be best to use for ns?
>>
>> Thanks a lot,
>>
>> Ranae
>
> --
> View this message in context: http://r.789695.n4.nabble.com/Do-YOU-know-an-equation-for-splines-ns-tp4632440p4632567.html
> Sent from the R help mailing list archive at Nabble.com.
>
> ______________________________________________
> [hidden email] 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.

--

Bert Gunter
Genentech Nonclinical Biostatistics

Internal Contact Info:
Phone: [phone]
Website:
http://pharmadevelopment.roche.com/index/pdb/pdb-functional-groups/pdb-biostatistics/pdb-ncb-home.htm

______________________________________________
[hidden email] 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.

[William Dunlap]

Do you have to include the grouping variable, plotF, in your newdata
argument?  E.g., after fitting the model with
  rcn10G<-groupedData(N ~ day | plotF, data=rcn10)
  fit10 <- lme( N~ns(day, 3), data = rcn10G)
try checking the predictions when you've include plotF in newdata:
  par(mfrow=c(2,1))
  plot(N ~ day, subset=plotF=="12", data=rcn10G)
  points(num, predict(fit10, data.frame(day=num, plotF=rep("12", length(num)))), pch=".", col="red")
   
  plot(N ~ day, subset=plotF=="43", data=rcn10G)
  points(num, predict(fit10, data.frame(day=num, plotF=rep("43", length(num)))), pch=".", col="red")

I am no expert on the lme and groupedData, but the general rule is that all variables involved
in the model, except the response, must be given to predict.

Bill Dunlap
Spotfire, TIBCO Software
wdunlap tibco.com

> -----Original Message-----
> From: [hidden email] [mailto:[hidden email]] On Behalf
> Of Ranae
> Sent: Wednesday, June 06, 2012 10:39 AM
> To: [hidden email]
> Subject: Re: [R] Do YOU know an equation for splines (ns)?
>
> I have not been able to get "predict" (or most functions) to run well with
> grouped data in nlme.  I may not have it coded right, but this is what it
> looks like:
>
> http://r.789695.n4.nabble.com/file/n4632567/spline.txt spline.txt
>
> library(nlme)
> library(splines)
>
> rootCN<-read.table("spline.txt", header=TRUE)
> rootCN$plotF<-as.factor(rootCN$plot)
>
> rcn10G<-groupedData(N ~ day | plotF, data=rcn10)
>
> fit10 <- lme( N~ns(day, 3), data = rcn10G)
>
> plot(augPred(fit10))
>
> num<- seq(88,300, len=200)
> lines(num, predict(fit10, data.frame(day=num)))
>
> -Ranae
>
>
> Does
>  ?predict.ns
>  not do what you want without having to explicitly manipulate the spline
> basis?
>
> -- Bert
>
> On Tue, Jun 5, 2012 at 1:56 PM, Ranae <[hidden email]> wrote:
>
> > Hi,
> >
> > I am looking at the change in N concentration in plant roots over 4 time
> > points and I have fit a spline to the data using ns and lme:
> >
> > fit10 <- lme( N~ns(day, 3), data = rcn10G)
> >
> > I may want to adjust the model a little bit, but for now, let's assume
> > it's
> > good.  I get output for the fixed effects:
> >
> > Fixed: N ~ ns(day, 3)
> > (Intercept) ns(day, 3)1 ns(day, 3)2 ns(day, 3)3
> >  1.15676524  0.14509171  0.04459627  0.09334428
> >
> > and coefficients for each experimental unit in my experiment:
> >
> >   (Intercept) ns(day, 3)1 ns(day, 3)2 ns(day, 3)3
> > 24    1.050360 -0.42666159 -0.56290877 -0.10714407
> > 13    1.104464 -0.30825350 -0.53311653 -0.05558150
> > 31    1.147878 -0.14548512 -0.78673906 -0.07231781
> > 46    1.177781 -0.22278380 -0.80278177 -0.02321460
> > 15    1.144215 -0.04484519 -0.06084798  0.07633663
> > 32    1.213007  0.00741061  0.03896933  0.15325849
> > 23    1.274615  0.16477514  0.00872224  0.23128320
> > 41    1.215626  0.57050767  0.11415467  0.10608867
> > 43    1.134203  0.48070741  0.72112899  0.18108193
> > 12    1.091422  0.39563632  1.01521528  0.22597459
> > 21    1.100631  0.44589314  0.98526322  0.23535739
> > 35    1.226980  0.82419937  0.39809568  0.16900841
> >
> > NOW, I want to write a spline function where I can incorporate these
> > coefficients to get the predicted N concentration value for each day.
> > However, I am having trouble finding the right spline equation, since
> > there
> > are many forms on the internets.
> >
> > I know it won't be a simple one, but can some one direct me to the
> > equation
> > that would be best to use for ns?
> >
> > Thanks a lot,
> >
> > Ranae
>
> --
> View this message in context: http://r.789695.n4.nabble.com/Do-YOU-know-an-
> equation-for-splines-ns-tp4632440p4632567.html
> Sent from the R help mailing list archive at Nabble.com.
>
> ______________________________________________
> [hidden email] 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.

______________________________________________
[hidden email] 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.

[Spencer Graves-2]

       I agree with Bill and Bert:  "predict" is the proper tool for
making predictions.  Pinheiro and Bates (2000) Mixed-Effects Models in S
and S-Plus (Springer) includes several entries in the index for
"predictions".  Please note, however, that there are a few lines of code
in that book the work in S-Plus but not R.  Fortunately, the corrections
are available in script files distributed with the package, which you
can find as follows:


> system.file('scripts', package='nlme')
[1] "c:/Program Files/R/R-2.15.0/library/nlme/scripts"


       The TaylorSpline{fda} function will give you explicit
coefficients each segment of a spline.  However, if you want model
predictions, you are probably best using predict with objects produced
by functions in nlme.  That package has seen lots of use and attention
by the R Core team, and should be pretty good -- especially with the
documentation provided by Pinheiro and Bates.


       Hope this helps.
       Spencer


On 6/6/2012 1:48 PM, William Dunlap wrote:

> Do you have to include the grouping variable, plotF, in your newdata
> argument?  E.g., after fitting the model with
>    rcn10G<-groupedData(N ~ day | plotF, data=rcn10)
>    fit10<- lme( N~ns(day, 3), data = rcn10G)
> try checking the predictions when you've include plotF in newdata:
>    par(mfrow=c(2,1))
>    plot(N ~ day, subset=plotF=="12", data=rcn10G)
>    points(num, predict(fit10, data.frame(day=num, plotF=rep("12", length(num)))), pch=".", col="red")
>
>    plot(N ~ day, subset=plotF=="43", data=rcn10G)
>    points(num, predict(fit10, data.frame(day=num, plotF=rep("43", length(num)))), pch=".", col="red")
>
> I am no expert on the lme and groupedData, but the general rule is that all variables involved
> in the model, except the response, must be given to predict.
>
> Bill Dunlap
> Spotfire, TIBCO Software
> wdunlap tibco.com
>
>
>> -----Original Message-----
>> From: [hidden email] [mailto:[hidden email]] On Behalf
>> Of Ranae
>> Sent: Wednesday, June 06, 2012 10:39 AM
>> To: [hidden email]
>> Subject: Re: [R] Do YOU know an equation for splines (ns)?
>>
>> I have not been able to get "predict" (or most functions) to run well with
>> grouped data in nlme.  I may not have it coded right, but this is what it
>> looks like:
>>
>> http://r.789695.n4.nabble.com/file/n4632567/spline.txt spline.txt
>>
>> library(nlme)
>> library(splines)
>>
>> rootCN<-read.table("spline.txt", header=TRUE)
>> rootCN$plotF<-as.factor(rootCN$plot)
>>
>> rcn10G<-groupedData(N ~ day | plotF, data=rcn10)
>>
>> fit10<- lme( N~ns(day, 3), data = rcn10G)
>>
>> plot(augPred(fit10))
>>
>> num<- seq(88,300, len=200)
>> lines(num, predict(fit10, data.frame(day=num)))
>>
>> -Ranae
>>
>>
>> Does
>>   ?predict.ns
>>   not do what you want without having to explicitly manipulate the spline
>> basis?
>>
>> -- Bert
>>
>> On Tue, Jun 5, 2012 at 1:56 PM, Ranae<[hidden email]>  wrote:
>>
>>> Hi,
>>>
>>> I am looking at the change in N concentration in plant roots over 4 time
>>> points and I have fit a spline to the data using ns and lme:
>>>
>>> fit10<- lme( N~ns(day, 3), data = rcn10G)
>>>
>>> I may want to adjust the model a little bit, but for now, let's assume
>>> it's
>>> good.  I get output for the fixed effects:
>>>
>>> Fixed: N ~ ns(day, 3)
>>> (Intercept) ns(day, 3)1 ns(day, 3)2 ns(day, 3)3
>>>   1.15676524  0.14509171  0.04459627  0.09334428
>>>
>>> and coefficients for each experimental unit in my experiment:
>>>
>>>    (Intercept) ns(day, 3)1 ns(day, 3)2 ns(day, 3)3
>>> 24    1.050360 -0.42666159 -0.56290877 -0.10714407
>>> 13    1.104464 -0.30825350 -0.53311653 -0.05558150
>>> 31    1.147878 -0.14548512 -0.78673906 -0.07231781
>>> 46    1.177781 -0.22278380 -0.80278177 -0.02321460
>>> 15    1.144215 -0.04484519 -0.06084798  0.07633663
>>> 32    1.213007  0.00741061  0.03896933  0.15325849
>>> 23    1.274615  0.16477514  0.00872224  0.23128320
>>> 41    1.215626  0.57050767  0.11415467  0.10608867
>>> 43    1.134203  0.48070741  0.72112899  0.18108193
>>> 12    1.091422  0.39563632  1.01521528  0.22597459
>>> 21    1.100631  0.44589314  0.98526322  0.23535739
>>> 35    1.226980  0.82419937  0.39809568  0.16900841
>>>
>>> NOW, I want to write a spline function where I can incorporate these
>>> coefficients to get the predicted N concentration value for each day.
>>> However, I am having trouble finding the right spline equation, since
>>> there
>>> are many forms on the internets.
>>>
>>> I know it won't be a simple one, but can some one direct me to the
>>> equation
>>> that would be best to use for ns?
>>>
>>> Thanks a lot,
>>>
>>> Ranae
>> --
>> View this message in context: http://r.789695.n4.nabble.com/Do-YOU-know-an-
>> equation-for-splines-ns-tp4632440p4632567.html
>> Sent from the R help mailing list archive at Nabble.com.
>>
>> ______________________________________________
>> [hidden email] 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.
> ______________________________________________
> [hidden email] 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.
>

--
Spencer Graves, PE, PhD
President and Chief Technology Officer
Structure Inspection and Monitoring, Inc.
751 Emerson Ct.
San José, CA 95126
ph:  408-655-4567
web:  www.structuremonitoring.com

______________________________________________
[hidden email] 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.

[Ranae]

I was able to get the predicted values from the splines.  Thanks so much
for the help.

I wrote a loop with some of the code that Bill suggested.  It seems that
when using predict with nlme, it is important to be specific with what one
is using as newdata.  This does come through in Pinheiro and Bates, I just
didn't recognize it to begin with.  Bert, I did try your code, but was only
getting coefficients, so I may have neglected a step.

##The successful code:
library(nlme)
library(splines)

rootCN<-read.table("spline.txt", header=3DTRUE)
rootCN$plotF<-as.factor(rootCN$plot)
rcn10G<-groupedData(N ~ day | plotF, data=3DrootCN)

fit10 <- lme( N~ns(day, 3), data =3D rcn10G)

plot(augPred(fit10))


t<- 152:305
subject<-rootCN[11:22,2]
sim<-NULL
for(i in 1:12){
sim<- cbind(sim, predict(fit10, data.frame(day=3Dt, plotF=3Drep(subject[i],
length(t)))))
}
colnames(sim) <- c(subject)


par(mfrow=3Dc(4,3))
for(i in 1:12){
plot(t, sim[,i], type=3D"l", main=3Dsubject[i])
}


-Ranae